Advice needed; Building new workstation for Keyshot

Hi everyone,

I’m looking to build a new PC workstation specifically for high-resolution product rendering in KeyShot. I currently use an Intel i9-9820X with an RTX A4000 and 32GB RAM. While the GPU is still okay, the CPU is aging fast and becomes a bottleneck, especially when working with large, high-poly scenes (30+ million polygons), or semi-complex materials.

At the moment I do not have a clear budget in mind. I want to work out a few different options, cost and performance wise. Lets say;
Tier 1: 5000,-
Tier 2: 10.000,-
Tier 3: 15.000,-

One thing I find difficult is figuring out what actually results in noticeable time savings, and more importantly, how much. Benchmarks are helpful, but translating those scores into real-world gains (especially in KeyShot) is tricky.
My benchmark sheet; Keyshot Benchmark Data - Google Sheets

I’d really appreciate any advice on what parts of a build give the most real-world time savings in KeyShot, and what kind of setup you would recommend based on your experience.

Thanks!
Merijn

My old systems were 6 years old, dual xeon gold hp Z8 workstations. 10k each when new.

Just took delivery of 3 new Boxx Apex workstations with intel 285k CPU and RTX5090 gpus. $7500 each.

the difference of course is amazing. To be honest, the CPU speeds for rendering are about the same, but that’s comparing 6 year old dual xeons @$6k a pair to the new 285k @ $499. We switched to GPU rendering and I still can’t comprehend how it can render so fast. We went from rendering scenes overnight for 4+hours to rendering the same scene in about 20 min. Insane.

The good thing is that we use these machines for AE and PS as well, so the faster clock speeds of the 285k is super nice and AE really appreciates the faster core speeds.

RAM type or speed is not going to matter much, just get the most and best you can afford. Last workstations had 64gb and never ran low, new workstations came with 96 by default.

Going from SSD 2.5" sata drives to M2 NVME type drives made a HUGE difference in load speeds. Boxx uses the Taichi based motherboards in this, and its got 6 m2 slots.

We went full bore with the 5090’s as we need these workstations to last a long time. Oscar has done a lot of testing too saying that the 5090’s may not be worth the extra cost, and he may be right. We are banking on the drivers and KS extracting more and more performance out of them though.

In the end I did not see the gains from spending extra on Threadripper or Threadripper Pro chips, and their stability issues that they were still having would not have worked well in a corporate environment. In my opinion if you are going for GPU rendering, base your budget on that, and build the system around it. Figure out what is the largest time suck you deal with on a daily basis, and that for us is rendering and look dev while setting scenes up. Change a matieral, wait to see what it looks like. That workflow due to the GPU’s has sped up at least 50%.

My gut says that there will be a larger and more perceptible speed difference between $5k and $10k budget than there will be between the $10k and $15k budget. This has not been updated in a while, but its worth a look. My own gathered spreadsheet.


1 Like

Great post Matt! And well, since there are no new 4090s a 5090 is nice because of the extra VRAM but the jump from a 3090 > 4090 is much larger in speed than the jump from a 4090 > 5090.

And Merijn, since you’re located in the Netherlands, a look at the Tweakers pricewatch makes sure you don’t want a 4090 since they are listed now as more expensive than the 5090. Of course that makes no sense, I would happily trade a 4090 for a 5090.

As Matt mentions, I think the difference between a € 5000 machine to a € 10000 is more noticeable than between € 10000 and € 15000.

And even for around € 5000 you can have a computer with the fastest consumer CPU from AMD/Intel, 192GB of memory and a 5090. If you would put in an extra 5090 or 2 extra 5090s the price will rise with like 2500 for a card and your render speed will double or triple.

Multiple GPUs will make it more expensive because of the custom liquid cooling needed or you need a big case and buy some AIO cooled 5090s, you still need to have the space for multiple radiators though. And multiple cars will only increase speed, not the amount of VRAM.

Instead of multiple cards in one machine you could also have multiple PCs and a network license. If nobody works behind those other PCs they can be pretty simple besides the GPU. I’ve no experience though with having multiple machines running KS over network and the network itself will always slow things down.

All the workstation kind of hardware is actually a waste of money if you are using it mainly to render and use other graphic software. A lot of suppliers will try to convince you otherwise, but it really makes no sense. It might would if you run very critical simulation software which also often uses CPU cores but for graphics the consumer grade of these days is perfect. Nvidia and SolidWorks like you also to believe otherwise, it’s not based on anything.

While memory is not that expensive and I was also orientating a bit I did notice that using for memory slots can cause instabilities if you want to run the ram at max speeds. Which means that if you like it to run full speed you are tied to 96GB of ram (2x48). As Matt says it’s not really noticeable if your ram is running at somewhat lower speeds and if you have applications that like more ram, slow but more will win from less but faster.

I do wonder if it’s actually the CPU that becomes a bottleneck with your current system or the amount of RAM. I’m still on a i9-9900K with 64GB and I notice more that if I for example try to photogrammetry software it’s my RAM what holds me back which will make it crawl in the end.

It’s also fun to ask CoPilot about hardware and what your plans are. I did it about a week ago. It actually was pretty clever when I asked I wanted to house multiple cards but it also made mistakes. Like it suggesting a motherboard that doesn’t exist. It must felt embarrassed when I asked for the website :wink: It did however consider my wish to have more spacing between the two first PCIe slots so the intentions were all good.

I just got this link from a friend from Finland, it’s about new Threadrippers which might be interesting is you do use very heavy CPU related tasks as well.