new video card questions

[s_gru]

hello. so i'm starting to look at new video cards to replace 4 GTX1080ti. and have a few questions.

1. i thought i read a while ago that multiple newer nvidia cards could have their memory combined/bridged so that max memory wasn't as much of an issue as it used to be. for example it used to be individually limited by the card with the lowest memory (doesn't apply to me, but wondering if i could get away with say a set of 10gb cards instead of 4 12gb or 4 24gb cards
2. i've got my eye on RTX 3080ti any thoughts on those cards for high-performance/usability etc.

my renders are typically still-frame, however i will continue to use the GTX1080ti in a 2nd machine if that's possible as IPR helpers or simply additional rendering power for my main machine. any thoughts on that?

i will also need to do the occasional animation so i would probably split it up between the two machines or just use all available per frame. not really sure what the best approach is.

i'm sure there's some issues between RTX and GTX. plus i'm checking my mobo to see what it can handle.

thanks, appreciate any feedback. it's been a while since i researched all this and i'm relatively new to Vray GPU.

[Muhammed_Hamed]

Hey s_gru

multiple newer nvidia cards could have their memory combined/bridged so that max memory wasn't as much of an issue as it used to be

Yes, this is called memory pooling. V-Ray GPU supports memory pooling on NVlinked GPUs, basically doubling your GPU memory
NVlink is supported on GPUs like A6000, A5000 and Geforce 3090s
If you have 2 Nvlinked 3090s, this means you have access to 48 GB of GPU memory

2. i've got my eye on RTX 3080ti any thoughts on those cards for high-performance/usability etc.

Try to find 3090s, it should be close in price and have double the amount of memory, plus NVLink support for future proofing the setup.
3080Tis are fine as well, depending on your projects, if you don't run into memory issues on your 1080tis I guess 3080Ti will be fine in your case

my renders are typically still-frame, however i will continue to use the GTX1080ti in a 2nd machine if that's possible as IPR helpers or simply additional rendering power for my main machine. any thoughts on that?

You can use the second machine as render node for your animations/high resolution renders. it should be great for that.
And the new machine is going to be plenty of speed for IPR and look-dev

i'm sure there's some issues between RTX and GTX. plus i'm checking my mobo to see what it can handle.

Yeah, mixing different GPU families is not recommended on the same job. best is to have a separate render job with Cuda mode for your older cards
On the old machine with 4x 1080tis you should use Cuda mode, and on the new machine with 3xxx use RTX mode
For animation usually what people do is render one frame per GPU, which is more efficient than rendering the same frame using 4 cards

so i'm starting to look at new video cards to replace 4 GTX1080ti

Keep in mind that 4 GPUs per machine is not as easy now because GPUs get much hotter, and are massive in size
2x 3090s + NVlink for the new machine should be great in my view. a single 3090 is faster than 4x 1080tis, not even considering RTX speed boost


Best,
Muhammed

[s_gru]

thanks muhammed! ok i forgot about the Cuda mode (which i use now) and RTX mode... interesting. I could still definitely use the 1080ti's which sounds great! i will look up rendering 1frame per gpu i remember reading about that, but haven't tried it yet. so maybe then 2 3090's sounds like the way to go. i'll check out dimensions and sizes for cards.

[Muhammed_Hamed]

Good luck!

If you have questions let me know

Best,
Muhammed

[haniel1107]

Hello Muhammed_Hamed
may I add a similar question? I am using Cinema 4D.

I am mainly rendering with CPU, as Cuda crashes very often.

I have 2 x 2080 Ti.

Regarding Interactive (VFB) rendering, what is better or what responds faster?

2 x 2080 Ti or 1x 3080 Ti or 1x3090?

Would the investment be worth it?

Thanks for your help.
Best,
Hani

[Muhammed_Hamed]

I am mainly rendering with CPU, as Cuda crashes very often.

I have 2 x 2080 Ti.

Hello!
This is not normal, does this happen for all of your scenes
I'm able to use the GPU IPR for hours for my look-dev, it can be quite reliable
Any crashes we are able to reproduce, will get fixed.

Regarding Interactive (VFB) rendering, what is better or what responds faster?

2 x 2080 Ti or 1x 3080 Ti or 1x3090?

Would the investment be worth it?

a 3090 is faster for sure, and has more VRAM, it is a great GPU
3080Ti is also faster than 2x 2080Tis, specially in RTX mode. 3000 series GPU has second Generation of tensor cores, much better than 2000 series in this

Best,
Muhammed

[haniel1107]

Thank you very much Muhammed_Hamed !

The crashes happens only when rendering big scenes.. I think because of the ram... I haven't studied it in detail to be honest
IPR is stable..

