Be sure you have checked Part 1 & Part 2 of this report:

• SideFX Houdini 21’s System Requirements Analysis - Part 1: Background, Benchmarks and Objectivity

• SideFX Houdini 21’s System Requirements Analysis - Part 2: O.S. & Memory

Welcome to the third, final and most important part of these report! Thanks for being here! Please share this report with a few Houdini users you know or share it on your favorite Houdini User Community or Forum. Thanks!

Share

Processor (CPU)

• Processor is one of the most interesting ones, although that paragraph would give a nerd like me a great puzzle to solve, let’s make this simple.

  • What SideFX is saying above is:

    • Intel:

      • If you have a Core i5/i7 and its model is not 4000 series or higher, don’t run Houdini 21 on it. i.e. Intel Core i7-4770K was an amazing processor 12 years ago, I loved each one of its 4 cores and its high frequencies for 2013 standards.

    • AMD:

      • Excavator Architecture include Processors like Athlon X4 900 series and many AMD A6/A8/A10/A12 8000/9000 series. CPU World has a full list of all Excavator processors1

  • All these mentioned processors deserve a place in the Computer History Museum, we do not recommend to use any of them, why?

    • In the last 10-12 years, processors have become orders of magnitude faster, both in single threaded performance for 3D manipulation and multi-threaded performance for fluid simulation and rendering.

    • Modern processors are also way more power efficient

    • These old processors do not support Windows 11, so you are already exposed given Windows 10 EoL

    • And finally, I value your physical and mental health, you do not want the stress of getting anything meaningful done on this kind of hardware.

• So what do we recommend:

  • For 3D modeling, animation and fast vector manipulation work, the super high single thread performance is a must:

    • Intel Core i7 or i9 Processors 12th gen or newer. The leap from 11th gen to 12th gen is so profound, that I do recommend 12th gen or newer. The state-of-the-art CPU here is the Intel Core Ultra 9 285K

    • AMD, the Ryzen 7 or 9 7000 series or newer are ideal. The state of the art CPU here is the AMD Ryzen 9 9950X or 9950X3D. The 3D cache should in theory help with latency for simulation, it’s just an educated guess at this point, I’d love to someday research it. Hey AMD, give me a ring if you are interested.

    • Please notice that these CPUs have been released in the last few years, 7 to 12 years newer than the ones recommended by SideFX.

  • If beyond the 3D stuff above, you want to do fluid simulation, render cloning, Karma XPU rendering on 1 or multiple GPUs, the recommended CPUs are from higher end product lines like: Intel Xeon W and AMD Threadripper Pro.

    Mid-tier:

    • Intel Xeon W-2595X (up to 26 cores)

    • AMD Threadripper Pro 9965WX (24 cores)

    Top-tier:

    • Intel Xeon W-3595X Series (60 cores)

    • AMD Threadripper Pro 9985WX (64 cores)

    • What about the Threadripper Pro 9995WX with 96 cores? Early tests by Level1Techs.com2, suggests there is better performance even at such high core count, however, it’s important to find the right model, since 50% more cores will bring you 2% - 25% better performance depending on the test.

• There is a lot that can be improved about how much we the tech nerds know about Houdini 21, however I’m not surprised that Puget Systems has the best analysis I have seen based on the simple benchmark from VFX Arabia plus their deep hardware expertise: https://www.pugetsystems.com/blog/2025/02/24/sidefx-houdini-performance-analysis-and-lessons-from-production/

• A note about Qualcomm. It’s important to notice that although Snapdragon CPUs are innovative and power efficient, it’s completely unknown if they could run Houdini 21, even in emulation, since ARM architecture do not support AVX instructions. I’d love see at least a blog from Qualcomm and SideFX covering any collaboration and enabling efforts.

Graphics Cards (GPU)

Let’s start with the GPU vendor that most people will use…

Nvidia

There is so much to discuss about the paragraph that SideFX dedicated to Nvidia. GL4.0 (do they refer to OpenGL?), >12GB VRAM, OpenCL 1.2, OptiX Denoiser support and all this just covers the functionality of Houdini, not the immense performance spectrum between all the GPUs that support these features.

The Nvidia Quadro K-series (Kepler family) were released in 2012-2013. Only the top of the line Quadro K6000 has the 12GB of VRAM required by Houdini. GPUs don’t launch every year, but when they do, the gen-on-gen performance is big enough to make CPUs be ashamed of their typical single digit improvements. GPUs typically go up >50% faster every generation. Sure, with high demand and limited competition, prices have gone up too. But enough talk about GPUs that are almost old enough to get a driver license.

We can go to the Maxwell generation (2015), where again, only the top of the line Quadro M6000 (12GB) and M6000 (24GB) would meet the criteria, but still these GPUs are obsolete.

Then comes Pascal (2016-2017) and this time, the Quadro P5000 and P6000, have enough memory and features, but compared to today’s standards, you can get the same performance at 1/20 of the price ($300 vs $6000), so let’s talk about what makes sense to buy today, so I’ll skip Volta & Turing (2018). All those are definitely obsolete. If you have an Ampere based card (2021-2022), specially the RTX A4000 or above and you are happy with it, it will work but the advances that have happened since are significant.

For a new workstation, you can try to find a good deal on an RTX Ada Generation supporting not only way higher compute, but all the SKUs have enough memory for Houdini from the RTX 2000 Ada (16GB) to the RTX 6000 Ada (48GB). As discussed earlier, if you are planning to do 3D work, 16GB is enough to get going, but when it comes to simulation, more memory and compute enables you to simulate bigger boxes, with more particles and way faster.

Finally we get to the NEW state-of-the-art Blackwell generation, definitely the obvious choice if you can find it and if you can afford it. If you are planning to render with Karma XPU or even a 3rd party renderer like the new Pixar Renderman XPU will benefit from a higher end model. I had the pleasure to attend the Renderman Art & Science Fair at Siggraph 2025 and they claimed that 98% of the Pixar scenes created so far, can fit in the 96GB that the Nvidia RTX 6000 Blackwell offers. Also, as we discussed on the Processor section, rendering in both Karma XPU or Renderman XPU will share the load with the CPU, so the fact that new powerful GPUs are available and enabled, doesn’t mean that these workflows won’t benefit from powerful high core count CPUs. A balanced system will always win!

What about AMD, Intel and Apple GPUs?

There is very little guidance from SideFX on GPUs from these vendor besides the fact that... “They work!” How well? are there other denoising libraries besides Nvidia Optix? maybe Intel Open Image Denoise (OIDN)3 ? Intel should work closer to SideFX to provide better guidance. After some investigation, I found this, deep into the Karma XPU documentation: “Karma currently ships with Intel’s OIDN fully functional out of the box”4. How is a user supposed to know this?

What about Apple? In our Foundry Nuke analysis, we found amazing performance scaling across Apple M-series SKUs and generations. Only a good comprehensive benchmark would show the impact of going from Pro to Max to Ultra SKUs from M1 to M5.

Skipped

Due to the length of this report, I skipped the following sections of the Houdini 21 MSRs, please let me know if you want me to go deeper in any of them

• Disk Space

• Input Devices

• Monitor

• Network

• Remote Desktop Solutions

Conclusions

• While it is a testament of the quality of Houdini’s code that run on ancient obsolete hardware, doesn’t mean that you should. If you value your productivity, the investment on your artists’ salaries and giving them room to experiment so your project has the best quality possible instead of good-enough, you should invest on a decent workstation. I’ll suggest three profiles, one for 3D and Motion Graphics, one for Simulation and, the dream setup we all want but can’t afford. I suggest you to use them as guidelines and adjust to your needs.

  

  Next Steps

  By working with SideFX and some of the XPU and Memory vendors above, I hope we can go deeper into Houdini with real world workloads to build a data-driven and objective analysis of the impact that each component has on Houdini workflows for procedural animation and simulation.

  This would be a Herculean effort that I look forward to go through, in order to serve the Houdini users’ community. Please ping me if you have any interest in this effort or give me feedback on what you would like to see coming out of a project like this.

  ---

• Do you need help selecting the right Workstation for your Workflow or Professional Application? Ping me

We need your help

In order to grow our mission to empower users with great tech, we need your support. These are a few things you can do to support us:

• If you haven’t already, please Subscribe to this newsletter, there is a free option available or you can support this project with a paid option

  Subscribe now

• Share this article with 3 people you think will find it valuable. I appreciate introductions introductions into Houdini community owners in order to reach more users.

Share

• Like this article,

• Repost it if you think it will help others

About the Author

This article was written by Hernán Quijano, Workstation Performance and Market Analyst at CavalryHQ. Our mission is to bridge the workstation industry and power users; improving guidance and removing technical roadblocks, so users unleash their talents focusing on accelerating software applications and workflows in Engineering, Media & Entertainment and A.I.

Disclaimers

Unless explicitly stated, this article has not been sponsored by any brand or organization | The author might personally own stock in one or multiple of the mentioned companies.

References:

Be sure you have checked Part 1 of this report:

• SideFX Houdini 21’s System Requirements Analysis - Part 1: Background, Benchmarks and Objectivity

SideFX Houdini 21 Official System Recommendations

Source and credit for all screenshots below: sidefx.com/Support/system-requirements/

Note: Screenshots taken at date of writing of this report. We noticed that the look and feel of the website changed recently, but not the content, which is what we are focusing on this article.

Operating System

• About Windows: Now that Windows 10 has reached End-of-Life (EoL), does it make sense for it to be supported? This has nothing to do with SideFX, so no problem on their part, but as a user, every day that passes using Windows 10 increases the exposure and risk on an unsupported Operating System, unless of course you are paying for the extra security maintenance. If you wanted an excuse to move to Windows 11, on fresh hardware or switch to a Linux distro, this is it.

• About Linux: Even if Linux is the path, I’d stay away from some of the listed distros, specially CentOS for the same reason as Windows 10. For example, CentOS was EoLd in 20211. In fact, if we look deeper into Linux distributions, these are the current versions, vs the ones listed on SideFX’s site and when they were EoLd. It’s also recommended to follow the VFX Reference Platform guidance. By following this guidance, you know there are many users with similar configurations and if anything goes wrong, there will be a high priority to fix any problem. For more information, visit: https://vfxplatform.com/

  

  *Note on Pop!_OS: See footnote2

• Regarding MacOS, Apple generally offers security updates and full support on the most recent 3 versions, so today, we would not recommend anything earlier than:

  • macOS Tahoe 26

  • macOS Sequoia 15

  • macOS Sonoma 14

  • … a far cry from the Minimum Recommended of macOS Big Sur 11

• O.S. conclusion: Just because you can, doesn’t mean you should! Whether on Windows, Linux, or MacOS, O.S. selection should be current, at least for security reasons. It’s reasonable to apply a Minimum Common Requirement rule, so in this case, the minimum requirements of the desired <current> O.S. should overwrite the ones for the SW application. Nobody doing professional work should run it on outdated tech. If orders-of-magnitude more performance isn’t enough of a reason, security and support should.

Memory

• This should be read as:

  • less than 16 GB? don’t even try

  • 16GB: you can install Houdini and start exploring it, maybe go through some tutorials with low/limited complexity

  • 32GB: is the real minimum for a good experience, doing mid-complexity/polycount 3D scenes, but if you are planning to go beyond the basic features and get into high-complexity fluid simulations, you will need more, how much? RAM is like oxygen or money, everything seems fine until you run out of it.

• It’s recommended that you open some of your most complex projects and as you work with it, observe on how much memory you are using. Nothing too sophisticated, simple tools like Windows’ Task Manager (Ctrl+Alt+Esc) can show you. i.e. My Mobile Workstation has 64GB of RAM, this is my Memory Usage right now, I’m using a bit less than half of my memory (27.5GB out of 64GB) and Houdini is not even open.

  

• Always leave room for increased complexity of future projects, upcoming versions of Houdini, other applications you will be using in parallel, your O.S. needs and always plan for a buffer for unknowns.

• Of course, this is easier said than done, considering the spike in Memory prices in the last few weeks. Having said that, having more than enough memory is critical since when your system runs out of memory and starts paging to your SSD, you will feel a significant impact in responsiveness and performance.

A final and very important note on memory:

• Bandwidth! Houdini, and any other application calculating many particle simulations, needs to access massive amounts of memory in parallel. The amount of memory in your system (total capacity) will dictate how much information it can hold, it is the memory bandwidth what will dictate how much of that information can be accessed in parallel. Think about when you go to the supermarket and there are long lines at the registers to pay, there are 8 registers but only 2 are open, so you wish there were 4 or all 8 registers open, everyone found all the items they need, but only 2 customers can leave at a time! That is exactly what happens with memory bandwidth, when doing simulations there are thousands, hundreds of thousands or even millions of particles’ data to be loaded and computed, so you want to have as much memory bandwidth as you can. Not all applications need this, but Houdini definitely benefits from that big pipe of data. This is one of the key differentiators between different CPU Product lines, as shown here:

  

  Perfect segway to discuss the next important component: the Processor.

