NVTX Stack Trace
$GPU_LABEL | Window: $TRIM_LABEL
Duration heat:
Fast
Medium
Slow
Bar width = % of parent
ðŊ Scope:
All regions
Clear
ðē Tree
⥠Kernels
ð Scan
ð ScanV2
ðģ ScanV3
ð Streams
ð Aggregate
ðŪ Patterns
âķ
Exit Focus (F)
Expand All
Collapse All
Depth 2
Depth 4
ðĄïļ Heat
ðŊ Scope
ð Demangled
ðĪ AI
path
glob path: sample_0/**/flash*
ðĪ AI Profiler
â
I can analyze your GPU profile. Try asking:
âĒ "What's the bottleneck?"
âĒ "Why is kernel X slow?"
âĒ "Compare streams 21 and 56"
Configure LLM endpoint in settings to enable.
Send