Planning a new render rig

Hurdy3D

Hey,

I'm currently planning a new render rig.
Maybe you can give me a few tips.

I will propably get two RTX 2080Ti and want to reserve a thrid slot for maybe later a quadro or the Titan with 24GB or a GPU with more VRAM of the next generation ;)
Yes, I know, I can only use the VRAM amount of the lowest GPU. But the idea would be to use all three togehter and if I have a heave scene, I would disable the 11GB VRAM GPUs.

Does with a RTX GPU the CPU still matter (so, should I use with RTX the GPU and CPU together while rendering?)

I think I'll need 32 or 64GB RAM.

Some CPUs or Mainboards have an integrated GPU. Would it make sense to have such a addtional GPU as the default GPU for the Desktop and the DAZ3D viewport? I experienced that espiacally the DAZ3D viewport steals a lot of VRAM from the render GPUs, if these grahic card is also used for the viewport.

Which CPU, Mainboard and RAM modules would you recommened?

3141592654

I do not have an RTX, but I have a quadro M6000 (24Gb vram) and an Intel i9 with 14 cores. When I add the CPU to the GPU it speeds up sometimes up to 30% faster. However, you NEED to have a good cooling system (think radiator) to have that many cores working hard. I have my system set up on an MSI X299 motherboard.

Sevrin

Here's an article you may find helpful.  It doesn't mention Iray, and predates the recent RTX-Iray announcement, but it's still useful as it covers all aspects of our rendering machines.

https://www.cgdirector.com/best-hardware-for-gpu-rendering-in-octane-redshift-vray/

tj_1ca9500b

You can usually fit a fair amount of stuff into a 1080Ti or 2080 Ti, so while 24 GB of VRAM is nice, if your scenes aren't too complex, the 11 GB should be plenty.  Plus, you'll probably get faster render times with the two smaller cards working together than the RTX Titan can crank out on it's own.

If you like to build complex scenes, though, that regularly exceed the 11GB, yeah that's when spending twice as much to grab the RTX Titan makes more sense.  Sure, you can attempt to shoehorn scenes into the VRAM using Scene Optimizer and such, but of course that takes extra time to do.  So your strategy of 'down the road' isn't a bad one

In the Iray benchmarks thread, it's noted that as far as actual render times, x4 PCIe is marginally slower than x16 PCIe, as most of the work is done 'on the card' so there's not a lot of cross-talk with the CPU.  That being said, every little bit helps, so if you go the Threadripper route, well 3 cards at x16 each, with leftover PCIe lanes for storage just isn't going to be an issue.

7nm Threadripper STILL hasn't been given an official release date.  The thought is October maybe, but we don't really know.  That being said, really even a 1900x 8 core 14nm Threadripper may be enough for your needs, and those can be had for fairly cheap.

7nm Ryzen Matisse is absolutely amazing, and the 3950X 16 core AM4 part should be arriving in September.  We already have the 3900x 12 core part that has really decent CPU rendering scores.  HOWEVER, you are limited in PCIe lanes with an AM4 build.  As I noted, x4 or x8 PCIe isn't that big of a deal for rendering, but the x570 boards with 3+ 'full length x16 slots' tend to be a bit more pricey.  Myself, I'm holding off for 7nm Threadripper, just because. 

As for PCIe gen 4 on X570, well right now the main advantage will be in faster PCIe 4 NVME drives, if you want faster boots and such.  So it's not a big deal, but as far as 'looking to the future' yeah there's that.

On the flip side, large core counts give you the option of 'that render just went CPU only, but I have lots of cores so I don't mind just letting the render run anyways' option.  Sure you may be talking 4x or more slower, but when you factor in the time you need to spend to optimize your scene to fit in your graphics card and such, well sometimes just using this as an opportunity to take a break for food, etc. while the render bakes might make sense. Most of the time, though, yeah put those dual 2080 Ti's to good use, they will be MUCH faster, so optimizing your scene is probably worth the effort.

@3141592654

I'm curious as to your comment.  In the Iray Benchmarks thread, we haven't seen even large core count CPUs making much difference in actual render times when doing CPU + GPU, so your 'up to 30% faster' comment seems to fly in the face of that.  Would you care to elaborate?

Hurdy3D

thank you for all the answers so far.

I thought of getting an AMD Ryzen 7 or 9... but than I figured out that these CPUs has only 24 PCIe Lanes... which is good for one single RTX 2080ti. But after the performance drop is marginal, I can still use 3 GPUs in 4x mode, right?

If I want something future proofed I should propably get a board with PCIe 4.0 slots, if the next GPU genreation is maybe PCIe 4.0.

From an other source I heared that using CPU and GPU together is mostly slower than using only the GPU. So I shouldn't borther about the CPU speed/cores for rendering?

 

Dim Reaper

If you are thinking of having more than 2 cards, I would recommend getting a full tower case.  I have a midi tower case, and although ti has plenty of space for drives and fans, graphics cards are getting bigger with each generation.  I did an upgrade today, so right now I have a 2080Ti using one PCIE slot and covering another.  I also have a Palit 1080Ti that takes up 2.5 slots.  So although I have potentially 5 slots, I can't fit another card in there.  I removed an older 980Ti card, but in order to use it again I'll have to use a riser extension and either modify the case to mount the card somewhere, or buy a bigger case and then go through the hassle of moving everything over.

As for your question about CPU's in rendering with the GPU, my CPU is getting a little old now - an 8 core 5960X - and I never render with it in iray because it adds very little to the overall render speed.

tj_1ca9500b

I just wanted to note that I DO think that having 8 cores in Daz vs 4 cores, with SMT on top of that, does make a difference in working with the Daz Studio interface.  It may not help much with render times, but my 4 core seems a bit slower on the responsiveness than my 8 core did when it was working, when using the non-Iray viewport modes.  I'd be curious to hear from others as to whether 12 or more cores helped more on the Daz Studio interface responsiveness vs smaller core counts.

It may help with scene load times as well.  Again, I think I saw faster load times with the 8 core, not significantly faster unless we were talking large scenes, but still noticeable.  Again, what have other people observed along these lines?  And also with larger core counts?

I'm using the same NVME drive on my 4 core system that used to be in my 8 core laptop, so in this case access and read times should be similar from the drive performance standpoint as far as my observations go.

Hurdy3D

tj_1ca9500b said:

I just wanted to note that I DO think that having 8 cores in Daz vs 4 cores, with SMT on top of that, does make a difference in working with the Daz Studio interface.  It may not help much with render times, but my 4 core seems a bit slower on the responsiveness than my 8 core did when it was working, when using the non-Iray viewport modes.  I'd be curious to hear from others as to whether 12 or more cores helped more on the Daz Studio interface responsiveness vs smaller core counts.

It may help with scene load times as well.  Again, I think I saw faster load times with the 8 core, not significantly faster unless we were talking large scenes, but still noticeable.  Again, what have other people observed along these lines?  And also with larger core counts?

I'm using the same NVME drive on my 4 core system that used to be in my 8 core laptop, so in this case access and read times should be similar from the drive performance standpoint as far as my observations go.

As far as I know, DAZ can only use two CPU cores at the same time (expect for rendering).
Maybe your new CPU has more HZ?