Processor (CPU)

SideFX Houdini 21’s System Requirements Analysis - Part 3: CPU, GPU & Conclusions

• Do you need help selecting the right Workstation for your Workflow or Professional Application? Ping me

We need your help

In order to grow our mission to empower users with great tech, we need your support. These are a few things you can do to support us:

• If you haven’t already, please Subscribe to this newsletter, there is a free option available or you can support this project with a paid option

  Subscribe now

• Share this article with 3 people you think will find it valuable. I appreciate introductions introductions into Houdini community owners in order to reach more users.

Share

• Like this article,

• Repost it if you think it will help others

About the Author

This article was written by Hernán Quijano, Workstation Performance and Market Analyst at CavalryHQ. Our mission is to bridge the workstation industry and power users; improving guidance and removing technical roadblocks, so users unleash their talents focusing on accelerating software applications and workflows in Engineering, Media & Entertainment and A.I.

Disclaimers

Unless explicitly stated, this article has not been sponsored by any brand or organization | The author might personally own stock in one or multiple of the mentioned companies.

References:

Background

At Siggraph 2025 in Vancouver, BC, SideFX launched Houdini 21! The latest version of an application that changed and continues to push the industry forward when it comes to procedural 3D animation, including state-of-the-art fluid and particle simulations.

How important is Houdini in the Media & Entertainment industry? according to the State of 3D 2024 survey by Poliigon.com1, Houdini is an absolute beast:

• Houdini has the widest usage across Media and Entertainment, including animation, advertising, Games & VFX, but obviously the largest userbase in VFX

• When asked: “How well does the 3D software meet your needs?” Houdini scored at the top of all other applications and very close to Extremely Well

• Regarding User Experience, Houdini won top spots in Stability and Performance

• For Developers, Houdini took the top spot in both Speed of Development and Confidence in Developers

• When users were asked about which M&E sub-industry this software is best suited for, Houdini also took the combined top spot, including (cumulatively) Animation, Advertising, Games and VFX

So Houdini, is useful, fast, stable, SideFX listens to their users, etc. Does it have any weaknesses? Well, not many. In fact, the only area where Houdini scored last was Ease of Learning, and I get it, it’s a very powerful and complex tool. Having said that, I can see that SideFX is taking action, with tons of learning resources, most of them free and providing a platform for experts to sell courses, right on the Learn tab at the top of SideFx.com. Also, SideFX just launched a contest to promote How To - Tutorials from experts. Contest goes from Oct 2025 to Jan 2026. For more information, visit: https://www.sidefx.com/community-main-menu/contests-jams/how-to-contest/

Update: Poliigon’s State of 3D 20252 report came out while I was writing this article, and no surprise, Houdini is still scoring very high in all categories, so I see no statistical need to write this section again, having said that, go to Poliigon and check it out! There are new questions around AI usage, demographics, salaries and more!

With New software, comes New Systems Requirements

I’m sure Peter Parker says that in at least one of the multiverses.

What kind of system do you need to run Houdini 21? Answering this question for any application from any Software Vendor (ISV) should be as easy as sending you to their website, but if you browse my previous articles, it’s never as easy as it sounds.

If you want to dig deeper, I recommend reading:

In Houdini 21’s case, these are their System Requirements:

sidefx.com/Support/system-requirements/

As I argue on the article above, these pages need some help, sometimes to bring obsolete hardware specs to the present or at least to this decade. So let’s see how well SideFX did with Houdini 21’s System Requirements and let this be a useful article to facilitate the journey of any Houdini user that want to answer not only questions like:

• Can I run Houdini 21 on my current PC? The answer is yes, unless your system is like older than ~10 years old. But we are not here to make a cube and simulate less particles than what you can find in the air of a semiconductor fab, are we? (Those fabs are really clean… get it? Nevermind, let’s move on)

We want to know more useful things like:

• Is my current PC ideal for Houdini 21?

• What am I missing if I run Houdini 21 on my current PC vs a state-of-art PC/Workstation?

• What workstation should I buy to run Houdini 21 at all its glory?

Benchmarks and Objectivity

In order to make objective (please read: data driven) recommendations and not subjective ones (please read: educated opinions), we would need a good Houdini benchmark and a lot of data.

There are no official benchmarks for Houdini, and the only one I have seen from the community a this benchmark created by VFX Arabia3. I applaud their contributions bringing industry expertise for the benefit of all Houdini users! Some other day we can double down on this benchmark, like I digged on PxF Nukebench for Foundry Nuke on the article below, but that’s a topic for another day.

Today we are analyzing SideFX’s System Recommendations and making sense of them, so let’s get into it.

SideFX Houdini 21 Official System Recommendations

• Do you need help selecting the right Workstation for your Workflow or Professional Application? Ping me

We need your help

In order to grow our mission to empower users with great tech, we need your support. These are a few things you can do to support us:

• If you haven’t already, please Subscribe to this newsletter, there is a free option available or you can support this project with a paid option

  Subscribe now

• Share this article with 3 people you think will find it valuable. I appreciate introductions introductions into Houdini community owners in order to reach more users.

Share

• Like this article,

• Repost it if you think it will help others

About the Author

This article was written by Hernán Quijano, Workstation Performance and Market Analyst at CavalryHQ. Our mission is to bridge the workstation industry and power users; improving guidance and removing technical roadblocks, so users unleash their talents focusing on accelerating software applications and workflows in Engineering, Media & Entertainment and A.I.

Disclaimers

Unless explicitly stated, this article has not been sponsored by any brand or organization | The author might personally own stock in one or multiple of the mentioned companies.

References:

Pixels! Whether you use the left side of your brain to calculate them or the right side to paint them, once you fall in love with them, there is no way back. Some of you love fast pixels you can interact with, some of you don’t mind waiting hours for the highest quality one.

Sometimes we need them pure, sometimes we mix them with video to enhance storytelling. Some of you want to put some pixels on a small screen, some of you on a high-refresh-rate monitor, some of you on a headset or maybe on a massive screen; what we have in common is that we love those little tiny suckers! They are a form of expression, the minimum component of our arts.

For more than 50 years, we, the <pixel lovers> people, have been gathering at Siggraph, to share advances, insights, successes and failures, from on-boarding new blood to recognizing pioneers; to hear how the best in the industry solved their technical, production or artistic challenges and by listening to their deconstruction, understand what otherwise would look like magic on screen.

My Siggraph Journey

I have been to Siggraph several times in the last 15 years. At the beginning of my career, Siggraph was only a dream, even when I was an Engineering student / Intel Intern, building Render Farms with spare parts from the Intel warehouse in Colombia or later on in Brazil as a recent-college-graduate, it was completely unfeasible to attend Siggraph due to the combination of long distance plane tickets, unfavorable exchange rates and a starting salary of ~USD$900 / month (pre-tax and pre-caipirinhas)

Since I came to the U.S. in 2009, going to Siggraph was a no-brainer! It’s magical to get together with thousands of people that care about computer graphics of any kind (2D, 3D, video, VFX, real-time/interactive/miliseconds per frame, non-realtime/hours per frame). Siggraph attendees are technical, or artistic, or like me, love the interception of the left and right side of brains. There is something for everyone!

Must see events

As Julie Andrews would say: These are a few of my favorite things (at Siggraph):

• Keynotes - Industry Visionaries describing recent achievements, challenges, sparking curiosity about the future, and a call to action for the industry. This year’s keynotes are:

  • Dr. Anjali Tripathi, Future Earth: Storytelling From Space and the Exoplanet Revolution and ACM SIGGRAPH Award Presentations

  • Autodesk: Mike Haley, Maurice Patel, Nikola Todorovic -- Redefining Creativity: The Future of Media & Entertainment With AI

• Production Sessions - a deconstruction of some of the most challenging VFX in award winning movies and productions. If you grew up enjoying the bonus material on movie DVDs, you can’t miss these sessions!

• Computer Animation Festival - What happens when talented artists have freedom to create? not just good-enough, deadline-driven, cost-optimized productions, I’m talking about a few dozen of the best animations in the industry, artists unleashed! Hint: bring tissues, be ready to experience many different emotions!

  This year’s Siggraph will feature a very important event, titled: “30th Anniversary Toy Story Screening” presented by Ed Catmull, Pixar Co-Founder and Siggraph Pioneer. This Movie is so important to me and to the industry, that it will have its own separate article. Subscribe and stay tuned

  All Aliens: [pointing up] The clawwwwwwwww!

  Alien #1: The claw is our master

  Alien #2: The claw chooses who will go and who will stay

  

• Exhibition - As a tech guy, this is candyland! Top Workstation and Hardware components vendors, the latest Software booths full of demos and new capabilities, training / learning companies, many talks, experts everywhere. Many vendors offer free day passes, let me know if you need one!

• Courses - usually a handful of presentations about a specific topic from the best of the industry. i.e. Are you into 3D Rigging? see many industry experts talk about how they rigged their latest many-leg, wobbly, shape-shifting creatures on their latest movie. Some of my favs this year are:

  • A Whirlwind Introduction to Computer Graphics - great for beginners!

  • Advances in Real-Time Rendering in Games Part I & Part II - a must attend every year

• Birds-of-a-feather - if you want to find a room full of people that care about a specific niche, this is for you. Being a performance / benchmarking obsessed engineer, my obvious favorite this year is:

  • Benchmarking Graphics Performance on Modern Computing Systems, presented by the SPEC Committee with presenters from Nvidia & Intel. Curious fact, I co-presented this course last year and with new versions of SPEC Viewperf and SPEC Workstation launched since, I can’t wait to see what the SPEC GWPG Committee will share on this class!

And these are just a few of the ~25 different program categories! Believe me, there are many more things going on! Academic Posters, Emerging Technologies, Labs, you name it! See the full program at: Siggraph 2025 - Full Schedule

To Siggraph and beyond!

If everything I mentioned above is not enough, there are plenty of amazing side events organized by vendors and organizations of all kinds. Be sure to subscribe as I’ll be compiling and sharing as many side events as I can find, they fill up fast! Also, I’ll be re-posting events on CavalryHQ’s feed on LinkedIn, since this will become a real-time feed as we get closer to the event.

The biggest gems (IMHO) related but not part of the official Siggraph program are:

Note: all of them require registration, so do your homework, find that Eventbrite or Lu-ma link, and register ASAP, they fill up fast!

• The Pixar’s Art & Science Fair - Think about it as the Renderman’s State-of-the-Biz. A definite MUST for all Pixar and Renderman lovers. Ping your Pixar friends to get invited, or go to their exhibit booth and make a friend :-). The Pixar A&SF happens off-site, after hours and lasts a couple of hours. As a timeless bonus, you will walk out with a celebratory “walking little friend”

  

  While you are there, you might enjoy one of these amazing cocktails. Great job Pixar, Dell, AMD & Nvidia with your cocktail naming! 🤓

  

• Linux Foundation’s Open-source Days is hosted by the Hosted by Academy Software Foundation (ASWF) If you care about OpenUSD, OpenColorIO, MaterialX, OpenVDB, OpenPBR, VfxReference Platforms, Blender and more, this off-site multi day event is for you. Part of the agenda is open to anyone, while some parts required a Siggraph attendee badge. Virtual options available too

• Foundry Education Summit - How to train not your dragon, but the next wave of VFX talent: join Foundry’s Education Summit at SIGGRAPH 2025 for tools, insights, and industry-led inspiration.

And of course, between Happy Hours and community events, there is something for everyone at #Siggraph2025 ! See you there!

Help us help others

These are a few things you can do to support us, Thanks!

• If you haven’t already, please Subscribe to this newsletter, there is a free option available or you can support this project with a paid option

  Subscribe now

• Like this article

• Repost it on Substack if you think it will help others

• Share this article with 3 people you think will find it valuable

Share

About the Author

This article was written by Hernán Quijano, Workstation Performance and Market Analyst at CavalryHQ. Our mission is to bridge the workstation industry and power users; improving guidance and removing technical roadblocks, so users unleash their talents focusing on accelerating software applications and workflows in Engineering, Media & Entertainment and A.I.

Disclaimers

Unless explicitly stated, this article has not been sponsored by any brand or organization | The author might personally own stock in one or multiple of the mentioned companies.

While bragging is considered rude in Latin culture, I do believe on the power of celebrating wins and being grateful! After all, life is a series of experiences and while nobody wins every time, it’s up to us to keep trying, get better and celebrate wins, big or small, along the way.

I’m happy to share that I will be attending Siggraph, the most important event in Computer Graphics, but for the first time, not representing Intel, but CavalryHQ, as a Workstation Tech Analyst!

This is a big milestone and I’m grateful to all of you for your support and to the Siggraph Media Committee for helping me achieve this so early! It’s a partnership made in heaven, after all, every person that cares about Siggraph & Computer Graphics, will benefit from the insights, articles and reports we are creating at CavalryHQ.

A few months ago, CavalryHQ only existed on a napkin. Thanks to your support, I have reached thousands of power users and planted a few seeds to drive positive change at ISVs and Hardware companies.