But I am happy to render with CPU, with TR 3990..

But I like to have a better IPR responds at large scenes.. therefor my question..

When you say it is worth a RTX 3090, I think I'll go with a 3090 and replace the two 2080TI..

Thanks again,
Hani

[Muhammed_Hamed]

But I am happy to render with CPU, with TR 3990

It is a great CPU
One note is that if you add a 3090 you will need to use V-Ray GPU to see any speed benefits. There is a Hybrid mode where you can use both your CPU and GPU for rendering, this is recommended for you specific case

When you say it is worth a RTX 3090, I think I'll go with a 3090 and replace the two 2080TI

Yup, as long as you are willing to use V-Ray GPU (Cuda/RTX modes)
Best case scenario is leaving one of the 2080tis and using it together with the 3090 if possible
So the 2080ti is used for viewport performance and your monitors, and 3090 memory and performance are all dedicated to your GPU rendering

On another note, I have some tips on optimizing GPU memory for rendering. See my post here

Best,
Muhammed

[haniel1107]

Muhammed_Hamed
this is great, thank you!!

Do you think a 1300W PowerSupply can handle both cards, 1x 2080 Ti and a 3090?
And if I leave a 2080 Ti in for monitors... which card will be responsible for the IPR?

Thanks for your big help so far..

Best,
hani

[Muhammed_Hamed]

Do you think a 1300W PowerSupply can handle both cards, 1x 2080 Ti and a 3090?
And if I leave a 2080 Ti in for monitors... which card will be responsible for the IPR?

1300 Watts are enough yes, for 2 GPUs + 3990X

You can specify which card is used by V-Ray in the device manager in V-Ray's renderer settings, if you go to Edit then preferences and scroll to V-Ray under Renderers
You can specify which card is used by C4D itself for its viewport as well, In Nvidia's control panel, check the link I shared above it has some screenshots about that

You are welcome

Best,
Muhammed

[s_gru]

Hello again, where do i start researching how to do an animation rendering 1 frame per card (i have not upgraded cards yet, still using the 1080ti's)

EDIT: sorry i searched for the right phrase and found this thread! https://forums.chaos.com/forum/v-ray...secondary-card

[s_gru]

Hi Muhammed, so how do you do it via the standalone renderer? from that other thread i found - "...this is also possible with commend line and Vray standalone, but easier with the tools I listed above." thanks!

EDIT: so i purchased a copy of the vrsceneGUI. i set the number of batch segments to 4 then checked the box for multi-GPU mode then hit Export Batch. i double-clicked each of the batch files and that seems to have launched 4 v-ray frame buffer windows and 4 cmd windows where it's listing out what is happening.

so is that how you do it? i do not have distributed rendering enabled (it's set to Do No Use Dist. Rendering ... but if i wanted to could i? and add my other machine?

something seems suspicious because i thought it auto-detected the # of cards in my machine and each of the batch files compared only showed 1 difference between them and that is the -frames= tag. in other words it seems like it's just using all 4 cards together instead of individually?

EDIT2: my cards are pretty pegged to 100% utilization though. like i said though is it really 1 frame per card?

[Muhammed_Hamed]

Hello s_gru

Sorry for the late reply, I had a very busy week.

so is that how you do it? i do not have distributed rendering enabled (it's set to Do No Use Dist. Rendering ... but if i wanted to could i? and add my other machine?

Yeah, you are doing this right! now each card is rendering one frame as a separate job
You can experiment with DR to start new jobs for more animation frames,

EDIT2: my cards are pretty pegged to 100% utilization though. like i said though is it really 1 frame per card?

You can verify as well, if it is 4 different jobs shared between the GPUs you will have 4x higher memory usage
We are having internal discussion about this workflow at the moment, it is quite an important topic. I will update the thread with more information soon

Best,
Muhammed

[3deeit]

Hello s_gru

Yeah, you are doing this right! now each card is rendering one frame as a separate job
You can experiment with DR to start new jobs for more animation frames,

Hi Mohammed, I find this very interesting as I usually render GPU (3ds max, backburner) in parallel with 2x RTX3080's in one box. Is there a way to render 1 frame per GPU using the 3ds max/backburner interface?

I just did some tests, and it does make sense if I can render one frame per gpu - it would increase my speed by 20%. However I'd prefer to stick with the 3dsmax/backburner method

Thank you!
Phil
3deeit.com

[Muhammed_Hamed]

Hi 3deeit

I don't think there is a way to do this with Backburner, it has to be a vrscene manager or Deadline. I will discuss this topic with the team soon

Best,
Muhammed