2x Nvidia RTX 4070ti

[eragon16gaming]

Hi. 

I have 2 4070ti's in my system. 

Every monitoring software I use tells me only one of them is ever be used at once. 

How do I configure my system to best utilise the 2 cards? 

My job has me running Blender and Keyshot on a daily basis for around 10 hours a day. 

Any help would be appreciated. 

[Kilrah]

Blender should be able to use both during a render but that's about it, most things can't use 2 GPUs.

[venomtail]

SLI isn't supported. Dead tech, abandoned for gaming at least.

[igormp]

36 minutes ago, eragon16gaming said:

My job has me running Blender

On blender you should select both devices under Optix/CUDA settings for the cycles renderer.

 

Your other stuff likely won't make use of both GPUs.

18 minutes ago, venomtail said:

SLI isn't supported. Dead tech, abandoned for gaming at least.

I don't think they're talking about games here.

[venomtail]

19 minutes ago, igormp said:

I don't think they're talking about games here.

Doesn't multiple GPU use still require a bridge? Communication via PCIE lanes through CPU is slow that's why the bridge exists. I know for gaming the SLI bridge is dead that's why I've mentioned it but no idea for its support for productivity apps. If something does support multiple GPU's it's cause that app's devs gone out of their way to support it and I'd be surprised if bridge is no longer needed.

[Kilrah]

11 minutes ago, venomtail said:

Doesn't multiple GPU use still require a bridge?

No.

That was for both GPUs to be able to share memory and act "as one GPU". Current productivity / rendering apps just treat them as the 2 separate units they are and give each one a different job to do.

[igormp]

21 minutes ago, venomtail said:

If something does support multiple GPU's it's cause that app's devs gone out of their way to support it and I'd be surprised if bridge is no longer needed.

The nvlink bridge isn't required for anything afaik, but it helps in some cases to speed up transfers between the GPUs instead of using PCIe. RAM to GPU is limited by the PCIe bandwidth and the bridge won't help at all in this case.

9 minutes ago, Kilrah said:

That was for both GPUs to be able to share memory and act "as one GPU".

I don't think anything ever made multiple GPUs look as a single one, but yes, the bridge was just meant for faster transfers and to keep the GPUs in sync easily for workloads that required it (such as games).

[YoungBlade]

To configure Blender for multiple GPUs, you go to Preferences > System > Compute Device Panel

 

I'm not familiar with Keyshot. Looking at this page implies that, for a single project, you would need the now defunct-for-consumers NVLink (SLI) bridge. https://manual.keyshot.com/manual/render-4/gpu-mode/

 

This old article mentions that there are way to configure the application to use multiple GPUs, and mentions that NVLink support was upcoming. So it's possible that there are ways to configure it without NVLink. I'd poke around in the settings.

[Bagzie]

47 minutes ago, Kilrah said:

No.

That was for both GPUs to be able to share memory and act "as one GPU". Current productivity / rendering apps just treat them as the 2 separate units they are and give each one a different job to do.

Sli was never able to share memory as far as I'm aware.

[eragon16gaming]

Thanks for all the information so quickly. Yeah for gaming a single 4070ti can pretty much handle extreme all the things for every game. i have a second one specifically for work purposes. the problem i run into is that thew HUD on keyshot shows as only using up to 12 GB vram, which is one GPU. this is leading me into bottleneck problems where the program runs out of VRAM and crashes on me, and there's still no autosave feature for keyshot as of yet. i also don't know if both GPU's are actually being utilized by the programs during rendering.

[Kilrah]

8 hours ago, Bagzie said:

Sli was never able to share memory as far as I'm aware.

Yeah, confused with NVLINK.

 

8 hours ago, eragon16gaming said:

the problem i run into is that thew HUD on keyshot shows as only using up to 12 GB vram, which is one GPU.

You can't pool memory from both GPUs. So yes you have 12GB. If an app like Blender is able to use 2 GPUs for rendering it's going to need to put a copy of the whole scene in each GPU so the 12GB are still going to be the limit.