There is no better time to ask you, if you have not already, please subscribe for Free, it matters a lot:

Subscribe now

If you want to do even more to support such mission, please share articles with a handful of colleagues; experts in VFX, 3D, video or any Media & Entertainment, Engineering or A.I. workflow or application, or technology experts in the workstation market! Go ahead, I’ll wait for you :-)

Share CavalryHQ by Hernán Quijano

Thanks for that!

Now that you have subscribed, you will be the first to read our Siggraph 2025 articles and a few surprises I can’t share yet!

#Siggraph2025

Quoting some Black Eyed Peas wisdom: “Let’s get started!”. This is our first article on out Siggraph 2025 series:

Be sure you have checked:

Nvidia DGX Spark and DGX Station, redefining the Future of AI Workstations?

Part 1 | Part 2 | Part 3

In previous articles, specially on Part 3 linked above, we talked about the PC OEM implementations of the Nvidia DGX Spark and their importance to Nvidia’s ramp of this innovative product line of DGX Workstations.

Today we will focus on the DGX Spark and we will cover DGX Station in a future article. Please subscribe for free to receive this report in your inbox and of course to support this movement to empower users with the appropriate compute to unleash their talents.

Subscribe now

When will the DGX Spark and variants be available? of course availability will vary by partner (PC OEM), having said that, reservations for many have been open for a while and according to Nvidia, availability starts in July. I’m typing this on July 7th, so let’s see where they are.

According to Nvidia’s Press Release, these are the PC OEM partners that are bringing to market implementations/variants of the DGX Spark: (in alphabetical order)

• Acer

  • Almost no information out there. No info from Acer and we were only able to find 1 pic from Nvidia’s “petting zoo” at Computex

    

    

• Asus Ascent GX10

  

  • Product Page & Availability Notifications |

  • Videocardz.com shared an invitation to a Webinar on July 22nd, 2025 (U.S.) and July 23rd, 2025 (APAC & EMEA). This doesn’t mean availability on that day, but for sure we will learn more details then

• Dell Pro Max with GB10

  

  • Product Page | Reservations

  • Availability: Per the reservation website: Get it as soon as July 14th, 2025. Having said that, what would I get on July 14th is uncertain, would I get the actual Dell Pro Max with GB10? or the Pre-order coupon for Priority Access? The website could be more clear.

• Gigabyte AI Top Atom

  

  • Product Page & Availability Notifications

  • It will be interesting to see what Gigabyte will do here given all the experience they have from the Gigabyte BRIX product line.

• HP ZGX Nano AI Station

  

  • No public Product Page, maybe more information is available behind registration wall | Availability Notifications

• Lenovo ThinkStation PGX

  

  • Blog | Availability Notifications

  • Availability: Q3 - 2025

• MSI EdgeXpert MS-C931

  

  • Product Page & Availability Notifications

• Channel Partners

  • Nvidia refers to more “Channel Partners” coming with DGX Spart variants, we were not able to find any other designs based on DGX Spark.

  • It’s a different story for DGX Station with partners like BOXX, Lambda, and Supermicro, but that is a topic for the upcoming DGX Station article.

Conclusions

• As of this writing, DGX Spark variants are not out there yet, however, it’s clear that partners are not on the same level of readiness. Acer is the laggard in this line up with only a mechanical sample shown at Computex and no product information (that we were able to find)

• While every other PC OEM mentioned above (please read: everyone mentioned but Acer) is gathering customer interests with notification forms, it’s clear that Dell, HP & Asus are ahead; they have shown since day one that they are bringing this platform to market as soon as possible. Lenovo is typically ready soon after.

• Dell and Nvidia itself are the only ones actually taking public reservations, although Asus was taking reservations initially for GTC 2025 attendees

• It is unknown at this point to know how successful DGX Spark will be, so I’m sure that all PC OEMs are moving with caution, having said that, not being time-to-market with a niche workstation like DGX Spark can be costly in market adoption.

• It is unknown how many A.I. Developers will adopt DGX Spark (including variants) as their main stand-alone development system, or as a head-less companion to their Mobile or Desktop Workstation, or not adopting it and staying on A.I. Data Science Workstations (Intel Xeon W + multiple RTX Pro GPUs) or Cloud Development environments. To the people at these companies forecasting volumes for this innovative first-gen product, I feel your pain, reach out if you need emotional support :-)

• Differentiation continues to be our biggest question mark, so far, all offerings from all PC OEMs are listing the same specs and positioning from Nvidia. Will each partner design just a way for them to say, “me too” to their customers? or will there be any differentiation between them? Time will tell!

We need your help

In order to grow our mission to empower users with great tech, we need your support. These are a few things you can do to support us:

• If you haven’t already, please Subscribe to this newsletter, there is a free option available or you can support this project with a paid option

• Like this article

• Repost it on Substack if you think it will help others

• Share this article with 3 people you think will find it valuable

Thanks!

Share

About the Author

This article was written by Hernán Quijano, Workstation Performance and Market Analyst at CavalryHQ. Our mission is to bridge the workstation industry and power users; improving guidance and removing technical roadblocks, so users unleash their talents focusing on accelerating software applications and workflows in Engineering, Media & Entertainment and A.I.

Disclaimers

Unless explicitly stated, this article has not been sponsored by any brand or organization | The author might personally own stock in one or multiple of the mentioned companies.

Many articles and reviewers do their best, yet they get it wrong

A new Workstation is out, a new CPU is launched, a new GPU arrives; we can’t wait to see what it can do and we rush to read the reviews. There is a lot of commentary around specs, core counts, frequencies, bits and bytes, then comes the hardest question: “So what?”

What can all that new technology do for my software application and therefore my productivity, quality of my work, income, etc? There has to be a way to objectively measure performance so we can understand the benefits of new technology right? right?

Well, of course! that’s what benchmarks are for, Maxon Cinebench1 or Blender Benchmark2 there are plenty of them and most of the time, reviewers are not experts on most software applications, so they trust that the Software Vendor (ISV) and its benchmark will properly guide them. I mean, why wouldn’t they. Note: I’m ignoring synthetic benchmarks (i.e. Passmark) on purpose because they are even more misleading!

Even if some of these benchmarks are based on real parts of applications, there is one problem, they should be taken in context, since they are not a good proxy for the application performance as a whole. They are a good proxy for a very specific set of workloads in such application; in the 3D world, it’s typically Render Time! This is the case for Blender Benchmark and Cinebench. So if a Workstation can render fast, it’s the right machine for you? Not so fast (pun intended)

Typically, a CPU with many cores will be good for Rendering, after all it is an embarrassingly parallel3 workload, where a frame can be splitted in tiles or multiple frames or batches of frames can be processed separately and independently. We won’t do a deep dive on Amdahl’s Law4 today, but maybe in the future, would you like me to? The point is that when you are doing 3D modeling, texturing, animation, rigging, and the majority of the tasks that happen on your 3D application when you are not rendering, are massively different in nature and therefore their compute requirements are also very appart. In fact, most of the time you are manipulating vectors (CAD) and interacting with your application, you need very few, really fast cores, so having a top of the line CPU from the newest architecture makes all the difference when you are in front of the system, regardless of rendering time.

Other factors can also play a role depending on scene complexity, like having plenty of RAM available, CPU cache sizes, among others.

Based on the example above, my hot take is that you would be way more productive on a 3D application, using a new Intel Core Ultra or AMD Ryzen CPU than an older and higher core count Intel Xeon W or AMD Threadripper. Having said that, when export operations arrive, the big brother CPUs (Xeon W and Threadripper) will leave Core Ultra and Ryzen in the dust.

If CPUs were vehicles. Intel Core Ultra and AMD Ryzen are sports motorcycles they will always win the quarter mile races. However, when you need to move a Ton of dirt, you will definitely go with a pick-up truck, like Xeon W or Threadripper.

How do we fix this?

1. Be aware and trust your gut and experience

If you are reading this, my guess is that you are an experienced user on at least a handful of professional applications in your industry/workflow. Go try different Workstations, from PC OEM samples or try one from a colleague; run some of the projects that made you shed a tear. If it feels off, it probably is. Keep digging until you see your project running smoothly and if not, identify the bottleneck. Think about where the complexity will increase in future projects and try to simulate that.

I feel this can be yet another topic for the Bin List, let me know if you agree. If you need help on this process, reach out. This should be objective, but there are a lot of educated guesses that will help you along the process.

2. ISVs: Build Better Benchmarks

BBB - trademark pending :-D

It would be wonderful if Blender Benchmark or Cinebench also had workloads for other part of the application, not only the “Export” operation of Rendering. I like to look at an application as a whole, as flow that a user follows, even if every user can do things in a different way, there are many commonalities. Such flow goes like this:

Import

The process of bringing all the footage/media/assets you need into your Workstation. This phase is more about networking, I/O ports and peripherals than compute

Load

This phase is about opening and bringing your project and dependencies into the application, in complex projects this can take several minutes. This process is often ignored, yet there is a lot that happens in terms of storage, memory sub-system, decompression, etc

Active Editing

This is when you are actually creating! Productivity is defined by you being unleashed on this phase, fast viewports, responsiveness, being able to experiment and never be afraid of “sliding a slider” because your Workstation can handle anything you want to try in the pursuit of quality and excellence!

Export

This is when you pick your settings and typically go away! tests and quick experiments typically allow for a coffee refill and final exports mean progress bars that cook while you sleep or even you submit to a higher compute solution. Interestingly, this is what most benchmarks test!

Example: Epic Games - Unreal Engine

There is no benchmark for Unreal Engine out there. Luckily there are experts that have come to the industry rescue, and have found ways to test Unreal Engine, creating some workloads, typically such workloads are focused around:

• Compiling Unreal Engine from source code. This is a great workload for programmers, not artists

• Light Baking / Build Lights

• Compile Shaders

All this a good, but what about really important and modern features like:

• Lumen performance

• Nanite performance

• Loading time

• Viewport performance / preview FPS

• Memory usage by tile on World Partition

• Metahuman Creator

• Blueprint compiling performance

… to name a few.

Hope: There are great benchmarks out there!

There are people, companies and organizations that “Get it” and are developing great Workstation benchmarks and the best news is that YOU can help them with your expertise!

SPEC Graphics & Workstation Performance Group (GWPG)5

This Committee has amazing benchmarks with complementary focuses:

• Graphics: SPECviewperf

  • Focused on the Graphics hardware for professional applications, this benchmark includes workloads for popular titles like 3ds Max, Catia, Creo, Maya, Solidworks and yes, there are new ones we needed, like Unreal Engine for Lumen, Nanite and Temporal Super Resolution. For a full list and details, visit their website (listed on References)

• System: SPECworkstation

  • This benchmark shows you individual scores per industry:

    • AI & Machine Learning

    • Energy

    • Financial Services

    • Life Sciences

    • Media & Entertainment

    • Product Design

    • Productivity & Development

  • Independently testing the major Hardware Subsystems:

    • Accelerator

    • CPU

    • Graphics

    • Storage

• Application: SPECapc

  • There are multiple benchmarks, one for each application, including (individually): Siemens NX, Solidworks, Maya, Creo, 3ds Max

  • If your goal is to deeply understand one of these apps, this is the way to go

Puget Systems - Pugetbench6

From the PC OEMs, Puget Systems is the one that took the “So What?” question more seriously, dedicating resources to develop and maintain benchmarks. This is not an easy task and in my honest opinion, this is a very smart investment since their objective recommendation has become a tool for Influencers and Reviewers too, giving them big visibility that shows their leadership and they can translate into sales.

Pugetbench currently covers:

• Adobe Premier Pro

• Adobe After Effects

• Adobe Photoshop

• Adobe Lightroom Classic

• Blackmagic Design - Davinci Resolve

Puget has a developer program to encourage experts like you to contribute knowledge, expertise and ultimately useful workloads! Please join them, your contributions have industry-wide impact. https://www.pugetsystems.com/pugetbench/development-program/

Conclusions

• It’s important to make informed and objective (data-driven) decisions when purchasing Workstations for Professional Applications and Workflows. Having said that, a simple chart from an easy to run benchmark is not enough; think about what your workflow and identify the phases where you need to increase productivity

• I celebrate the fact that there is little coverage overlap between SPEC benchmarks and Pugetbench, giving the industry a broader scope by complementing each other. At the end of the day, we are on this together and it makes no sense to compete in a space as specialized as this

• It’s critical that YOU, as an expert in your industry, contribute to the SPEC Committee7 and/or the Pugetbench Developer Program8 . This is particularly important if your expertise is on an application not covered, yet! (by “yet”, I mean: “until you help them”)

• Do you need help selecting the right Workstation for your Workflow or Professional Application? Ping me, I’m happy to help.

We need your help

In order to grow our mission to empower users with great tech, we need your support. These are a few things you can do to support us:

• If you haven’t already, please Subscribe to this newsletter, there is a free option available or you can support this project with a paid option

  Subscribe now

• Like this article,

• Repost it if you think it will help others

• Share this article with 3 people you think will find it valuable

  Share

About the Author

This article was written by Hernán Quijano, Workstation Performance and Market Analyst at CavalryHQ. Our mission is to bridge the workstation industry and power users; improving guidance and removing technical roadblocks, so users unleash their talents focusing on accelerating software applications and workflows in Engineering, Media & Entertainment and A.I.

