raf802
-
Posts
8 -
Joined
-
Last visited
Content Type
Profiles
Forums
Downloads
Store
Gallery
Bug Reports
Documentation
Landing
Posts posted by raf802
-
-
23 hours ago, ich777 said:
I think because you are tried to save power when the VM wasn't running maybe?
Exactly, because there is a newer driver for the legacy cards available and it can't find your specified driver version, it will fall back to the latest available.
Glad to hear that everything is up and running again.
I think I spoke too soon....
I noticed today that the issue has come back. Again, plex was working fine and used the GPU for transcoding. I only noticed the issue because another docker container that was to use the GPU failed to compile / run.
In the attached setup, I have "nvidia-persistenced" in the go file and no custom scripts running.
Here is the diagnostics file.
-
47 minutes ago, ich777 said:
Keep in mind that if you have nvidia-persistenced on when you try to start the VM it is possible that the VM won't start and/or even crash the your entire server.
I would recommend that you bind the one card to VFIO that you maybe plan to use in a VM because then nvidia-persistenced will only work for the card that is not bound to VFIO.
This is then basically the same as what the script from spaceinvader one does since it will also pull the cards into P8 and they can of course ramp up to whatever power mode they need, but keep that with the VM in mind.
As a workaround if you don't want to bind the GPU to VFIO you can also do something like this in your go file:
nvidia-persistenced sleep 5 kill $(pidof nvidia-persistenced)
This will basically start nvidia-persistenced, wait 5 seconds and kill it after the 5 seconds so that the cards go to P8 when you boot the server.
Of course there are more advanced ways to also pull the card into P8 again after you've ended a VM.
Thanks.
I have bound the second card to vfio. Can't remember why it wasn't to begin with tbh.
Is there a way of putting the VM GPU in a low power P8 state when not in use (VM on) and when the VM is off?
1 hour ago, raf802 said:I will upgrade the OS and drivers to see if the GPUs keep working.
I have also updated the OS to 6.10.3, which upgraded the GPU drivers to v515 automatically. They are working fine still and my issue is resolved.
Thank you ich777!
- 1
-
19 minutes ago, ich777 said:
You don‘t use the GPU in VMs or am I wrong?
I have a second GPU set to be used in the VM. Nothing is bound to vfio on boot and these VMs are off.
In short, no it is not used by VMs.
22 minutes ago, ich777 said:If you are only using the GPU for Docker containers simply put this line in your go file and everything should work as it would with the script:
nvidia-persistenced
and of course reboot after that.
Ok, I'll do that, thank you.
- 1
-
21 minutes ago, ich777 said:
Please disable this script for now since this also includes a outdated command, nvidia-persistenced is now a dedicated application and nvidia-smi —persistence-mode should not be used anymore because this will deprecated in the future.
Please try to remove that script, reboot your server and try again.
The output seems fone to me…
Please keep me updated, could be possible that I wont answer very quickly in the next two days…
Thank you again Ich.
I disabled that powersave script and everything is working after the reboot.
The GPUs are idling in power state P0.
When I was getting the issue, their power state was P8.
I will upgrade the OS and drivers to see if the GPUs keep working.
- 1
-
Thank you for your quick reply!
12 minutes ago, ich777 said:What are the contents from: nvidia-powersave/script?
This is the script from Spaceinvader One. I added after the issue originally occurred. Here are the contents:
#!/bin/bash # check for driver command -v nvidia-smi &> /dev/null || { echo >&2 "nvidia driver is not installed you will need to install this from community applications ... exiting."; exit 1; } echo "Nvidia drivers are installed" echo echo "I can see these Nvidia gpus in your server" echo nvidia-smi --list-gpus echo echo "-------------------------------------------------------------" # set persistence mode for gpus ( When persistence mode is enabled the NVIDIA driver remains loaded even when no active processes, # stops modules being unloaded therefore stops settings changing when modules are reloaded nvidia-smi --persistence-mode=1 #query power state gpu_pstate=$(nvidia-smi --query-gpu="pstate" --format=csv,noheader); #query running processes by pid using gpu gpupid=$(nvidia-smi --query-compute-apps="pid" --format=csv,noheader); #check if pstate is zero and no processes are running by checking if any pid is in string if [ "$gpu_pstate" == "P0" ] && [ -z "$gpupid" ]; then echo "No pid in string so no processes are running" fuser -kv /dev/nvidia* echo "Power state is" echo "$gpu_pstate" # show what power state is else echo "Power state is" echo "$gpu_pstate" # show what power state is fi echo echo "-------------------------------------------------------------" echo echo "Power draw is now" # Check current power draw of GPU nvidia-smi --query-gpu=power.draw --format=csv exit
12 minutes ago, ich777 said:Is only one container affected or there are multiple container affected?
