Salad cannot access your GPU
Fix Degraded State errors 11, 12, 19, 21 and 22 ("Salad cannot access your GPU", "GPU drivers out of date"): the GPU isn't usable from inside the container environment.
What it means: Before each container job, Salad checks that the GPU can be used from inside the container environment. Errors 11/21 mean the card wasn't found there at all; 12/22 mean it was found but its status couldn't be read; 19 (and -7) mean the installed driver is too old or damaged for container jobs.
Why it happens: Windows can use the card perfectly well while the container environment can't. The link between the two is the NVIDIA driver, and it is the cause in nearly every case: a corrupt or partial driver install, a driver update that didn't take, or WSL losing the card until it's restarted. Less often: the card is an integrated GPU or an unsupported model, or a second GPU is confusing the check.
What to do
Restart the container environment
Exit Salad from the tray icon, open PowerShell and run wsl --shutdown, then start Salad. If the card had simply
"dropped out" of WSL, this is enough. If the error comes back within a day, continue.
Confirm the card is supported and dedicated
Integrated graphics (Intel UHD, AMD Radeon Graphics built into a CPU) can't run container jobs. Check that your card is listed on the Network Monitor and see Is my machine compatible.
Do a clean reinstall of the GPU driver
Even when the driver is "up to date", a clean reinstall is the fix most of the time: How to reinstall your GPU drivers. For NVIDIA, choose Custom and tick Perform a clean installation. Reboot afterwards.
Check what the container environment sees
In PowerShell (no admin):
wsl -d salad-enterprise-linux --exec /usr/lib/wsl/lib/nvidia-smiA table with your card means the fix worked; click Retry on the notification. "No devices were found" or an
initialization error means the driver still isn't reaching WSL - run wsl --update as Administrator, reboot, and check
again.
Two dedicated GPUs?
Only one GPU can earn. Disable the other in Salad under Settings > Configure Hardware, or physically, so the check tests the right card. See Can I use multiple GPUs.
Still stuck?
Send Salad Support the output of the nvidia-smi command above, your GPU model and driver version
(how to find them), and your logs folder zipped
(how to find it).
Last updated on
Failed to start workload
Fix the "Failed to start workload" Degraded State notification (errors -1, -4, -5, -7, -10), where the container environment runs but a job can't be downloaded or started.
Insufficient VRAM and high VRAM activity
Fix Degraded State errors 13, 24, 25 ("Insufficient VRAM") and 18 ("High VRAM activity"): too much of your GPU's memory is in use when the compatibility check runs.