Disclaimers

Unless explicitly stated, this article has not been sponsored by any brand or organization | The author might personally own stock in one or multiple of the mentioned companies.

References:

This is Part 3 of the report, be sure you already checked:

• Part 1: Foundry Nuke 16: Workstation Selection Guide

• Part 2: Projecting Foundry’s Tested Hardware to 2025

Let’s continue with our final and most exciting part, let’s cover Benchmarking, Sensitivity Analysis, CavalryHQ’s Hardware recommendations for Nuke, and Conclusions & Call to Action!

Benchmarking (#3)

Why are benchmarks so important? Because they change the discussion from subjective to objective. What makes a good benchmark? that can be a whole article, at some point I will get to it. For now, let’s just say that a good benchmark is:

• Easy to run

• Repeatable

• Based on real world workloads (from the application in question)

With good benchmarks, PC OEMs and component vendors (CPU/GPU) have something to execute, analyze and understand what the application in question needs. This will allow them to better position solutions for users. Also, they can understand the bottlenecks of their current products and make design changes on future ones!

After our previous article on Nuke’s Minimum System Requirements, I had the opportunity to meet and connect with two experts of the Nuke community:

• Frank Rueter, VFX Supervisor at OHUfx and Founder of Nukepedia.com. A leading community for all things Nuke

  “Currently so many AI tools produce amazing results that would benefit the VFX industry, but not everybody can afford the top-of-the-line Hardware (GPUs) to run them successfully in a production environment.

  Collecting meaningful industry benchmarks is essential for creating the most constructive relationship between hardware manufacturers, software developers and end users. When that triangle works well, everybody wins!”

  Frank Rueter - Nukepedia Founder

• Xavier Bourque, VFX Compositor and creator of Pixelfudger.com and very relevant to our discussion: PxF Nukebench.

  “VFX studios are very cost conscious. They require the best "bang for the buck" when building workstations for artists. Standardized performance benchmarks for DCC packages like Nuke are essential to achieve this goal. Without benchmarks, IT departments need to blindly guess what kind of hardware has the best cost/performance ratio for end users”

  Xavier Bourque

Meet PxF Nukebench!

Developed by Xavier Bourque at Pixelfudger, PxF Nukebench is a grass-root tool, born out of necessity, from a Nuke power user to all Nuke users!

Learn all about it here:

As you can see on their website, PxF Nukebench encourages users to submit their results, and I encourage as many Nuke users as possible to run and submit results too! The more data gathered, the more we can learn from it. There are only a few hundred submissions, so take all this with a grain of salt, however, I expect the high-level conclusions to stand.

After investing enough hours analyzing this data, shout out to Gemini PRO for giving me a hand directly inside Google Sheets, these are my findings and how they affect the Hardware recommendations:

Processor (CPU)

Complementing the data with discussions from forums and support blogs from Foundry, you will find that Nuke likes high single-threaded performance. That comes with the combination of a modern generation of CPU running at the highest possible frequency. Keep in mind that a CPU from an older generation, will typically run at a lower frequency and even at the same frequency will run less Instructions-per-clock (IPC), resulting in lower performance in general. When it comes to core count, it also matters; looking at the AMD Threadripper CPUs, you will see significant increase up to 24 core processors and performance will keep increasing all the way to 64 cores, but it’s “only” around 80% faster with 2.5X the cores, so far from the linear scaling you would see on a rendering workload.

In the charts below, I removed the Y-axis labels, because I want to stay high-level until you all submit more data, I just want to show certain observations:

Observation 1: Every CPU family gets faster and there is a big difference within the line up. Always pick the top of any tier you pick

One way to look at CPU performance per core and core count is to look only at the Apple CPUs, you can see that every generation gets better, and that there is a lot of value going to the M_ Max and Ultra models. This applies CPU performance, we will talk about the GPUs later.

Observation 2: High Frequency matters and core count too, specially up to 24 cores.

Another discovery is that given that Nuke really likes high frequency, there is a tier of processors we have not talked about in the recommendations and will define the “Value-Range” where you will get the best performance per Dollar invested, and that is the Intel Core or AMD Ryzen processors. As with Apple, you want to stay at the very high end of each family as shown on the chart below. There is significant difference between the bottom and the top of the family, so the best pick here is the Core i9 (Core Ultra 9) on the latest generation. I picked this 13th gen because we had more data on it, but there are 2 new generations newer than this one, 14th gen Intel Core and the new Core Ultra Series 2. I’ll add a recommendation below for this tier.

Observation 3: Be careful with the Tier you select: Value-Range vs Mid-Range vs High-End. Pick any family you can afford to richly configure, for example, it’s way better to have a Mid-Range Workstation fully loaded, than a High-End Workstation with the lowest components.

Let me illustrate this with some examples from the AMD line up. You will be better off with a top-of-the-line 16-core AMD Ryzen 9, top of the “Value-range”, than a High-End Workstation with a 12-24 cores AMD Threadripper, if you go High-End, go for at least 32 cores as suggested earlier on this article. I wish we had more data (please read: 32-core AMD Threadripper 7975WX) to make this point clearer, we will get there with your help.

Memory

We also know that Nuke can use large amounts of RAM Memory. How much will depend on the complexity of your project, so start with the recommendations on this article and grow from there. Think about a project you worked on that was particularly complex, if possible, load it on the Workstation with the highest amount of memory that you know, from a coworker or friend, and observe how much memory it takes. It’s always a good rule to have at 25% headroom, for multi-tasking or just because you can be sure that your most complex project is ahead of you, not behind!

Memory will also be important to pick your Workstation family, a “Value-range” workstation used to be limited to 128GB not too long ago, but advanced in Memory technology and density have risen (no pun intended) this limit to 256GB.

Final point on memory is Bandwidth, or how much memory is the CPU able to access simultaneously. Most Intel Core or AMD Ryzen CPUs support 2 memory channels, while Mid-range and High-End support 4 or 8 channels. Apple has a significant advantage since they are putting a lot of bandwidth on their M silicon CPUs, but it’s uncertain so far, how much this affects performance in Nuke. My experience tells me that we will see a difference when using CopyCat, due to the size of the Models used in A.I./M.L. Again, only benchmarks will answer the “so what” on things like this.

Graphics Card (GPU)

When it comes to Graphics Card (GPU), you will need a modern generation to have access to the latest features, and of course the higher the GPU model the more performance you will get. Specially if you want to use the Machine Learning / CopyCat capabilities in Nuke, quickly advancing with each release since version 13.

Again, I can’t wait to get for the community to submit more results, but these are some of my observations:

• The best Apple GPUs (in record) are the M2 Ultra (60 GPU-cores) and M4 Max. These have comparable performance to an Nvidia 2080Ti. I would like to see scores for M3 Ultra, the beefiest Apple-based GPU you can have to date.

• When it comes to AMD Radeon GPUs, the RX6900XT ranks well too, but again, I would like to see scores for the latest gen RX7900XT.

• The most obvious observation is that Nvidia has the best positioned GPUs, where any “Ada Generation” mid-range like the RTX 4000 or above have the best scores with the RTX6000 Ada taking the top performance spot. Having said that, a consumer GeForce RTX 4080 or a professional RTX 4000 have a better value with 3/4 of the performance at 1/6 of the price. The new RTX Pro cards should fly! I can’t wait to see those scores.

Conclusions & Call to action

Before we go over the conclusions, please support our work by subscribing, any Free or Paid tier is welcomed and tells us that you want to see more. Thanks

In summary:

• If you care about Foundry Nuke, I can’t put in words how important it is for you to take the time to run PxF Nukebench and submit your results. Even if you already see a similar config posted, your results will increase the consistency and confidence of the data.

• Intel, Nvidia, AMD, Apple: Run PxF Nukebench, submit scores, and have your experts look “under-the-hood” to see where your silicon is reaching bottlenecks and uncover opportunities for optimization in collaboration with Foundry.

• Dell, HP, Lenovo, BOXX, Puget Systems, Origin PC, how do these Workstation recommendations align with yours? Let’s collaborate to empower this talented community of experts in need for state-of-the-art solutions

• Foundry team, on your webinar you mentioned the existence of some performance tools or workloads you use internally and contemplated the possibility of releasing them. Please do! it would be a gift to have more Nuke workloads, whether they become part of PxF Nukebench or independent.

• Finally, regarding Hardware recommendations, you could safely go with the Projections of the Foundry Tested Hardware shown earlier; or even better, here are my recommendations for well-balance, state-of-the-art workstations, for 3 different levels of complexity and price points. They take into account my analysis of PxF Nukebench, I’ll review these recommendations when all of you (users are companies) have submitted your results! Each of these configurations could be more robust, but I wouldn’t add any value just recommending the best/max of each category. Add more RAM or Storage as needed, always taking into account the number of modules for max bandwidth and performance.

Now you know what you need to run Foundry Nuke and feel you are in 2025! There is no wrong choice if you pick from the ones above, it’s just about how complex you think your project will be! Now go show everything you can do; the limit is your talent and imagination!

Do you want to collaborate on our mission to bridge the Workstation industry with power users, empowering them to be productive?

Let discuss your strategy, technical or business challenges, user guidance for top software applications? We are one click away!

About the Author

This article was written by Hernán Quijano, Workstation Performance and Market Analyst at CavalryHQ. Our mission is to bridge the workstation industry and power users; improving guidance and removing technical roadblocks, so users unleash their talents focusing on accelerating software applications and workflows in Engineering, Media & Entertainment and A.I.

Disclaimer

Unless explicitly stated, this article has not been sponsored by any brand or organization. The author might personally own stock in one or multiple of the mentioned companies.

Be sure you have checked Part 1 of this report: Full Foundry Nuke 16: Workstation Selection Guide - Part 1

Based on the systems shown, this is CavalryHQ’s projection into state-of-the-art workstations to the writing of this article in June 2025:

Mid-Range Workstation Configuration

Characteristics

High Performance, balanced cost-value, limited expansion

Examples of Workstation models in this category

Dell Precision 5860 Tower, HP Z4 G5, Lenovo ThinkStation P5

About Upgradability

Memory (RAM): 64GB is just the equivalent, DDR5 is faster vs the DDR4 on the Foundry Tested config. About capacity, we know from experience that Nuke likes memory, so add more depending on your project peak complexity and your budget. In this platform, ideally always use 4 RAM modules for optimal performance and bandwidth, so before you buy, know that if you start with 4 memory RDIMMs and then add another 4, your memory frequency will go down a bit (~10%). I’d recommend starting with the amount of memory you think you will need in the next couple of years.

Similarly, on the Storage (SSD) front, 1TB is just the equivalent, but add as much as you will comfortably need for the project. Contrary to the RAM, adding more SSDs over time can potentially increase your performance, since you can have 1 SSD for your O.S. and Nuke, another for your Project/Data and even another one for Paging/cache files. Each one will have separate, dedicated lanes to the CPU. Check with your PC OEM how many SSDs you can have on your Workstation. For example, the HP Z4 G5, according to its datasheet, supports 4 NVMe m.2 SSDs with the possibility to add more via a PCIe card.

This Mid-Range Workstation is already configured robustly for its class, this means that:

• You won’t be able to upgrade to a better CPU

• You can add a second GPU or replace the listed one with a higher end model

• You can add more RAM and/or storage

That is not a lot of room to grow and that is what I meant by “limited expansion”

High-End Workstation Configurations

Characteristics

Top Performance, higher price point, most expansion

Intel

Examples of Intel-based Workstation models in this category

Dell Precision 7960 Tower, HP Z6 G5 / Z8 Fury G5, Lenovo ThinkStation P7

Fixing the baseline

There are multiple things off with the listed Tested Hardware configuration from Foundry:

• Intel Xeon w9-3495X has a 4.8GHz Turbo, not 4.4 GHz.

• It is a mystery to me why this Workstation has 96 GB of RAM, also 4x16 = 64 not 96. As I mentioned, for ideal performance, this system should be populated with 8 memory modules, so this system had 4x24GB = 96GB or 6x16GB = 96GB, either way, leaving memory channels unpopulated, which is a waste of a key capability of this platform. Also, this system has DDR5, not DDR4.

• Nvidia dropped the “Quadro” brand with the mentioned Ampere Architecture RTX A4500, which is a minor detail, having said that, this is card comes with 20GB of VRAM, not 8 GB.

I corrected the (outdated) Tested Hardware on the table below to have a good baseline to project to state-of-the-art hardware:

AMD

Examples of AMD-based Workstation models in this category

Dell Precision 7875 Tower, HP Z6 G5 A, Lenovo ThinkStation P8

Fixing the baseline

Again, there are multiple things off with the listed Tested Hardware configuration from Foundry:

• While is accurate that this CPU has a Base Frequency of 3.5 GHz, it’s important to mention that it can peak to 4.2 GHz Turbo, this is important for bursty lightly-threaded workloads, for example when the user is interacting with Nuke’s UI, this will affect latency and responsiveness

• Nvidia Quadro RTX 5000 has 16GB of VRAM, not 8GB

I corrected the (outdated) Tested Hardware on the table below to have a good baseline to project to state-of-the-art hardware:

The above are the results of just projecting what is listed on Foundry.com to the latest workstations available in the market. Just doing these would bring significant performance improvements to Nuke. Having said that, this method is still subjective since we don’t know exactly how much better these workstations are, or how much better is a Mid-range vs a High-End workstation. In order to have an objective discussion about this, we need benchmarks based on real-world Nuke workloads. And there is great news!