If it is only one container, please post the container configuration (Docker template).
It affects any container that I try to pass a GPU to. I have tried other plex/emby/jellyfin containers which I have never used and they too had the issue.
12 minutes ago, ich777 said:Please open up a terminal and post the output from:
cat /proc/sys/kernel/overflowuid
(screenshot preferred)
Please see attached
-
Hi,
I noticed that my containers which have GPUs passed through fail to restart after a CA auto-update this morning (15/06/22). One container which didn't stop during the night "appeared" to be working but was not transcoding a queued video. When I restarted this container manually I got the same error below.
The error I get when starting any GPU-passed container is a "Bad parameter" pop-up. When I edit the container config to see the compile error I get this:
docker: Error response from daemon: failed to create shim: OCI runtime create failed: container_linux.go:380: starting container process caused: process_linux.go:545: container init caused: Running hook #0:: error running hook: exit status 1, stdout: , stderr: nvidia-container-cli: initialization error: open failed: /proc/sys/kernel/overflowuid: permission denied: unknown.
I have the nvidia-driver plugin installed and have reinstalled most versions of the drivers, from v470.94 to v515.43.04. nvidia-smi shows my GPUs and the driver version correctly.
I am not sure what the cause of the error is, if its docker, the nvidia plugin, etc.
I noticed this issue occurring since upgrading the unraid OS from 6.9 to 6.10 (which included a nvidia driver update).
I have tried:
- a fresh docker.img, with previously configured container templates redownloaded.
- restoring the appdata folder from a backup.
- Checking GPU usage. Only the containers are set up to use the GPU (I even tried with a single container). I also made sure not to use the OS GUI mode. Currently nothing is using the GPU, according to nvidia-smi.
- Downgraded the OS from 6.10.2 to 6.10.1.
Here is the diagnostics file (had to use cmd line as GUI method doesn't do anything).
Any help will be greatly appreciated.
-
On 4/19/2022 at 4:25 PM, purplechris said:
Similar for me, just updated node stuck at Bad parameter when trying to start.
I had to remove --runtime=nvidia from my docker config for the node to start, but then it will not transcode.
Error shows as
docker: Error response from daemon: OCI runtime create failed: container_linux.go:367: starting container process caused: process_linux.go:495: container init caused: Running hook #1:: error running hook: exit status 1, stdout: , stderr: nvidia-container-cli: initialization error: open failed: /proc/sys/kernel/overflowuid: permission denied: unknown.
Resolved by going back to nvidia 510.54 driver, seems to like my p2000's
I also have this issue with tdarr_node and plex; or any container with a GPU passed through. I have no idea of the cause other than OS and driver update. I have rolled back both to no avail.
[Plugin] Nvidia-Driver
in Plugin Support
Posted · Edited by raf802
As a "start from scratch" approach. I removed the
command from the go file.
The docker containers all start up correctly with the GPU. Of course, they are in powestate P0, and I am trying to save power.
Do you mind elaborating on the following code?
Would this also put the container GPU into the P8 powerstate? Can it also go into P0 when needed and back to P8?
Some observations I have noticed with power save off (no commands run):
Is this discrepancy normal? Could nvidia-smi be reporting the incorrect power usage for P0, and hence P8 is <1W and causes an error?
For reference, when in P8, the GPU uses 7W according to GPU statistics.
When I run the nvidia-persistenced command from the cli, the nvidia-smi power usage reports correctly for P0 and then reports the correct P8 state too.
dailynas-diagnostics-20220616-1206.zip