With the A16 being 4 different GPUs in 1 card, can each of them split into different vGPU slices in proxmox ve? say 8 - 2gb chunks, 4 - 4gb chunks, 2 - 8gb, and 1 - 16gb chunk? or do all have to split equally?
Hi there @darrellsmoneymaker, welcome to the NVIDIA developer forums.
Memory allocation is per âGPUâ in this case, but of course depending on application design you have access to fast inter-chip memory communication.
But for more details on how to make the best se of this in vGPU scenarios, I moved your post into the vGPU category, I think that is a better place to discuss your plans.
Thanks
1 Like
Sure. You can run 4 different vGPU profiles. Please keep also in mind that we start to support heterogenous profiles starting with vGPU 17 which means you can run different FB profiles on the same GPU. (needs to be enabled, check our documentation).
1 Like
if the four GPUs show up individually in the OS, would i have to run a heterogenous profile? wouldnât i run 1 profile for each?
Depends on your requirements. With A16 this is not really necessary to have the heterogenous profiles feature but with GPUs like A40 or L40 it makes more sense as you could only run 1 profile pre-vGPU 17
you said âwe start to support heterogenous profilesâ. â multiple profiles for 1 gpu.
i said âif the four GPUs show up individually in the OS, would i have to run a heterogenous profile? wouldnât i run 1 profile for each?â <â the OS shows the A16 as 4 different GPUs, i dont need to split each of them with multiple profiles each. i want to split the 4 GPUs each with 1 profile each.
Yes, and I clearly said it is an additional new option. For sure it is not necessary in your case if you just want to run 1 profile per GPU