Benchmarking (#3)

Why are benchmarks so important? Because they change the discussion from subjective to objective. What makes a good benchmark?

Follow us to Part 3, the final section where we cover: Benchmarking, Sensitivity Analysis, CavalryHQ’s Hardware recommendations for Nuke and Conclusions

About the Author

This article was written by Hernán Quijano, Workstation Performance and Market Analyst at CavalryHQ. Our mission is to bridge the workstation industry and power users; improving guidance and removing technical roadblocks, so users unleash their talents focusing on accelerating software applications and workflows in Engineering, Media & Entertainment and A.I.

Disclaimer

Unless explicitly stated, this article has not been sponsored by any brand or organization. The author might personally own stock in one or multiple of the mentioned companies.

It would be easy to just check the Minimum System Requirements (MSR) on Foundry’s website and get to work; there are many reasons why those specs are simply not ideal. If you are curious, we just wrote a full report about it. Having said that, today we are in solution space and will make workstation recommendations for Nuke, based on the best data points we could find.

Hardware recommendations - For all applications

I think there are different levels of hardware recommendations, going from useless to useful:

#1. Minimum System Requirements (MSR)

Complexity: Low | User Productivity: Low | Cost efficiently: Low

Not recommended, just because you can, doesn’t mean you should! Why is this not cost efficient? because you lose more on employee salary, wasted talent, frustration, and SW licenses cost, than what you save by using Nostalgia-Hardware

#2. Hardware Projection

Complexity: Medium | User Productivity: High | Cost efficiently: Medium

In the absence of objective data from real world workloads, this is the best option.

If the ISV or application experts have a Best-Known Configuration, then project it to the latest generation available in the market.

#3. Benchmarking

Complexity: Medium | User Productivity: High | Cost efficiently: High

A benchmark based on real world workloads, specifically from the application in question is hard to find. I wish every top ISV would create one, but when not, sometimes amazing power users step in and develop a benchmark out of need; they understand the value!

Running the benchmark and gathering data is a big task, whether from a single company or influencer lab running many configurations or a community effort from users consolidating data.

Hardware recommendations - For Foundry Nuke

We are lucky that for Foundry Nuke we have #2 and #3!

Hardware Projection (#2)

As we covered in the previous article, Foundry had the great idea of sharing the configuration they have tested and likely use themselves, only 1-2 generations behind, which is way better than the ~20 generations ago from of the MSR. Apologies, I’ll stay positive :-)

These are the “Tested Workstations” from Foundry’s website:

All those 4 workstations are outdated, not obsolete, just outdated. What I mean by that is that I wouldn’t buy them as listed if I was in the market for a new system. They were launched between 2019 and 2020. The Lenovo ThinkStation P7 is the exception and the only I would consider current, with a configuration that was the state-of-the-art until summer 2024, the P7 is still the latest today, although it can be configured with a “refreshed” Intel Xeon W processor, with slightly higher core counts, frequencies and therefore, performance.

It’s important to clarify that the 4 workstations above do not belong to the same “tier”, there are “Mid-range” models like the HP Z4, all the way to “High-End range” models even with 2 CPUs. Things have changed so much that I’m not inclined to recommend dual socket workstations until we analyze Nuke with hard performance data (#3)

Based on the systems shown, this is CavalryHQ’s projection into state-of-the-art workstations to the writing of this article…

Part 2: Projecting Foundry’s Tested Hardware to 2025

About the Author

This article was written by Hernán Quijano, Workstation Performance and Market Analyst at CavalryHQ. Our mission is to bridge the workstation industry and power users; improving guidance and removing technical roadblocks, so users unleash their talents focusing on accelerating software applications and workflows in Engineering, Media & Entertainment and A.I.

Disclaimer

Unless explicitly stated, this article has not been sponsored by any brand or organization. The author might personally own stock in one or multiple of the mentioned companies.

Let’s start from scratch, or should I say from scraps?

In order to run Nuke 16, you need a system that meets Nuke’s Minimum System Requirements (MSR):

That’s it! thanks for coming! :-D

I wish it was that easy! As you saw on my previous article, I’m not a big fan of Minimum System Requirements. Let’s analyze the ones above, for sake of time, I’ll focus only on Processor (CPU) and Video Card (GPU)

Processor (CPU):

MSR: x86-64 processor, such as Intel Core 2 Duo or later

Observations:

• This is extremely old, Intel Core 2 Duo is from 2006, I remember it because I was part of the launch! Amazing processor, very power efficient… in 2006!

• It was actually hard to find this information because of how old it is, but here are some SPECint and SPECfp reports for one of the first-generation Intel Core 2 Duo:

  • SPECint_rate2006: 27.91

  • SPECfp_rate2006: 23.72

• Now just for comparison, there is a CPU (also old) that Intel launched on August 8th, 2015 (~10 years ago), I remember the date because I helped launch this CPU too! Intel Core i7-6700K, Skylake architecture

  • SPECint_rate2006: 2703

  • SPECfp_rate2006: 1994

• I couldn’t pick anything newer than this because these benchmarks are so old they evolved to represent more modern workloads. SPEC CPU 2006 retired, opening the door to SPEC CPU 2017 and it made little sense to even test this CPUs on the new version

Let’s normalize these results and chart the CPU performance from the Minimum System Requirement suggested by Foundry (from 2006) and a CPU from 2015 (already obsolete in my opinion):

• The Core i7-6700K (2015) is a 4 core CPU with a max frequency of 4.00 GHz, imagine a more modern CPU from the last couple of years like the 14th gen Intel Core i9-14900K (2023-2024); it has a peak frequency of 5.6 - 6.00 GHz and 24 cores (8P+16e). Let’s do some back-of-a-napkin math: 6X core count + >40% frequency + ~8% compound over 8 generations = A ton of performance! Easily be another 10X performance, over 2015 or 100X over 2006 performance levels.

• Also, and that is just Core processors from Intel (or Ryzen from AMD) there is a whole product line of more powerful CPUs like Intel Xeon W or AMD Threadripper processors that go up to 60 or 96 cores respectively! In summary:

  It would be insane to run Nuke on anything remotely close to the Minimum System Requirements (MSR) or even 10X that

Today, I called into Foundry’s webinar; not only was the Foundry team great at showcasing Nuke 16’s new features, but I was also lucky enough to be there live and ask them about System Requirements on their Q&A time! Thanks to the team for being so generous with your time and letting the community chime in!

Nuke 16.0 showcase: multishot workflows, Node Graph updates & live Q&A

Their point about the listed MSR is that many studios and Nuke users still have older hardware and they want to assure such users that they can still run Nuke.

While I agree that taking care of users is obviously important, encouraging users that a system 100X slower can do the job is, IMHO, a disservice. Think about it this way:

If you are a Media & Entertainment Studio, with Video, VFX and 3D workflows, you are investing in software licenses and salaries for talented <insert application names here> users, it is your duty to empower them with compute solutions that unleash the best they can produce in the limited time they have until the next project deadline/milestone.

I can keep going with the analogies, but let’s move to the GPUs…

Video Card (GPU):

Graphics card with at least 512 MB of video memory and driver support for OpenGL 2.0.*

*To enable optional GPU acceleration of Viewer processing, you need OpenGL 2.0 with support for floating point textures and GLSL.

OpenGL 2.0, according to Wikipedia, released in 2004, by no other than Silicon Graphics. I found this 2005 Nvidia paper: “Nvidia OpenGL 2.0 Support” by Mark J Kilgard5 . Here is a quote from it: “NVIDIA support for OpenGL 2.0 begins with the Release 75 series of drivers. GeForce FX (NV3x), GeForce 6 Series (NV4x), NV3xGL-based Quadro FX and NV4xGL-based Quadro FX GPUs, and all future NVIDIA GPUs support OpenGL 2.0”

I made a case for ~100X performance on the CPU side, that is just a fraction of the performance that the GPU has seen in this timeframe. here is a chart just covering early 2000s to 20146:

… and that was 10+ years ago! so put another 1-2 orders of magnitude over that to get an idea of where we are today!

Requirements for GPU Acceleration

To be fair, Foundry has a section called: "Requirements for GPU Acceleration”, so let’s say that I care about performance and I want to use GPU acceleration, who wouldn’t?

The Nvidia website listed doesn’t cover CUDA versions but what Nvidia calls: “Compute Capability”. After 10 minutes with Google, I found a Stackoverflow post7 that correlate CUDA 11.x to Compute Capability 3.5. With the help of Techpowerup.com’s GPU database8 and a different Nvidia website9, that should put us at “Nvidia Kepler 2 based cards”, available in 2013. All this to say: if you want GPU acceleration, have an Nvidia card from 2013 or newer. At least the AMD options have models from 2019 to 2023! Some of those might still be under warranty!

Foundry did something that I want to recognize as I wish all ISVs would do, and that is to list the exact configurations of the Workstations they have personally tested. This gives users a lot of trust in such models:

This is good! having said that, it still shows Workstations from 2019, 2022, 2019 & 2023 respectively. That’s 2-3 generations behind today’s available Workstations.

The Solution

I know this is probably the longest problem statement you have seen in your life, but it’s important to establish the size of the gap. Users need guidance and tools, that requires collaboration across ISVs and PC OEMs/CPU-GPU vendors.

This is what we need:

ISVs: (in today’s case: Foundry)

Please release workloads that stress key components of your applications. In Nuke’s case: Copycat, Multi shot sequences, low, medium and extreme complexity projects and the tools you use internally to measure Nuke’s performance

Benchmark Committees and Labs (SPECwpg, PugetBench, etc)

Incorporate such workloads or traces of such workloads into your testing suites

PC OEMs/CPU-GPU vendors

Run the ISV workloads across multiple CPUs, GPUs, system configurations and performance sensitivity analysis, then put out end user guidance in the form of system recommendations based your learnings

Nuke users

Are you a Nuke user and want to help the industry? reach out and help us build an ideal system configuration. The industry needs you, please pay it forward!

Conclusions

• Even if today, we focused on Foundry Nuke, it is time to get together as an industry to re-think Minimum System Requirements. This problem expands across the industry.

• I plan to continue analyzing other applications as new versions are released, my intention is not to focus on Nuke is just because it’s new and every new release, users are both excited and curious, about what they need to use the new capabilities, unconstrained, unleashed!

• Besides all this, systems this old will run O.S. like Windows 7. In October, 2025, Windows 10 will be EOLd and none of these systems will support Windows 11. Will that be the force that will help the industry to purge all this old specs? That might help once, but we don’t want to only move to the Minimum System that supports Windows 11 and stay there for a decade.

• Do you need any introductions to any of the key stakeholders needed to make this happen? do you have any roadblocks to make progress in this important mission? Please reach out, I’m happy to help!

Footnotes & References:

Share

This problem is not new

Imagine you are eager to start your project. Let’s use a Media & Entertainment workflow like 3D Modeling & Animation (although you could apply this to any other workflow), you or your company have invested thousands of Dollars into a 3D application like Autodesk 3ds Max or Maya, you install it on a Workstation that your I. T. department gave you and everything seems to be on to a great start!

You start placing a few low polygon primitives and the instant you start adding a little complexity to your 3D models or modifiers, your system starts choking and stalling, in a way that lets you keep going but becoming more annoying and making you less productive by the minute.

Let me illustrate this with a picture:

This happened to me in the 90s, I was in High School, I bought my first software application: 3D Studio R4 for D.O.S. I was so frustrated; I worked summers and holidays to save US$1,500 to by the student license (yeah back then students had to pay at least 50% of MSRP). Why I was able to model and animate cubes, but the second I made a sphere or a teapot, by poor 486DX 50 MHz couldn’t handle it?

I wanted to become a 3D Animator; instead, I went deep into performance tuning — a multi-decade journey to understand the performance impact of every component inside my PC. That journey led me to become an Electronic Engineer at Intel, instead of the Animator at Pixar that I had originally envisioned. By the way…

Back to the main story… Before I ordered the software, I had done my homework, after all, this was a huge financial investment, even by today’s prices. I had talked to the experts and my system met the needed requirements… The Minimum System Requirements (MSR)!

You might say: “Nice anecdote from 30+ years ago, we are in 2025, computer are orders of magnitude faster. Things have to be different today, right? … right?” Wrong. The complexity of the software has also increased, so when it comes to MSRs, we have not made a lot of progress.

The Problem with Minimum System Requirements

For decades, Software Vendors (ISV) have created Minimum System Requirements (MSR) to establish a minimum baseline to install and operate an application. The problem is that MSRs are not helpful — they are actively holding the industry back: Users, I. T. departments, PC OEMs and the ISVs themselves.

