Refine burn UI and NVIDIA stress flows

This commit is contained in:
2026-04-05 13:43:43 +03:00
parent 25af2df23a
commit 38e79143eb
18 changed files with 825 additions and 229 deletions

View File

@@ -52,6 +52,31 @@ else
fail "nvidia-smi: NOT FOUND"
fi
if p=$(PATH="/usr/local/bin:$PATH" command -v dcgmi 2>/dev/null); then
ok "dcgmi found: $p"
else
fail "dcgmi: NOT FOUND"
fi
if p=$(PATH="/usr/local/bin:$PATH" command -v nv-hostengine 2>/dev/null); then
ok "nv-hostengine found: $p"
else
fail "nv-hostengine: NOT FOUND"
fi
DCGM_PROFTESTER=""
for tool in dcgmproftester dcgmproftester13 dcgmproftester12 dcgmproftester11; do
if p=$(PATH="/usr/local/bin:$PATH" command -v "$tool" 2>/dev/null); then
DCGM_PROFTESTER="$p"
break
fi
done
if [ -n "$DCGM_PROFTESTER" ]; then
ok "dcgmproftester found: $DCGM_PROFTESTER"
else
fail "dcgmproftester: NOT FOUND"
fi
for tool in bee-gpu-burn bee-john-gpu-stress bee-nccl-gpu-stress all_reduce_perf; do
if p=$(PATH="/usr/local/bin:$PATH" command -v "$tool" 2>/dev/null); then
ok "$tool found: $p"
@@ -60,6 +85,12 @@ for tool in bee-gpu-burn bee-john-gpu-stress bee-nccl-gpu-stress all_reduce_perf
fi
done
if p=$(PATH="/usr/local/bin:$PATH" command -v nvbandwidth 2>/dev/null); then
ok "nvbandwidth found: $p"
else
warn "nvbandwidth: NOT FOUND"
fi
echo ""
echo "-- NVIDIA modules --"
KO_DIR="/usr/local/lib/nvidia"