Upgrading workstation for CGI creation

I am currently running 2700x on new Aorus 570x board. I have 32gb memory (3000Mhz G.SKill Sniper X), Samsung EVO 970 Nvme and 1070 ti with 8GB video memory.

At the moment I make VFX for a TV program and often have to deal with high poly models in ZBrush, Maya and Substance Painter. If it would be possible to increase performance on those applications, that would be really great.

Now, my system does the job, it’s not bad by all means. But I would be really happy if certain things would be faster, that would really increase my productivity. For example decimating models in ZBrush or manipulating several or tens of million poly models.

Mainly I would love to have more performance when running Substance Painter, as sometimes I have to deal with dozens of 4K texture tiles, and my use case is specific as I need to see what I am doing in realtime in 4K. I know this is depending largely on available video memory, but I doubt even 11GB would be enough, so I suppose my only option would be to invest into some kind of monster card that would probably cost over thousand dollars.

Having more or faster CPU cores would help on ZBrush and Maya side of things, but could also somewhat increase Substance Painter, perhaps, but I’m not sure.

What would you do? Would upgrading to say, 3900x help my machine to do the job faster?

Since I still lack the practical experience (want to get into UHD-HDR videography soon) I cannot give concrete recommendations.

What I can give are general guidelines how to check what is limiting your workflow:

  • During stuff that takes long/is laggy, open system monitoring tools (HWMonitor, HWiNFO or even Windows Task Manager)

  • There you can check the CPU and GPU usage spikes: Does the software use all available cores, so is total CPU utilization >80 %?

  • If not, how many CPU cores is the software actually using (cough, looking at Adobe), are there maybe CPUs that can offer a higher frequency on that amount of cores?

  • Regarding GPUs: There you can assume that consumer-class GPUs (everything non-Radeon Pro on AMD and non-Quadro on NVIDIA) are being artificially slowed down by the manufacturer’s drivers and the application you are using.

  • Unfortunately you cannot say something like “Radeon Pro GPUs are always better for that amount of money!”, you should always look for benchmarks comparing AMD and NVIDIA GPUs using application xyz.

  • The latest Quadro RTX cards knock the Radeon Pros around like an abusive drunk uncle but in some scenarios, for example Radeon Pros that are based on Vega can use the system’s memory as GPU VRAM speeding things up a lot or even avoid crashes (cough, Adobe Premier, cough), so an AMD Radeon VII would be a great buy if you can still get them (they were only released since Navi was unexpectetly not ready to be presented at CES 2019).

Also, if you are doing actual paid work with your system and can’t afford or want to get your computer pre-built from a “proper” company with support contracts I would advise considering moving to ECC memory to further reduce the risk of system crashes or currupted files.

Thanks a lot for advice!
My Aorus Pro x570 should support ECC memory, so I have been thinking about that actually.
And great advice about Quadro RTX, that would perhaps be good idea down the line.

Is there a benchmark for Substance Painter? Later this week I should be able to test 3700X + 128 GB ECC + Radeon Pro WX 8200 (Vega 56-based with 8 GB VRAM), maybe I can check if that software can also use Vega’s access to the main system memory for more “VRAM” to run faster.

1 Like

I don’t think there is benchmark for that. It sounds really neat if Vega could access system memory like that, it might just well be the solution for this.

Here is a comparison between “Pro” and “non-Pro” GPUs with various professional application, maybe this can help you how to research this topic better:

But of course, access to the system memory is limited by the PCIe 3.0 x16 interface (about 16 GB/s) of current Vega-based Radeon Pro cards.

1 Like

Oh cool, thanks for this. It is surprising to see how RTX 2080 Ti is doing so well on what would normally be Quadro’s area, namely the viewport performance which I desire. It even seems to run CAD pretty well.
I wonder how that is possible considering that it’s still consumer card and should be running on consumer card drivers.
It might be that I will choose this card, looks like can’t go wrong really.
This would really help Substance Painter too I bet.

The situation might even be better than shown in that review video for current consumer NVIDIA GPUs since NVIDIA released “Studio Drivers” that now support stuff like 10 bpc color in windowed OpenGL applications.

Unfortunately, AMD had quite a tough time financially prior to 2017’s Zen launch (and now Zen 2 Ryzen and Epyc) so they had to make a gamble to focus their resources on making the best they could out of their CPU products and look at GPU stuff more or less as an afterthought.

I hope that they can do something similar to Zen on the GPU side of things in the near future so better competition between the two houses brings down prices for consumers.

1 Like

Hey, nyankocool. A few thoughts…

  • If you’re doing a lot of Zbrush work, I’d recommend bumping your RAM to 64GB. ZB is surprisingly efficient but that would give you more room to breathe if you’re regularly working with those kinds of polycounts

  • If you’re doing freelance/contract work on your own machine, you should consider dropping the cash for a more capable graphics card/cards

  • For Substance Painter I’d make sure to take a look at the new UDIM/tiling features they announced at SIGGRAPH to see if they made any core changes to hardware support/usage

  • If you’re not looking to have more than one or two GPUs in your machine, the Ryzen 3900x is an insanely good deal. Just be careful if you think you’ll need more lanes in the future for GPUs or m.2 drives or whatever. The Ryzen 3900x only has 24 lanes IIRC, which isn’t a ton for a card-heavy machine. You’d want to look at something like the Threadripper 2950x for something like that.

  • Last, if you have the option to work in Linux on your box with any of those packages, do it. Memory management–both with RAM and VRAM–is much better under Linux. Disk write speeds tend to be better, too.

1 Like

You should be able to see what it is that is maxing out while you are messing with the slowest program that you work with. See if your cpu is hitting 100% use, if your ram is hitting 100% use, gpu, vram, etc. And try to figure out what exactly is slowing you down the most that way you don’t end up wasting money on stuff that is “better” but doesn’t actually help with your problems.

That said, I think that you should consider Radeon VII. You can still find them some places, and 16gb of HBM2 would likely really help you out. I also think that adding more ram would be beneficial. So that is my bet. After those, I would guess CPU, then storage speed, because you already have a pretty decent cpu and nvme ssd.

So order of priority for upgrading, my guess would be

  1. RAM to 64 or 128 GB
  2. GPU vram
  3. CPU core speed/core count (ie 3900x)
  4. Storage speed (PCIe 4 nvme ssd)
1 Like

Thanks for this. That’s solid advice.
About the memory upgrade, my motherboard QVL list doesn’t list anything above 32gb. It maybe unlikely to get anything working at 3000Mhz or so, I suppose.
I wonder how much impact would memory speed have on those applications.

Thanks for reply!
Oh yes, Radeon VII seems to be still available in Amazon Japan for quite reasonable price.
This is something I definitely will consider. Only I notice that Radeon VII isn’t on Maya supported GPU list, but neither is 2080Ti, they do list only 2080. I wonder would this matter in practice though.

Unfortunately, motherboard manufacturers’ system memory QVLs are about as reliable as single-layered toilet paper.

Tested 4 x 32 GB Samsung ECC UDIMM M391A4G43MB1-CTD on an ASRock X570 Taichi with a 3200G* (only Zen+ part I have right now) and it worked fine with DDR4-3200 and loose “Auto” timings - so your 2700X and Gigabyte X570 motherboard should be fine and more so potential future Ryzen 3000 CPUs used on it.

*Note: As expected, ECC is then not active since only Ryzen Pro APUs support ECC :frowning:
Ryzen CPUs without an integrated GPU however offer Multi-bit ECC.

1 Like