Let me illustrate my point:

• User’s experience with the application falls off a cliff the second their complexity increases from, you guessed it, the bare minimum. Creativity and experimentation die when users are afraid of suffocating their machine.

• I. T. departments need guidance to empower their users with computing tools that are capable of delivering the level of project complexity the company needs. They buy hardware based on MSRs and when the power user asks for more, you can see a sign coming out of their head that says one or many of the following: “Do you think you know more than the ISV? you don’t need more, you just want a fancier computer. Are you aware we are budget constrained?” and alike.

• The PC OEMs (Dell, HP, Lenovo, Puget Systems, BOXX, etc) and IHVs (Intel, AMD, Nvidia, Apple) test as much as they can, run some of the few benchmarks out there, but they are not experts at the applications, so there is a lot of subjective generational knowledge that gets used to position hardware solutions. The reality is that PC OEMs don’t have the deep application-level expertise and in many cases, lack the executive support to do in-house pathfinding to discover workloads from the latest workflows. Believe me, I tried for years.

• Finally, the ISVs, who set the lowest bar (MSRs) with the incentive of claiming that their application runs efficiently on very humble hardware, which sounds good, but it misguides users to feel the computer they already have is enough, while in reality, the latest version of the software is more capable, and therefore complex and computationally demanding. Bottom line, users get a subpar experience with the software.

Vague casts the wider net

To make things worse, MSRs have remained so vague that it lacks value and most importantly have not successfully provided a System Recommendation for an Optimal User Experience.

Let’s go through some examples from multiple ISVs’ MSRs:

• CPUs

  • 64-bit Intel® or AMD® with SSE4.2 instruction set
    For macOS, Apple Silicon arm64 or Intel x86

  • x86-64 processor, such as Intel Core 2 Duo or later

• RAM Memory

  • 8 GB of RAM (16 GB or more recommended)

  • At least 8 GB RAM

  • 32 GB or higher recommended and 64GB strongly recommended for fluid simulations

• Other

  • 3 Button mouse required

When it comes to GPUs, the MSRs are typically a list of Minimum VRAM, OpenGL, OpenCL and driver versions

But wait, some ISV’s also have “Recommended System Requirements”, aren’t those useful? Not really, or at least I haven’t seen a really good one. They are just “a little bit higher than the MSRs”. For example: Minimum: 4-core CPU, Recommended: 6-core CPU. Baseless and not even considering that there is a lot of performance difference across CPU generations & vendors, even at the same core count and frequency.

I particularly like the call above that implies that an “Intel Core 2 Duo or later” is what you need to get started. I was there for the launch of that CPU in 2006! 19 years ago. Would you dare to run the latest version of the most popular VFX compositing application on a 2006 system? I rest my case!

Industry strengths and weaknesses

Let’s talk about the areas of expertise for each type of industry player:

Note: I know there are exceptions, this is just a high-level summary of that I have seen in the industry.

• PC OEMs & CPU/GPU IHVs (Grouping both due to similarities)

  • Strengths

    • Hardware expertise

      • PC OEMs: Systems Level

      • CPU/GPU IHVs: Component Level

    • Strong testing & hardware validation capabilities

  • Weaknesses

    • Software Expertise. Application specific know-how

    • Access to Industry Experts mostly from marketing teams

• ISVs

  • Strengths

    • Software application specific expertise

    • Deeper Partnerships with Industry Experts (power users)

    • Deep Workflow understanding

  • Weaknesses

    • Hardware testing and validation for minor CPU/GPU/Memory SKUs

    • Most HW Certification process are costly, time consuming and focus on functionality rather than performance

• Industry Experts

  • Strengths

    • High expertise in Software applications and workflows (how they perform their craft)

    • Knowledge of application pain points

    • SW application experience to define low-to-high complexity workloads

    • Great at doing one-off experiments with their project. i.e.:

      • Is their PC good enough to get started? how does it “feel”?

      • Which application features should they avoid due to fear of lack of compute?

  • Weaknesses

    • Technical / computer knowledge is not their top area of expertise, their craft is

    • Limited access to multiple hardware SKUs

    • Finally, is not the user’s job to figure this out, but they could contribute a lot, so the industry should partner with them

Let me try to summarize this with a chart:

I applaud this FAQ developed by Lenovo about System Requirements, it touches on a lot of the points we cover today:

https://www.lenovo.com/us/en/glossary/system-requirements

Key players: Benchmarking Committees & Developers

Benchmarks are super important, they make it easy for the Hardware industry to run, however, not all benchmarks are created equal! There are synthetic benchmarks that compute random workloads for the sake of stressing the hardware and more often than not, they DO NOT correlate with the Real-World performance of the applications we care about. Therefore, I will pass-that-mark! (pun intended)

Let’s talk benchmarks based on Real-World workloads. The first think I must mention… Developing and maintaining a Benchmark is hard work and I want to applaud the effort of a couple of them that, in my professional opinion, are doing an excellent job!

• SPEC/GWPG: Short for The Standard Performance Evaluation Corporation, a non-profit corporation, establishes and maintains standardized, vendor-agnostic benchmarks and tools to evaluate performance and energy efficiency for the newest generations of computing systems.

  The SPEC group we care the most about for our industries is the Graphics and Workstation Performance Group (GWPG) develops a range of graphics and workstation benchmarks and performance reporting procedures. This committee is formed by members from participating companies like: Dell, HP, Lenovo, AMD, Intel & Nvidia. It’s important to recognize the companies that support this effort and the members that work on it beyond their daily job.

  GWPG produces 3 categories of benchmarks:

  • SPECviewperf: Focused on measuring Graphics Performance. This means that the workloads are selected to disregard other system dependencies and focus on what happens in the GPU. Important, yet very specific. GPUs should be selected based on it, but it is not complete picture for complete systems’ selection

  • SPECapc: This family of benchmarks is specific to each application. It combines multiple CPU, GPU and Storage bound workloads for Maya, SolidWorks, Creo, among others

  • SPECworkstation: This is the most broad and versatile benchmark, and it is great for system-level selection. It includes workloads for CPU, GPU and storage for multiple industries and applications. In my opinion, this is the most important of them all for PC OEMs, component IHVs and I.T. departments to select systems. They can always look at the Industry sub-score that is relevant to them including: AI & Machine Learning, Energy, Financial Services, Life Sciences, Media & Entertainment, Product Design, and Productivity & Development.

  • For more details, visit: https://gwpg.spec.org/about-gwpg/

• Puget Systems - PugetBench: Puget Systems is a computer OEM, very well regarded in the industry, not only because of their high-quality Workstations, but also because their capacity to give Objective (please read: data-based) recommendations. Like the rest of the industry, Puget Systems ran into the issue of how difficult it is to translate technical computer specifications into value for each workflow. The great differentiator is that Puget Systems decided to be proactive and do a lot to answer the hardest question: “So what?”

  • This workstation has many cores! So what?

  • This workstation has the latest GPU! So what?

  • What is the impact that each component and system bring to the user?

  • Just to give a few examples, for more information, visit: https://www.pugetsystems.com/pugetbench/

The common need that all benchmark developers have is simple, yet very complex: good quality workloads! They are the “thing”, the process, the application feature, you are trying to measure. After you have a good workload, there is work to do on automation, packaging, reporting, but without a good workload, you have…

ISV Benchmarking Initiative Award

If CavalryHQ had awards, it would go to: drum roll please…

Blender Benchmark

The Blender Foundation has been a role model to the industry, when it comes to helping the industry recommend hardware to make Blender fly! They provide a downloadable Benchmark, ready to run from Command-Line (CLI), which is very important for automation (remember, lab technicians don’t and should not need to be Blender experts to measure its performance). It also includes sub-scores for CPUs and GPUs. Could it be better? absolutely, I wish it includes other workloads not only related to rendering, but focused on Loading Times, Geometry Nodes and many other things that the user will do in front of the computer. Having said that, I with every ISV would have something like this.

This is how it looks like:

For more information, go to:

https://opendata.blender.org/

There are other efforts where ISVs collaborate with Committees to develop workloads, but they are the exception, not the norm.

There has to be a better way!

This article is the first of a series, with the goal to provide a proposal for a solution, or at least a big step forward in the right direction. Like any effort worth doing, it requires collaboration from all players mentioned.

This is the proposed solution:

ISVs

For your top applications, make available multiple High-Quality workloads*. Give permission to Benchmarking Committees to integrate them into their suites. Clarify that MSRs are barely ok to install the application and the level of complexity of the project should define the recommended hardware. Remove licensing barriers so anyone can run the benchmark

PC OEMs and CPU/GPU IHVs

When launching Workstations, CPUs or GPUs, use the performance data from such workloads ran at entry-level, mid-end and high-end systems or components so show the end users the impact of using different levels of hardware at corresponding price levels.

Industry Experts

Donate projects, classify them into low, medium, high or extreme level of complexity and why (model, textures, particle simulations, etc). Remember, nobody knows more than you and your pain points. Have a balance across your workflow, have workloads for Import, Edit and Export type of operations. Provide instructions to reproduce it. Bottom line, become part of the solution by participating at the committees, for example: https://gwpg.spec.org/participation/ or https://www.pugetsystems.com/pugetbench/development-program/

Benchmarking Committees

Continue to be clear about the help you need! great job at the websites I just mentioned! Remove barriers so Tech Press and Influencers can run them

* High-Quality Workload: I will likely expand on this in the future, but for now, let’s just say that a High-Quality workload is: repeatable, no-license required, has instructions for reproduction by a non-expert user of the application, covers different parts of the workflow (i.e. import/load, edit, export), includes variation toggles for low, mid, high and extreme complexity. Bonus: can be run from a command-line (CLI), if not, automation could be achieved from a trace capture

Conclusions

• We need to work together as an industry, no single entity or player can create great solutions

• Minimum System Requirements are just not useful, they are barely a recipe to avoid installation issues

• If power users, ISVs, PC OEMs and CPU/GPU/Storage/Memory ISVs support Benchmarking committees, WE ALL WIN! Better value-based guidance for users, better user experience with key ISV applications, better sales for hardware and invaluable feedback for future silicon and less subjectivity in the selection process.

Do you want to help but don’t know where to start?

Need an introduction to any of the key industry players?

Please reach out!

Part 3: OEM Partnerships, Conclusions & Recommendations

The role of the OEM Partners: “Together we go far!”

It was interesting to see that the DGX Station (the golden tower) shown at the keynote was not present at the showcase. Wait, what? Yes — Nvidia chose to highlight the implementation of the DGX Station platform by the OEM partners, specifically Dell, HP & Asus. Showing partner designs “Time-To-Market” is an important sign of industry adoption and readiness.

Nvidia also mentioned that other OEMs will follow, including BOXX, Lambda and Supermicro. The missing “GB10” player: Lenovo. While they had a full booth at GTC, there was no mention to a Nvidia GB10 or GB300 workstation. It will be interesting to see if they have a different plan or they will have a solution but are not ready to talk about it yet.

Just like there are DGX Station-based Workstations from partners, in their booths, there were DGX Spark-based Small-Form-Factor Workstations, again from Dell and HP. Great job by both companies being Time-To-Market! I expect that Lenovo will at least have a DGX Spark variant, the Lenovo ThinkStation P3 Ultra SFF is such a cool Intel+Nvidia machine, imagine it with an Nvidia GB10. Pure speculation at this point.

UPDATE (5/14/2025): My intuition was right, Lenovo just announced the ThinkStation PGX. More details here.

The Market Opportunity

So, with all this great technology, what do I think is the market opportunity here?

First, Nvidia has an opportunity with the Founder-Edition (please read: Nvidia-Branded systems). When a system is on a rack in the server room, maybe the looks don’t matter as much, but the DGX Spark and DGX Station are PC Clients, they will be on your desk and even if I’m not a fan of Gold-colored hardware, these systems are sure to call a lot of attention, be conversation starters and give developers bragging rights. Apple knows a lot about this.

Starting with the Mini PC (DGX Spark or OEM GB10 based), such prestige comes with a price premium of ~35% (extra ~US$1,000) versus what I expect from OEMs, based what I heard at the event and the Nvidia reservation website. The difference is actually smaller if you account for the BOM differences like storage, maybe other:

OEM implementations like the Dell Pro Max with GB10 (I think this is a placeholder name), and the HP ZGX Nano, will definitely be the volume runners, at least within the GB10-based options. It’s also great to see that Asus wants to keep growing in this space. Asus acquired the Intel NUC miniPC division, I’m sure such expertise has been very handy to bring the ASUS Ascent GX10 to market.

Are these systems here to compete with x86 SFF Workstations like the HP Z2 Mini or Dell Pro Max Micro? Based on current observations, that does not appear to be the case. This is a new category of SFF Workstation, it was positioned as a complement to your development laptop (or desktop). Can you use it on its own? yes, it has a display out, but I don’t think that’s the primary usage for it. Developers already have a development platform, typically a powerful laptop, there is less opportunity to try to replace such platform and instead try to “empower every developer” complementing whatever hardware platform they have.

