From 950d33aa3045581906c2db11f33d2a4c2bca3033 Mon Sep 17 00:00:00 2001 From: Daniel Hiltgen Date: Mon, 8 Sep 2025 11:39:00 -0700 Subject: [PATCH] docs: show how to debug nvidia init failures (#12216) This debug setting can help troubleshoot obscure initialization failures. --- docs/troubleshooting.md | 3 +++ 1 file changed, 3 insertions(+) diff --git a/docs/troubleshooting.md b/docs/troubleshooting.md index 6fdd3e85..7647b12f 100644 --- a/docs/troubleshooting.md +++ b/docs/troubleshooting.md @@ -92,6 +92,9 @@ If none of those resolve the problem, gather additional information and file an - Set `CUDA_ERROR_LEVEL=50` and try again to get more diagnostic logs - Check dmesg for any errors `sudo dmesg | grep -i nvrm` and `sudo dmesg | grep -i nvidia` +You may get more details for initialization failures by enabling debug prints in the uvm driver. You should only use this temporarily while troubleshooting +- `sudo rmmod nvidia_uvm` then `sudo modprobe nvidia_uvm uvm_debug_prints=1` + ## AMD GPU Discovery