Did you pass through any advanced config parameters from ESXi to get it working?
You made more progress than I on Ubuntu. There is a package on Debian called nvidia-driver and v460 was the version it “picked” for me. That was what I got working.
Nice catch on the driver,
I did, I used this YouTube video - PCI pass-through & Folding at Home on the VMware appliance (VMware vSphere ESXi 7) Jason Meers - YouTube
PCI pass-through & Folding at Home on the VMware appliance (VMware vSphere ESXi 7) Jason Meers
hypervisor.cpuid.v0 False
pciPassthru.use64bitMMIO True
pciPassthru.64bitMMIOSizeGB 32 - Single GPU option.
Memory - Check All guest memory (All locked)
Memory Limit - Unlimited
Boot Option EFI
Hi there, I met the same problem and my thread is Nvidia-smi shows "No devices were found", and dmesg shows "rm_init_adapter failed, device minor number 0"
By following @generix 's advice, I reinstall a Ubuntu 20.04 under ESXi, do passthrough as @twine.david said (except using EFI boot), downgrade the driver to 460.84, and hold my linux driver to 5.4.0-100-generic, now I can get my GPU in nvidia-smi as follows:
Because new users can’t upload more than 1 pictures, I split it up.
Interestingly, I don’t know why another hypervisor with nvidia-driver 470 can recognize GPU.
And I try another bare metal with Ubuntu 20.04, kernel 5.13.0-27-generic, nvidia-driver 510.47, GPU is GTX Titan X and it works. But the same conbination of kernel and driver is not work on ESXi.
As generix said, it seems like ESXi can’t support nvidia-driver >= 470 greatly.
Furthermore, to confirm kernel 5.13 can work with low version nvidia-driver, I made a bare metal with Ubuntu 20.04, kernel 5.13.0-28-generic, nvidia-driver 465.31, GPU is RTX 3090, it still works.
Because the apt source stopped to provide nvidia-driver under 470, I download .run file from nvidia website and install manually. To prevent the driver auto upgrade, I stopped autoupgrade function of apt.
Hope it can helps! <3
Did you install the driver via a .run file? I don’t see a v460 available via apt search.
EDIT: Missed your second post where you mentioned .run file.
I agree it seems to be ESXi/Nvidia related and not hypervisor related as I got it to pass through on Proxmox, even with the latest 510 drivers.
Yep. It looks like apt merged drivers from 418-470 to 470. I downloaded a 460 driver from https://www.nvidia.com/Download/Find.aspx?lang=en-us
Hi. Does anyone find a different solution other than using a lower version driver? The VM I’m using is managed by university so I couldn’t really turn off the auto upgrade.
My current situation is very similar to @DPCFish. One VM works perfectly fine with GTX 1080. But the other one with the same VM setting and Titan RTX failed to initialize the driver.
$ lspci | grep VGA can see the device.
$ dmesg | grep -i nvrm on 470.103 driver returns
NVRM: GPU 0000:03:00.0: RmInitAdapter failed! (0x26:0x56:1253)
on 510 driver returns error code 1463.
It seems work on workstation level card.
A4000 + Ubuntu 20 + ESXi 7.0U3
A5000 + Ubuntu 18 + ESXi 7.0U3
hello，I met the same question when i use 3080 on ubuntu 20.04.4 ESXi6. 7.U2 . nvidia-driver is 470.57.01. I want to use ubuntu20.04.3 and nvidia-driver6-460. Do you have installation package about ubuntu 20.04.3.iso and nvidia-driver-460 ?could you email to me (qq:1210586191)?
Hi,
I have just the same problem with my 1650. Trying right now with the 460 driver.
Update: 460 driver (first disable nouveau) works. I hope there wil be a fix fast for the new drivers.
Any Updates?
all Drivers above 470.57 wont work with ESXI → Ubuntu and my RTX 30er Series.
on Windows VMs no Problems.
need the new drivers (5XX)
i have the same issue, need working 510 drivers for ESXi
What changes were made in 470.86 because 470.82.01 works but 470.86 upwards doesn’t?
I’m just here to throw some wood on the fire.