There is more ROI for Nvidia to show the value of stacking multiple DGX Spark and I believe OEMs will be happier to sell multiple systems instead of cannibalizing one model with another.

Talking about replacement, there is a potential change on the host machine, now with the separate accelerator. Until now, there have been a strong reason for the Data Scientist or A.I. developer to buy a very strong configuration with a very robust Nvidia GPU. Now with the DGX Spark, I think the user will be very happy to trade a less robust configuration (specially GPU-wise) for better battery life or form factor, since now you can rely on the Accelerated Compute that the DGX Spark offers, just need to remote in to harness such power. The most common development environment is browser based, according to JetBrain’s State of Developer Ecosystem Report 2024. This facilitates the adoption of the MiniPC as a compute accelerator. We will know more, when we know how easy it will be to setup your environment. I would recommend Nvidia to have not only very solid setup tools, but also good solutions, support and overall documentation. Easy setup is key to ramp this product; many developers will be sharing their experience as soon as they (we) get their (our) hands on one. It has to be robust on Day 0.

How much volume will Nvidia sell?

The volume of these new category of device is to be seen, there are opposite forces at play; on one side, these Small Form Factor (SFF) Workstations are not general-purpose machines, they are here for a reason: Data Science / A.I. Development. While the rest of x86-based Workstations are general purpose systems, for many usages and workflows, CAD being the #1 usage with >40% of volume last time I checked Jon Peddie Research’s Workstation Report (highly recommended). x86-based Workstations are also used in multiple industries, like Media & Entertainment, Finance, Energy, Healthcare and Life Sciences, etc. This suggests that the DGX Spark is here to serve a niche, but such niche is growing roots into all industries, so will its volume be limited?

On the other side, the optimistic side of me knows there is a massive number of developers that don’t code on a Workstation! Starting with but not limited to, developers that code on MacOS, but also developers with traditional commercial client notebooks or even gaming machines (don’t get me started). If only a tiny fraction of developers adds an accelerator like this to their environment, this can be a big win for Nvidia and OEM partners! Also, this is a great system for users starting to get deeper into A.I. It will be exciting to see this product line ramp up!

I know this will be controversial, but I think the closest thing to a competitor for the DGX Spark is the MacStudio. They are not the same thing; they don’t run the same tools and clearly the MacStudio doesn’t support all the Nvidia A.I. software ecosystem. Having said that, both are small form factor desktops with Unified Memory space and powerful GPUs to actually do something with the data. Purists will rightfully call me crazy, but I have seen creators like Alex Ziskind running LLMs on a Mac Studio, quite successfully.

Enough talk about knives, let’s talk about Samurai/katana level swords! The DGX Station will easily cost 5-digits, I think if you even try to buy the 496GB of RAM memory for the CPU it would cost you a lot, let’s not even talk about the 288 GB HBM3e for the GPU.

Now, I’m not saying that paying a lot for a DGX Station is a bad deal, quite the opposite. Remember those “A.I. Workstations” that the OEMs have been selling since 2020? I think that is where there is potential for cannibalization of the market. In order to get 288GB of Memory for those advanced LLMs, you would need, oh wait, you can’t get that much with RTX Ada Generation, you are limited to 4x GPUs on a Workstation like the Dell Precision 7960, an HP Z8 Fury or a Lenovo PX, so are capped at 192GB of GPU GDDR7 memory after paying ~$25,000 in GPUs. The New Nvidia Blackwell-based RTX Pro 6000 have a very desirable 96GB DDR7 per card, so 3x GPUs would do the trick. Having said that, the HBM3e memory on DGX Station should be more expensive, yet higher bandwidth, more power efficient and lower latency vs GDDR7, so for A.I. Development, the DGX Station has a lot of advantages. And that is just the memory part of the equation, when it comes to compute, each RTX Pro has 4,000 (3x = 12,000) TeraFLOPs vs the 20,000 TeraFLOPs of the DGX Station.

I have seen vendors claiming that because their iGPU can address a lot of memory, they are good for A.I. but there is a balance between compute and memory, just because you can put a lot of food in your mouth, doesn’t mean that you can chew, swallow and digest it all. Think of it like a production line, but that is a topic for another day.

Conclusions + Recommendations to AMD, Intel & OEMs

First of all, THANK YOU for making it all the way here and walking with me until the end. Being my first deep down on CavalryHQ, I didn’t put this report behind a paywall, I kindly ask you to subscribe by entering your email below and supporting this effort in any way you can; if you can’t become a paid subscriber, becoming a free one or sharing with your stakeholders are greatly appreciated.

To show you my appreciation, here you have a bowl 🥣 of M&Ms (🔴🟢🟡🔵🟠) with the brown ones removed (🟤🟤) :-D (if you didn’t get it, check the last Reference below) Thanks for being here!

The Winners 🏆🥇🥈🥉

Of course, Nvidia, empowering developers with the most powerful GPU or multiple GPUs on the Intel-based Workstations, AMD-based Workstations and now giving Mac users a great way to add 100% Nvidia (CPU+GPU) to their setup!

OEMs also win, especially Dell, HP & Asus, if they keep working close to Nvidia, they can be very successful if this product line ramps up. Also, OEMs can now sell systems to Mac users, when was the last time you saw that? that is the textbook definition of TAM Expansion!

Most importantly, I think the Users are the biggest winners, these new systems will unleash their productivity with some upfront cost, but potentially lower TCO. Save in Cloud Credit’s cost early on, when exploring, experimenting and designing. Then, when ready, your code will be ready to deploy on an Nvidia-based cloud instance.

The Challenge: Differentiation

OEMs need to find ways to differentiate. When you have a platform with an Nvidia CPU, GPU, that also has soldered memory, I started to wonder, how will they differentiate? I assume everyone had exactly the same motherboard at GTC because of how new the platform is, I hope each OEM can have their own design, add more m.2 slots, PCIe ports, different thermal solutions, better software tools. Having said that, will there be enough volume to justify multiple designs? I’ll keep an eye on this as the launches and availability happen later this year. There are other non-technical ways to differentiate via bundling with other products, or technical ones like software tools to assure an easy and graceful transition when the accelerator (i.e. miniPC) is Online/Offline. I have a handful of other ideas that I’ll share with the OEMs when I have a chance to meet with them.

I challenge the OEM teams to use these systems internally as much as possible, this is a new way to work, and bridges will have to be built to assure smooth workflows and happy users. First one to figure this out will be rewarded in the market.

At risk

The x86 CPU vendors: Intel and AMD. Seeing Workstations with Nvidia GPUs is the norm, and GPUs do take a lot of the Share-of-wallet, but with Nvidia CPU and GPU should be a red flag to anyone paying attention inside these companies. If this product line ramps well, it will take time for the finance departments at Intel and AMD to notice that something is going on; initially it will feel like there is market softness, when in reality it is the effect of market cannibalization; a couple of years later, some market decline, by the time they realize what is really happening, it will be too late. Their time to act is now.

Recommended Action #1, call IDC and ensure they have a way to track this line up on the quarterly IDC Workstation Report. Same goes for Jon Peddie Research. The Workstation Market has very few consumption reports and, if this line up is not properly tracked, Intel and AMD will fight with a stealth competitor, that is already ahead in Share-of-Voice and Share-of-Compute. Only Jean-Claude Van Damme can win a fight while blind1 ;-)

Recommended Action #2, start measuring Share-of-Compute on key workflows in the biggest industries: Engineering (Product Design and Manufacturing and Architecture, Engineering and Construction) and Media & Entertainment (3D, VFX and Video/Audio Production). Only focusing on the workloads that run exclusively on CPU will blindside you on how relevant each of your products are to the power users in such industries. This relevance information is nice-to-have for the OEMs, but to Intel and AMD is way more critical to understand. How many workloads look like the SideFX Houdini example I shared on Part 1? What are other examples of the opposite balance where CPU is key? analyzing this is part of the CavalryHQ’s mission, reach out if you want to collaborate with me to understand this.

Let’s continue the conversation

Can I help you bring these products to market? Would you like an in-depth review of your latest Workstations?

Do you want an independent voice on your next salesforce training or webinar?

Do you need help selecting the best Workstation solutions for your company?

I’m one click away…

References:

• Nvidia DGX Station (2018): White Paper: https://images.nvidia.com/content/newsletters/email/pdf/DGX-Station-WP.pdf

• Nvidia DGX Station (2018): Data Sheet: https://www.nvidia.com/content/dam/en-zz/Solutions/Data-Center/dgx-station/dgx-station-print-explorer-datasheet-letter-final-web.pdf

• Nvidia DGX Station A100 (2020): Data Sheet: https://www.pny.com/en-eu/File%20Library/Professional/DATASHEET/DGX/DGX_Station_A100_Datasheet_PNY-WEB.pdf

• Nvidia DGX Spark: https://www.nvidia.com/en-us/products/workstations/dgx-spark/

• Dell Blog - Nvidia GTC coverage: https://www.dell.com/en-us/blog/pushing-boundaries-driving-ai-innovation-at-every-scale-with-dell-pro-max/

• Lenovo ThinkStation PGX: https://news.lenovo.com/all-new-lenovo-thinkstation-pgx-big-ai-innovation-in-a-small-form-factor/

• No Brown M&Ms reference: The Truth About Van Halen And Those Brown M&Ms : The Record : NPR

Be sure you already checked Part 1: The background

Part 2: Meet the 2025 Nvidia DGX <Workstation> products

For local development, as mentioned above, the top Workstation OEMs created with Intel and Nvidia, custom Data Science Workstations (later called: “A.I. Workstations” and such systems have been updated to newer generations of CPUs, GPUs and Software tools to make Developers more productive.

At Nvidia GTC, Jensen Huang announced the 2 new members of the DGX family: DGX Spark and DGX Station. I have to admit, just hearing the W-word coming out of a CEO brings chills down my spine and it is THE most boring drinking game I have ever come up with! :-D

Let’s talk about the new products:

DGX Spark

Formerly known as Project DIGITS, now officially called DGX Spark. Don’t be fooled by its Mini-PC form factor, the DGX Spark is very mighty, powered by Nvidia GB10 Grace Blackwell chip that combines the Grace CPU, Blackwell GPU and 128GB of Unified Memory. This means that both CPU and GPU can access the same memory space, reducing costly transfer operations over PCIe buses, that while great for graphics cards and peripherals, are not fast enough for XPU compute, where there needs to be collaboration between both compute subsystems. More on memory architecture later.

Before we go further, let’s talk about the bigger brother… Please set the green carper for the…

DGX Station

If you are a minimalist, this beast is not for you. The DGX Station is a Full-Tower Workstation, a Form-Factor designed for expansion and to have all the thermal management needed to unleash the last drop of performance out of any silicon on it. Obviously, it is way more powerful than the DGX Spark. How much? we don’t know for sure; we will have to wait for benchmarks when it is available. Having said that, it is significantly more capable, judging by its specs and Nvidia claims of 20X AI theorical perf vs DGX Spark. We will compare them later; you probably remember this picture from my LinkedIn post; I had the pleasure to witness this live:

About the system itself, this is the DGX Station (2025 Edition)

I specify that this is the 2025 Edition since there have already been previous 2 versions of the DGX Station:

• The first one in 2018, powered by a single (and already old for its time) Intel Xeon server CPU (20 cores, 2016 Broadwell architecture) and 4x Nvidia V100 GPUs. AI theorical throughput: 500 TeraFLOPS (Mix Precision)

• The second one in 2020, powered by a single AMD EPYC 7742 (64 cores, Rome architecture) and 4x Nvidia A100 GPUs. AI theorical throughput: 2,500 TeraFLOPS (unknown precision, assuming FP16)

• And now, the new DGX Station (2025 Edition), now powered by an Nvidia GB300, that combines 2 components: an Nvidia Grace CPU (72 CPU ARM cores) and an Nvidia Blackwell GPU. Such GPU does not have any graphics/display capabilities, so you will need an Nvidia discrete RTX Pro graphics card on a PCIe port. Well played Nvidia! AI theorical throughput: 20,000 TeraFLOPS.

Overwhelmed by all the specs? I think this chart will help you appreciate the progress Nvidia has done in such a short period of time:

The Memory Subsystem - an architectural note

Before we talk about the business side of things, I feel the need to make a clarification, it seems that there is a significant difference between the memory architecture of the DGX Spark and the DGX Station.

There are 2 concepts at play here: Unified Memory and Coherent memory.

Unified Memory refers to the fact that multiple devices (CPU and GPU) can access the same memory space, in other words they share memory. This has advantages since this avoids expensive (time consuming or just inefficient) copy operations of data between CPU memory (RAM) and GPU memory.

Coherent Memory on the other side, happens when the computing system, which has multiple memory levels on its hierarchy (i.e. cache, RAM) assures that all copies of the same data stay synchronized, so when any part of the system access certain variable, they all have the same information and when one changes, they all do.

The DGX Spark has 128GB of Unified Memory, so CPU and GPU on the GB10 can access the same memory and address such memory space.

