You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I am not affiliated with NVIDIA but I can comment a bit here.
I think it would be good if we had an automated method for installing GDS on bare metal or VM. Seems Ansible would be awesome for this.
Now if we did have GDS software installed in a VM we have to consider what hardware is real and what hardware is emulated. QEMU allows the user to emulate NVMe but there is no emulation of an NVIDIA GPU. So the GPU would have to be passed in.
Now you have to consider peer to peer DMAs and what that even means when the NVMe SSDs are emulated. I know quite a bit about this from my p2pdma work and I know you can make this all work but it only emulates p2p.
All in all I like this VM idea as a learning tool. But it won't work for an actual GDS deployment.
I am not affiliated with NVIDIA but I can comment a bit here.
I think it would be good if we had an automated method for installing GDS on bare metal or VM. Seems Ansible would be awesome for this.
Now if we did have GDS software installed in a VM we have to consider what hardware is real and what hardware is emulated. QEMU allows the user to emulate NVMe but there is no emulation of an NVIDIA GPU. So the GPU would have to be passed in.
Now you have to consider peer to peer DMAs and what that even means when the NVMe SSDs are emulated. I know quite a bit about this from my p2pdma work and I know you can make this all work but it only emulates p2p.
All in all I like this VM idea as a learning tool. But it won't work for an actual GDS deployment.
could you please shed light if we can set up one VM to quickly understand GDS and play it inside VM?
The text was updated successfully, but these errors were encountered: