How does Deepstack handle a PC with multiple discrete GPUs?

[mrc545]

I am about to add a second NVIDIA GPU to my Blue Iris Windows 10 PC. The GPU that is currently in there is handling Deepstack. How does Deepstack treat a PC that has two discrete GPU's? Does it share the load? Is there a way to specify which GPU (or both) for Deepstack to utilize?

 

[Coldair]

what is a discrete gpu?
I have built PCs with 4 watercooled GPU but they were right there on the motherboard not discreet you can plainly see them

 

[bp2008]

 

[Coldair]

I knew I should have paid attention in high school instead of smoking wacky weed with my teachers

anyway with my pcs that multiple GPUs the NVidia software set the sharing of the load

 

[mrc545]

No worries. I had problems with the differentiation until my 30's. Hah. Thanks for the feedback. Adding an RTX 3080 to supplement the RTX 2060 that is in there. Will be interesting to see how they work together.

 

[phtp]

Does anyone know if there is a way for it to force which GPU it uses with deepstack?

 

[kferrero]

I doubt deepstack has this functionality built in, but in the nvidia control panel you can assign specific GPUs to specific processes in Windows. I have had to do that before with some games that were defaulting to the built-in Intel GPU and not using my GTX 1070. I had to go into the nvidia control panel and I forget where it is but you locate the .exe of whatever program/game you want to manually assign and then tell the nvidia control panel you want that exe to launch with the discrete card not the integrated Intel card. I would imagine with dual nvidia cards you can do the same, deepstack's exe is py.exe

 

[phtp]

I doubt deepstack has this functionality built in, but in the nvidia control panel you can assign specific GPUs to specific processes in Windows. I have had to do that before with some games that were defaulting to the built-in Intel GPU and not using my GTX 1070. I had to go into the nvidia control panel and I forget where it is but you locate the .exe of whatever program/game you want to manually assign and then tell the nvidia control panel you want that exe to launch with the discrete card not the integrated Intel card. I would imagine with dual nvidia cards you can do the same, deepstack's exe is py.exe

Oh that's a great suggestion. I will have to try that. Thanks!

Edit: FYI for those interested - doesn't work with limiting Cuda - GPUs and OpenGL rendering GPU in Nvidia Control Panel. Globally or any deepstack + python executables.

 

[mrc545]

I tried running Deepstack in Docker Desktop (in tandem with Deepstack GPU for Windows) with the "--gpus X" option to force it to use the other GPU, and it just launches on the GPU that the Windows version uses.