The DGX Station on the other hand, claims to have a total of 784GB of Memory, but it a combination of 496GB of LPDDR5X RAM for the Grace CPU + 288GB of HBM3e for the Blackwell GPU. In this case, the 2 memory subsystems are Coherent and Unified, thanks to the NVLink-C2C Interconnect.

This can mean one of two things:

Option 1: The boring but OK scenario. There is pretty much a copy of all the 288GB of GPU Memory on the CPU RAM and when one changes, they both do. Also known as traditional coherence. Not the most efficient since 2 memory spaces duplicate and synchronize data, but it’s the traditional approach for separate subsystems.

Option 2: The paradigm shift! The NVLink-C2C Interconnect’s bandwidth (BW) is 900GB/s, this is 2.3X the CPU to RAM BW of 396GB/s. If the memory is truly Unified, the CPU would be able to read the GPU VRAM without making a copy to CPU RAM. The GPU might maybe feel a little tickle, but it should not affect performance too much since the HBM3e has an insane 8TB/s of BW. No, I won’t compare this to school buses, nor football fields, just picture that is 20X the BW of CPU-to-RAM. Also, this would be an extremely efficient use of memory while eliminating PCIe bottlenecks; that would be an architectural achievement to say the least.

Not only Nvidia’s slide claims the memory is Unified, but also the Dell launch blog claims the same.

The last time I saw an attempt to alter the memory hierarchy this much was with Intel Optane!

Ok, let’s switch gears and talk about the business side.

The role of the OEM Partners: “Together we go far!”

It was interesting to see that the DGX Station (the golden tower) shown at the keynote was not…

Let’s continue the conversation

Can I help you bring these products to market? Would you like an in-depth review of your latest Workstations?

Do you want an independent voice on your next salesforce training or webinar?

Do you need help selecting the best Workstation solutions for your company?

I’m one click away…

References:

1. Nvidia DGX Station (2018): White Paper: https://images.nvidia.com/content/newsletters/email/pdf/DGX-Station-WP.pdf

2. Nvidia DGX Station (2018): Data Sheet: https://www.nvidia.com/content/dam/en-zz/Solutions/Data-Center/dgx-station/dgx-station-print-explorer-datasheet-letter-final-web.pdf

3. Nvidia DGX Station A100 (2020): Data Sheet: https://www.pny.com/en-eu/File%20Library/Professional/DATASHEET/DGX/DGX_Station_A100_Datasheet_PNY-WEB.pdf

4. Nvidia DGX Spark: https://www.nvidia.com/en-us/products/workstations/dgx-spark/

5. Dell Blog - Nvidia GTC coverage: https://www.dell.com/en-us/blog/pushing-boundaries-driving-ai-innovation-at-every-scale-with-dell-pro-max/

Part 1: The background

A few of weeks ago, at Nvidia GTC 2025, the company announced and showcased two new and innovative Workstation platforms: DGX Spark (formerly known as Project DIGITS) and DGX Station. This is my take on what these products mean and the potential impact they can have. So many of you asked me about this launch! Will they capture a lot of segment share (MSS) in the Workstation Market? Will they cannibalize existing Workstations? and if they do, whose lunch would Nvidia eat? Let’s unpack all this!

What has been Nvidia’s role in Workstation until now?

From Pro-Viz Graphics to GPGPU to XPU

Before we talk about the new products, it’s important to build a baseline of the role that Nvidia has historically played in the Workstation Market. I don’t want to go too far into the past, but let’s just say that over the like couple of decades, Nvidia has patiently and persistently built a presence in the Workstation market that should be a case study in business schools. Today, approximately 4 of every 5 Workstations have a discrete GPU in them (in many cases more than one), and the vast majority (as in 9 out of 10) are Nvidia based.

Now, what is the role that such GPUs play in the system? It all started as Graphics Accelerators for 3D and video applications, then around 2010 they started shifting into not only graphics but becoming more programmable which make them great choices for parallel compute workloads, a trend known as General Purpose GPU or GPGPU.

Let’s move to the 2010s and Nvidia’s SW enabling efforts (and Hardware evolution of course) were mature enough to start Complementing the CPU in complex compute workloads like 3D rendering (think of 3D animation, not real-time like Video Games). Software Vendors (ISVs) rightfully saw the opportunity to put that powerful piece of Silicon known as GPU to actually compute (think export operations), beyond the traditional interactive graphics tasks in the applications. Such effort is known as XPU, as a way to say that you can compute on CPU and GPU. This transition is still on-going but every year, every generation, more applications and more (parallel) features are capable of producing exactly the same result (or negligible differences) on CPU (gold quality standard) and the GPU.

Applications like SideFX Houdini (sidefx.com), users’ favorite VFX tool for parametric 3D modeling, particle effects (smoke, water, fire, etc) are a great example of the XPU transition during the last handful of versions. For the ones paying attention, if there is such a thing as “Share of compute”, once a workload (specific function within an application) is XPU, even if both CPU and GPU are collaborating and giving it all, the majority of compute happens on the GPU. It is not uncommon when rendering on SideFX Houdini (Karma XPU renderer) to see System Resource Utilization like the image below, where Optix refers to Nvidia and EmbreeCPU to x86 CPU. Sure, that share of compute can be altered with higher end CPUs (more and newer cores), but that argument can also be made for having higher end GPUs and/or more GPUs on the systems (in the case of Desktop Workstations)

That is all great news for Media & Entertainment workflows, but in the last 5-7 years, Data Science/A.I. development needed very powerful systems, so in 2018 Nvidia launched the first DGX Station, a tower form factor designed to be at your desk, not on a rack, but more on this later.

Around 2020, Data Science and later on A.I. Development keep aggressively growing. When the world was blown away with OpenAI’s ChatGPT, developers had been developing A.I. for years. In order to satisfy their need for compute, the Workstation industry (CPU/GPU Vendors and Workstation OEMs) built on the concept of the 2018 DGX Station and launched Workstations with the specific purpose of facilitating A.I. Development: Meet the “Data Science Workstation”, later called “A.I. Workstation”. This kind of system was (and still is) a recommended configuration of CPU, GPU, memory and a strong layer of Software, from O.S.s to optimized distributions of Python, Libraries and Frameworks that could have orders of magnitude better performance vs “vanilla” (non-optimized) versions.

At the end of 2020, Nvidia updated the DGX Station to support A100 GPUs, Intel Xeons were replaced by an AMD Epyc CPU. With this refresh and the birth of the D.S./A.I. Workstation, this new class of system was ramping fast!

Everyone is a winner with this kind of solution, CPU vendors, especially Intel, added value with better core-count CPUs with modern Instruction sets like AVX-512 and AMX instructions. AMD got a CPU win on the DGX Station A100, and Nvidia was the real winner with a proven solution, regardless of the system vendor, to add not only a single high-end GPU, but up to 4 of them on a single system. And of course, the end user, the data scientist / A.I. Developer wins with a super powerful, local, low latency system; with a high upfront cost, but no cloud credits required for transmission, storage or compute, probably lower Total Cost of Ownership (TCO) in the long term.

Tons of system memory is necessary to clean and manipulate big Data Sets (RAM = 2-3X data set size) and tons of GPU Memory (VRAM) to then train the models. Having one of these “A.I. Workstations”, the user can now raise the limit on the level of complexity that could be achieved locally, especially if they had some funds for the project, because these systems can get pricey, a typical starting configuration would be >US$10K and with multi-GPU could easily go to US$30K. Each top end GPU goes for ~$7K-8K, times number of GPUs, you can estimate how quickly the system price can build up. The now “gen minus one” Nvidia RTX Ada generation packed up to 48GB per card, so scaling GPU memory (VRAM) is limited to 192GB when using 4 cards (~US$25K just in GPUs on such system)

There was also a remote/non-local compute cousin from Nvidia, the DGX-1. It was introduced in April 2016 as a rack mountable solution for A.I. Development. Calling this system a Workstation would be a stretch, due to its rack form factor and features, it would be more appropriate to call it a GPU-accelerated server. Regardless, its usage was the key point: At launch it was a GPU accelerated Rendering Node; then A.I. ramped up and the DGX-1 pivoted to A.I. Development, for A.I. to flourish few years later, I’d consider this kind of solution one of the seeds. Fun fact, “A. I.” was not mentioned once in the original DGX-1 data sheet.

Its specs were impressive even by today standards and although there was not a drop of co-marketing, it was based on Intel Xeon Processors, in combination with the latest Nvidia GPUs.

Then the update in 2018 was all about A.I. with the integration of Nvidia Volta architecture. Great for remote A.I. development, but there are advantages to local solutions and when you are Nvidia, why not split your hand and have products for both remote and local development? The Nvidia DGX-1 and DGX Station were the answer, until now…

Meet the 2025 Nvidia DGX <Workstation> products

For local development, as mentioned above, the top Workstation OEMs created with Intel and Nvidia, custom Data Science Workstations (later called: “A.I. Workstations” and such systems have been updated to newer generations of CPUs, GPUs and Software tools to make Developers more productive.

At Nvidia GTC, Jensen Huang announced…

Let’s continue the conversation

We can help you bring workstations to market (4P/GTM).

Would you like an in-depth review of your latest Workstations?

Do you want an independent voice on your next salesforce training or webinar?

I’m one click away…

References:

1. Nvidia DGX Station (2018): White Paper: https://images.nvidia.com/content/newsletters/email/pdf/DGX-Station-WP.pdf

2. Nvidia DGX Station (2018): Data Sheet: https://www.nvidia.com/content/dam/en-zz/Solutions/Data-Center/dgx-station/dgx-station-print-explorer-datasheet-letter-final-web.pdf

3. Nvidia DGX Station A100 (2020): Data Sheet: https://www.pny.com/en-eu/File%20Library/Professional/DATASHEET/DGX/DGX_Station_A100_Datasheet_PNY-WEB.pdf

4. Nvidia DGX Spark: https://www.nvidia.com/en-us/products/workstations/dgx-spark/

5. Dell Blog - Nvidia GTC coverage: https://www.dell.com/en-us/blog/pushing-boundaries-driving-ai-innovation-at-every-scale-with-dell-pro-max/

How did we end up here?

Hi, thanks for being here, I’m Hernán Quijano.

My passion for Workstations started when I was in High School in the mid-90s, my real passion is 3D animation, and due to its overwhelming hardware demands, sent me on a path to discover technology, high performance compute and a technical career. All this deserve a full detailed article, but my point is: “I have cared about Workstation for multiple decades!”

After a 22-year career at Intel, that included 3 countries, 12 roles in almost every division and culminated with the last 5 years leading Global Sales for Workstations, I want to build a safe space to discuss some aspects of the Workstation market and how we technologists can better empower the most demanding users across all Professional Workflows and industries and work together to move the world forward.

Who am I looking for to join this mission?

Do you belong to any one of these groups?

• You belong to an Engineering, Product Marketing, Sales or Go-To-Market of:

  • High-performance Workstation CPUs or GPUs at a Hardware Vendor like Nvidia, Intel, AMD or Qualcomm

  • Full workstation solutions at an OEM like Dell, HP, Lenovo, BOXX, Puget Systems, Supermicro, etc.

• You are part of a Software Vendor (ISV) of a key application in Workflows like Media & Entertainment, Architecture, Engineering & Construction, Product Development/Manufacturing, Data Science or A.I. Development

• You are a power user for any Workflow using one or multiple applications for the mentioned industries, i.e. key applications for 3D Modeling, texturing & animation, video & VFX, CAD, simulation or Data/AI frameworks & tools

I’m doing this to connect all of us and have discussions around how to work together to build better experiences from the tech industry for power users!

This is my invitation, let’s unleash users’ talents with technology, so they can move the world forward!

“Please stay, I want you, I need you” Benson Boone :-)

What to expect…

Initially, I’ll be writing like crazy, covering the state of the Workstation Market, my take on new hardware or software product announcements and trying to unveil the: “So What?” behind as many as I can take on. Then I want to keep my plan flexible since I want YOU to help me build the path forward.

Getting tactical: I need your support

1. Join the movement. I need you to follow/subscribe to CavalryHQ on:

   1. LinkedIn: I’ll be posting about new CavalryHQ articles and relevant posts from the industry.

      LinkedIn

   2. Substack: This is where the content will live. I expect all readers to join as free subscribers and please consider becoming a Paid Subscriber. Full transparency, the support I get will define how big and for how long I’ll be able to carry this effort on.

      Subscribe now

2. Contract my services. Do you need help with your Workstation Strategy? want to improve Workstation or CPU/GPU sales in 2025? Do you need to train your sales force? I can help you! Please Message me:

3. Technology Companies in the Workstation Market: Please connect me to your Press/Analyst Relationships team. In order to better represent your products, and be aligned with your launch timeline and events, please help me get in touch with your PR/Analyst team. Please send an intro message:

4. Other ways to follow: (more to come)

Bluesky

Any or all the items listed above are appreciated! Please support this effort as much as you feel comfortable with. I firmly believe we can drive a lot of innovation in the Workstation space.

Thanks!

Hernán Quijano