Quadro and Gtx same system?

Been wondering this for years, but has anyone had any luck using gpu assignment where a gtx card (same family) runs as gpu 0 and a quadro card runs on gpu 1?
I want to use the main card for multi camera trackong gpu tasks and the second gpu for content and output

I’ve never had a successful mixed assignment with affinity :confused:

In theory things should be better with Vulkan, however Quadro’s use different drivers than Geforces, so your system isn’t in a great spot with that setup. I certainly not recommend it.

Thanks team for the great answers. Ahh good point about the drivers. Shame we cant gpu assign gtx cards.

In the 2022.20000 series GPU affinity now works on Geforce cards.

1 Like

Wait WHAT??? For real??

Yep, it works on all cards as a base feature of Vulkan.

wow! Ok I’d say that should probably be added in a bold font to the release notes :grin:…actually I don’t think this info is anywhere public yet?
But this opens quite a few perspectives, amazing!

pretty sweet to push the heavy load of calculating fluids, particles or flocking etc on a screaming fast Geforce, and then render the final output on a another card in another instance.
But @malcolm now you know geforce works with affinity, would you still not advise running one TD instance on a Geforce and other TD instance on a Quadro?

Pffff then imagine a GPUDirect P2P TOP, we could link 2 cards that support it both in their own TD instance directly to each other’s textures… :grimacing:

It’s mentioned here where we talk about GPU Affinity:

It doesn’t change the rules for Geforce/Quadro mixing, due to the driver issues mentioned above. Fast cross GPU communication is still a huge task that doesn’t get any easier with this (and is generally Quadro only anyways).

Ah I totally missed that in the docs! thx

I was hoping gpudirect P2P would be easier now, as I thought it was now fully supported in recent CUDA versions: cuda-samples/Samples/0_Introduction/simpleP2P at 26665bf33b2df3f03f52467d954090f8a19d0e71 · NVIDIA/cuda-samples · GitHub
but I’m no expert in this field so probably there’s much more to it…

Things aren’t getting easier :slight_smile: Vulkan is massively more complex than OpenGL.

1 Like

Omg affinity on all gpus is massive. I totally overlooked that.