Guess what? The beta release of 515.43.04 open source driver works with Ampere cards, but the proprietary driver shipped in the same installer does not. Hmmmmmmmmmmmmmmmmmmm
I am in the same position as others in the thread trying to use a 3090 in ESXi 7.0U3c with DirectPath I/O (PCIe passthrough). I confirm that the 470.82 drivers were the last proprietary drivers to work until 515.43.04/kernel-open I also want to reiterate that the properietary kernel driver 515.43/04/kernel still fails with the same errors as discussed in this thread.
Per the instructions here, install with: ./NVIDIA-Linux-x86_64-515.43.04.run -m=kernel-open and make sure to include options nvidia NVreg_OpenRmEnableUnsupportedGpus=1 in /etc/modprobe.d/nvidia
I lowkey suspect this whole “bug” is making it difficult to do this by design, and whatever chicanery NVidia is up to in the proprietary driver isn’t gonna fly on open source. Hooray for open source!
$ dmesg | grep vmware
[ 0.000000] vmware: TSC freq read from hypervisor : 2992.968 MHz
[ 0.000000] vmware: Host bus clock speed read from hypervisor : 66000000 Hz
[ 0.000000] vmware: using clock offset of 18001067827 ns
[ 5.414545] systemd[1]: Detected virtualization vmware.$ nvidia-smi
Thu May 19 16:38:19 2022
±----------------------------------------------------------------------------+
| NVIDIA-SMI 515.43.04 Driver Version: 515.43.04 CUDA Version: 11.7 |
|-------------------------------±---------------------±---------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
| | | MIG M. |
|===============================+======================+======================|
| 0 NVIDIA GeForce … Off | 00000000:03:00.0 Off | N/A |
| 30% 38C P8 35W / 350W | 5MiB / 24576MiB | 0% Default |
| | | N/A |
±------------------------------±---------------------±---------------------+±----------------------------------------------------------------------------+
| Processes: |
| GPU GI CI PID Type Process name GPU Memory |
| ID ID Usage |
|=============================================================================|
| 0 N/A N/A 2942 G /usr/lib/xorg/Xorg 4MiB |
±----------------------------------------------------------------------------+
1 Like
Had the same issue, running RTX 3060ti with ubuntu 20.04 on esxi. Hardware is Asus x370 crosshair 3700x. Was working fine, then updated and no longer had a gpu. As of 8/15/22 I tried updating to the latest patch of esxi, was on 7.0 update 1 and installed to update 3 U3f, no change. Tried 22.04 and Manjaro with preconfigured 515.65.01 drivers, no luck. I was able to downgrade to 460.84 in Manjaro and get that to work. I can also confirm opensource driver 515.65.01 does work on Manjaro with kernel 5.4.210-1 using the manual install path from guide https://linuxconfig.org/how-to-install-the-nvidia-drivers-on-manjaro-linux and using -m=kernel-open with the .run download. Then added options nvidia NVreg_OpenRmEnableUnsupportedGpus=1 in /etc/modprobe.d/nvidia.conf
and rebooted and I now have a working nvidia-smi output.
Hi! Is it possible to explain how you install the (latest) open source drivers on a ubuntu server? Thanks!
So this is the full solution! use the open source drivers of nvidia. Here is how to do it (on my ubuntu server):
First Disable nouveau & enable unsuported GPU’s for open source drivers:
1. Go to: /etc/modprobe.d/
2. Make a file: blacklist-nvidia-nouveau.conf
3. Put this in the file:
blacklist nouveau
options nouveau modeset=0
4. Make a other “nvidia.conf” file and put this in the file: options nvidia NVreg_OpenRmEnableUnsupportedGpus=1
5. Updat kernel init ram fs: sudo update-initramfs -u
6. Reboot
7. Go to the Nvidia site to the page to download the driver that you want.
8. Copy the URL of the download butten and past it behind wget to download it to the current folder
9. Sudo chmod 700 the file
10. Run the install file: sudo .\filename.run
a. For open source: sudo .\filename.run -m=kernel-open
b. Watch out! See that the nvidia.conf file exist in the modprobe folder and you have rebooted (and run the update-initramfs commando). Then only the GTX/RTX/QUADRO cards wil work!
11. After the instalation reboot the server
12. Test with nvidia-smi
13. Great succes!
4 Likes
thanks for Excellent Anser. blacklist-nvidia-nouveau.conf & sudo update-initramfs -u is needed for ubuntu 22.04, but needless for 20.04