Chaos Vantage multiple RTX GPU, different series

[davide_paolini]

Hi everyone, I currently use Vantage and I have 2 questions:

- I have a RTX 3060 and I'm considering to add a new RTX 4070Ti. I know that they have different architecture inside ( Ampere and Ada Lovalace) so thay can work weel togheter with Vantage or do you think that is better to have 2 GPUs with same archetcure/serie , for example RTX3060+RTX3090Ti ?

- I'm mostly interested in reducing the total time calculation of the final animation: in this case having 2GPUs (for example 3060+3060) will half the time needed or not?


Many thanks to Chaos for developing Vantage, such a beautiful software!

[npg]

Hi and welcome to the forum,

There is no issue with mixing RTX GPUs. (Just a reminder, Vantage allows maximum 2 GPUs)

The other question is more complicated and "Your Mileage May Vary". There are a few additional factors here:
- A couple of features are not supported in multi-GPU mode. Most importantly, the Light Cache (LC). The effect of not using LC depends on your scenes. You could try disabling it even on the single GPU (File->Advanced scene settings) to see the difference.
- We do post-processing on each sampling pass and this is done entirely on the main GPU. The denoiser takes almost 100% of this post-processing time. However, we try to balance the load between the two GPUs so that the secondary GPU does a bit more rendering work to compensate for the time spent denoising.
- The computation is split between the two GPUs, so if they are of different models one would be slower and we try to balance the load to keep both busy as much as possible. This is not perfect, so if the GPU is 99% busy when used alone, it could be something like 92% in dual mode.
- Depending on the complexity of the scene, the render resolution and whether render elements are used (EXR) the frame rate may be limited by the bus between the GPUs. So if you're rendering on a more modest resolution (fewer pixels to transfer) and the scene is heavy (more time spent rendering a single pass), the transfer of data between the GPUs could be fast enough to fit in the time that the sampling pass takes. If it is the other way around and the transfer is slower than rendering, then rendering speed will be capped by the rate of transfer. Using PCI-e v4 would be better than PCI-e v3 in this regard - it will allow a greater maximum FPS for a given resolution.

I wish the answer was simple, but as you can see there are several things to consider. So it would be best if you could borrow a GPU from someone to test how it works with your typical scene. It's possible that it will indeed speed up near 2x (if the same model, etc.) but if you're rendering 4K EXRs you might be limited by bus bandwidth. And if renders without LC are unsatisfying there's no workaround for that currently.

p.s. If you go for a mix of different cards, it's preferable to have the more powerful card as your primary (display) GPU. And another note: you would be limited by the memory of the smaller GPU, so if they have 12GB and 8GB you can load scenes that fit in the 8GB. There is no sharing of the memory pool.

[davide_paolini]

npg Many thanks for your comprehensive and precise answer

[davide_paolini]

ngp I did some test today that demonstrate that one fast GPU is better than 2GPUs and also that 4070Ti is near 3x faster than 3060 !!!!

So I report her the results because I think they could useful to you:


SCENE: CHAOS VANTAGE Vrayscene from SketchUp 2022 + Vray 6 (image attached)

WORKSTATION: AMD Ryzen Thredripper 3970X 32-Core 3.69 Ghz

RAM 64 Gb DDR6

TEST 01 : RTX 3060 12 GB

• Still Image 2K res / 1000 sample == TIME 4,32 min
• Still Image 2K res / 500 sample == TIME 2,14 min
• Still Image 4K res / 500 sample == TIME 8,50 min
• Movie 1080p / 512frame / 100 sample == TIME 17,05 min

TEST 02 : RTX 4070ti 12 GB

• Still Image 2K res / 1000 sample == TIME 1,22 min
• Still Image 2K res / 500 sample == TIME 52 sec
• Still Image 4K res / 500 sample == TIME 3,22 min
• Movie 1080p / 512frame / 100 sample == TIME 6,15 min

TEST 03 : RTX 4070ti + 3060

• Still Image 2K res / 1000 sample == TIME 1,34 min
• Still Image 2K res / 500 sample == TIME 54 sec
• Still Image 4K res / 500 sample == TIME 3,28 min
• Movie 1080p / 512frame / 100 sample == TIME 6,30 min

As you can see the better results are using the 4070ti alone, wich is faster than the other!!

It seems that having 2 GPUS when one is much faster than the other one it give worst results than the faster GPU alone.

[npg]

Thanks for sharing! Just one question: did you make sure to disable Light cache from the advanced settings dialog? Because it gets auto-disabled in multi-GPU mode and affects the comparison if it was on (which is the default) while there was a single GPU.

[davide_paolini]

npg Yes, sure, I disabled the Light cache in all the tests.

[thalles1990]

Bom dia, uma pergunta sobre a RTX 4070 TI, em V-Ray no modo RTX, a renderização é muito maior em velocidade também?

[davide_paolini]

thalles1990 Yes! It depends from the scene naturally, but my 4070ti is about 2x or 3x time faster in V-RAY GPU RTX mode comparing to the CPU mode with my AMD Threadripper 3970X 32-Core 3.69 Ghz. Obviusly you know that there are some little differences in reflections and in GI between CPU and GPU. mode