On the top of every thread there is a "Watch" button. With this you can subscribe to a thread without posting "empty" posts.
[SOLVED] Kernel 5.11 & NVIDIA Linux vgpu-kvm
- Thread starter mishki
- Start date
-
- Tags
- kernel 5.11 nvidia vgpu
On the top of every thread there is a "Watch" button. With this you can subscribe to a thread without posting "empty" posts.
updated drivers: <removed by moderation>
unlock: VGPU-Proxmox
only works for kernel 5.15 and below (5.15.60-2-pve)
unlock: VGPU-Proxmox
only works for kernel 5.15 and below (5.15.60-2-pve)
Last edited by a moderator:
updated drivers: <removed by moderation>
unlock: VGPU-Proxmox
works for kernel 5.19 and below (5.19.17-1-pve)
unlock: VGPU-Proxmox
works for kernel 5.19 and below (5.19.17-1-pve)
Last edited by a moderator:
Yes but did anyone get this vgpu to work in COMBINATION with cpu HOST? I always get a BSOD bluescreen due to nvlddmkm.sys if I try to have BOTH cpu HOST (so that Ubuntu WSL2 can work in Windows 11 VM). I can DISABLE the RTX6000 in device manager and have no crash and Ubuntu WSL2 working, OR I can use QVM common cpu instead of cpu HOST on the VM and that way I can have RTX6000 vgpu working inside VM but no WSL2. Does anyone have BOTH working at the same time?
I'm using Linux 5.19.17-1-pve and NVIDIA-Linux-x86_64-525.60.12-vgpu-kvm....
I'm using Linux 5.19.17-1-pve and NVIDIA-Linux-x86_64-525.60.12-vgpu-kvm....
Last edited:
Thanks, I tried args: -cpu 'host,kvm=off' but sadly same crash. Sometimes right away, sometimes it boots and crash comes seconds later, same nvlddmkm.sys
Do you have it working with both? I mean does for example Ubuntu work via WSL2 while you're booted with vgpu and RTX6000 etc?
no, no pve with gpu project here yet, it was just an idea about nvidia blocking virtualization, but your problem, imo, is about
wsl2 that require virtualization so it's nested and nvidia gpu which lock/reserve something which prevent the nested virt needed by wsl2.
just a guess , sorry for language or noob idea.
wsl2 that require virtualization so it's nested and nvidia gpu which lock/reserve something which prevent the nested virt needed by wsl2.
just a guess , sorry for language or noob idea.
@Dizel958 Why can't you use the more recent NVIDIA-GRID-Linux-KVM-525.60.12-525.60.13-527.41 ?
I'm still waiting for anyone who can confirm they have vgpu working inside a VM but with CPU "HOST" option in proxmox vm setting.
I'm still waiting for anyone who can confirm they have vgpu working inside a VM but with CPU "HOST" option in proxmox vm setting.
I have a Tesla P4 video card installed in my server and vgpu profiles work on the NVIDIA-Linux-x86_64-510.85.03-vgpu-kvm.run driver
But I want to install exactly this version NVIDIA-Linux-x86_64-510.108.03-vgpu-kvm.run
agent: 0,type=isa
args: -cpu 'host,-hypervisor,kvm=off,vendor=GenuineIntel,+pcid,+vmx'
audio0: device=ich9-intel-hda,driver=none
balloon: 0
bios: ovmf
boot: order=sata0;net0
cores: 2
cpu: host,hidden=1
efidisk0: local-lvm:vm-100-disk-0,efitype=4m,pre-enrolled-keys=1,size=4M
hostpci0: 0000:0a:00.0,mdev=nvidia-63,pcie=1
localtime: 1
machine: pc-q35-7.1
memory: 6144
meta: creation-qemu=7.1.0,ctime=1670767022
net0: e1000=B1:F9:0B:06:A2:EA,bridge=vmbr0
numa: 0
ostype: win10
sata0: local-lvm:vm-100-disk-1,aio=io_uring,backup=0,cache=directsync,size=117125940K,ssd=1
scsihw: lsi53c810
smbios1: uuid=0f3ce76e-1949-4480-999a-82aceb287310
sockets: 1
tablet: 0
vga: none
vmgenid: 42056cd0-c470-4be6-8916-b04192a9d8bb
I have already tested the latest version of the NVIDIA-Linux-x86_64-525.60.12-vgpu-kvm.run driver, I didn’t like the driver specifically for Windows
Last edited:
@Dizel958 Do you mean you just want this file? Here it is direct from Nvidia:
https://storage.googleapis.com/nvid...U14.3/NVIDIA-Linux-x86_64-510.108.03-grid.run
Other versions as well:
https://cloud.google.com/compute/docs/gpus/grid-drivers-table
Linux display:
https://www.nvidia.com/download/driverResults.aspx/195121/en-us/
https://storage.googleapis.com/nvid...U14.3/NVIDIA-Linux-x86_64-510.108.03-grid.run
Other versions as well:
https://cloud.google.com/compute/docs/gpus/grid-drivers-table
Linux display:
https://www.nvidia.com/download/driverResults.aspx/195121/en-us/
No, these are not the right drivers. There is no NVIDIA-Linux-x86_64-510.108.03-vgpu-kvm.run driver on these links.
I need exactly this archive NVIDIA-GRID-Linux-KVM-510.108.03-513.91.zip
Last edited:
Ok my bad then, I notice now the version is correct but the ending is grid.run instead of the vgpu-kvm.run
@Dizel958 Maybe ask @dcorp on here since seems he was using the 510.108.03-vgpu-kvm driver version:
https://forum.proxmox.com/threads/nvidia-vgpu-on-proxmox-ve-7-x.113983/page-2
Seems you need 14.3 package? https://forum.proxmox.com/threads/nvidia-vgpu-on-proxmox-ve-7-x.113983/post-515439
Apparently we can download drivers directly from Nvidia if we apply for a "TRIAL":
https://www.nvidia.com/en-us/data-center/resources/vgpu-evaluation/
Last edited:
I applied for an NVIDIA Enterprise Account twice. And I don't get an answer. Therefore, I can not go in and download the necessary driver.
@Dizel958 Apply again, but this time, if you are not located in USA or Germany or UK or some other wealthy country, then use a VPN that gives you an IP address inside one of these countries and apply with an email address that is not 'free' and not an email "provider", but a company email. Do not use the same email address you used before if you applied with it from an IP address located in a not so wealthy country. If you have a decent sized 'employer' it might help to use your 'work' email address. (I haven't tried yet myself, I'm using drivers found online already.)
The third time I sent a request and also no answer. I changed the vpn to the United States, set the language to English and entered all the real data.