Grafikkarten

[max_romju]

Yes, one shortcut for "-d 0" and one for "-d 1".
It's usually easiest to copy and modify an existing shortcut.

Greetings,
Vladimir Nedev

HI vladimir.nedev, How can I comprehend one vantage instance Supportmaximum of two Graphics cards, and when I maked two vantage instances, a computer has 4 Graphics cards will work well indeed. am I right?

[npg]

We haven't tested it, but yes, it sounds like you should be able to do this. You would of course put more strain on your CPU RAM and especially the PCI-e bus between the cards. I'm not quite sure how 4 cards share the PCI-e bandwidth, but PCI-e is a bottleneck even with just two cards. It's actually the main reason we have to limit devices to 2. If the GPUs have NVLINK it would be faster.

[Muhammed_Hamed]

but PCI-e is a bottleneck even with just two cards

How did you guys get to this conclusion? Is this different about Vantage compared to say Vray GPU or GPU rendering in general ?
PCIe Gen 3 x8 is more than enough bandwidth for any GPU, in my case here I'm testing with RTX 3090 SupremeX, it is the fastest 3090 in the market out of box (I get the highest score on the benchmark page)
Using 2 of these SupremeX 3090s on PCIe Gen 3 x8 I don't lose any performance.. compared to running a single card at PCIe Gen 3 x16
I'm using Vray benchmark, Furmark, Heaven, Octane Bench..etc scores are exactly double what I get with a single GPU at x16

I'm not quite sure how 4 cards share the PCI-e bandwidth

Depends on Motherboard, but plenty of consumer boards on X99, X299, X399, TRX40 is able to run 4 GPUs at x8 speeds PCIe Gen 3
Basically most Threadripper Motherboards, and Core i9 XE baords.. Dual Xeon boards as well
Again PCIe bandwidth here is out of question, it never gets in the way and scores are near 100% linear compared to running a single GPU at x16

This was my build a few years back, 5 GPUs all running at x8 with linear scaling and no bandwidth issues. 4 and up to 8 GPUs (per machine) are still common amongst the GPU rendering community

If the GPUs have NVLINK it would be faster

I don't understand this either, it doesn't seem to make any difference on my NVlinked 3090s, turning off NVlink the performance is identical
Do you have a scene I can benchmark the performance with ?

I can test with 3x 3090s and 1x 3080 all at PCIe Gen 4 x8 .. if there is a way to enable 4 GPUs in Vantage.. or with 2 instances of Vantage

[vladimir.nedev]

Is this different about Vantage compared to say Vray GPU or GPU rendering in general ?

Yes, it's very different, Vantage updates the image much more often compared to V-Ray CPU/GPU and in order for this to happen, each GPU has to send its part of the image to the main one.

Greetings,
Vladimir Nedev

[npg]

It has been discussed on the forum already. The PCI-e is a bottleneck because it takes quite a long time to transfer the many megabytes of rendered buffers each frame (several render elements are used in postprocessing). If the raytracer is faster than this transfer speed, it has to wait for the transfers before rendering another frame (e.g. 10ms to render, 16ms to transfer). If the scene is complex, the raytracer will need more time and you won't notice the bottleneck (e.g. 30ms to render, 16ms to transfer). The transfer time grows with render resolution, so the maximum achievable FPS goes down with resolution on multi-GPU.

[Muhammed_Hamed]

Thanks!
It makes a lot of sense now, I'm going to try one GPU only with my setup

(e.g. 30ms to render, 16ms to transfer). The transfer time grows with render resolution, so the maximum achievable FPS goes down with resolution on multi-GPU.

It is not just about rendering speed, I see

[Muhammed_Hamed]

I have done more testing, while monitoring PCIe bandwidth for each card using Bus usage graph in AfterBurner

On using a single GPU,



And on dual GPUs,



Vantage uses 7 times more PCIe bandwidth compared to running on a single GPU, in my case here I have my GPUs running at PCIe Gen 4 x16 which is best case scenario (3990X CPU + TRX40) and verified the PCIe link in GPUz
So on paper anyone on PCIe Gen3 x8 is gonna get bottlenecked by PCIe bandwidth, see this table


Basically anyone that didn't adapt to PCIe Gen 4 (11th generation Intel CPUs, 3rd and 5th generation Ryzen CPUs, and 3rd Gen Threadrippers) might see a bandwidth bottleneck on 2 GPUs..
Very cool seeing Vantage take advantage of such new technology

[max_romju]

Yes, they should work together, however you are limited by the smaller amount of VRAM when you want to use them together. In this case 10 GBs for the 3080.

Greetings,
Vladimir Nedev

hi vladimir.nedev ,I have 2 3080ti,each 12g vram,is there any ways to share the vram to more than 12G? now I couldn't set big resolution

[max_romju]

by the way ,when i render,one card is inefficient，am i set wrong?
I use Remote Desktop,havn't link any monitor.

[max_romju]

[Alexander.Atanasov]

Hi max_romju,

There is currently no way to share memory between GPUs. As for the GPU inefficiency, the Windows task manager is not very reliable for that. Use instead MSI Afterburner or GPUz.

Best regards,
Alexander

[vladimir.nedev]

hi vladimir.nedev ,I have 2 3080ti,each 12g vram,is there any ways to share the vram to more than 12G? now I couldn't set big resolution

No, it's not possible at the moment.

by the way ,when i render,one card is inefficient，am i set wrong?
I use Remote Desktop,havn't link any monitor.

There is nothing that you could set wrong I believe.

Can you try monitoring the activity with GPU-z ? I am not sure Windows Task Manager shows the correct values, it's complicated.

Greetings,
Vladimir Nedev

[alex_obrist]

Hi all, is it possible to use 4 graphics cards for Ventage ? if so how do I connect the other two because I can't see them in Render Devices ?

[npg]

No, Vantage is currently limited to a maximum of 2 GPUs.