Skip to content

Commit e441545

Browse files
Becker-ZHGoogle-ML-Automation
authored andcommitted
Expose system memory buffer size for PM Sampling
PiperOrigin-RevId: 813830951
1 parent bf86259 commit e441545

File tree

1 file changed

+1
-0
lines changed

1 file changed

+1
-0
lines changed

docs/profiling.md

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -355,6 +355,7 @@ The following options are available for GPU profiling:
355355
[NVIDIA's CUPTI documentation](https://docs.nvidia.com/cupti/main/main.html#metrics-table).
356356
* `gpu_pm_sample_interval_us`: Sets the sampling interval in microseconds
357357
for CUPTI PM sampling. Defaults to `500`.
358+
* `gpu_pm_sample_buffer_size_per_gpu_mb`: Sets the system memory buffer size per device in MB for CUPTI PM sampling. Defaults to 64MB. The maximum supported value is 4GB.
358359
* `gpu_dump_graph_node_mapping`: If enabled, dumps CUDA graph node
359360
mapping information into the trace. Defaults to `False`.
360361

0 commit comments

Comments
 (0)