-
Notifications
You must be signed in to change notification settings - Fork 1.5k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Add documentation for PA metrics feature #4871
Conversation
docs/user_guide/perf_analyzer.md
Outdated
|
||
Perf Analyzer can optionally collect and output server-side metrics, such as GPU | ||
utilization and GPU power usage. To enable this collection/output, use the | ||
`--verbose-csv` (with the `-f <filename>` option) and `--collect-metrics` |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I would split up talking about collection (--collect-metrics) vs output to csv since soon it will also output to stdout. Also the csv stuff should go after talking about the metrics-url option
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Rearranged 👍
b04e8fa
to
c19c126
Compare
Note that all metrics are per-GPU in the case of multi-GPU systems. | ||
|
||
To output these server-side metrics to a CSV file, use the `-f <filename>` and | ||
`--verbose-csv` CLI options. The output CSV will contain one column per metric. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Just curious, why is the extra --verbose-csv
needed? Shouldn't specifying a filename -f
be sufficient to say we want the output to the file?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I don't remember the full rationale behind --verbose-csv
. @tgerdesnv @nv-braf can discuss. What I remember is that --verbose-csv
was added as a way to preserver "the old" csv output format (i.e. not using --verbose-csv
) for backwards compatibility. Somewhere in the design discussions for this metrics feature, we decided that the metrics would only be output on this "new" csv output format (i.e. using --verbose-csv
).
Co-authored-by: Ryan McCormick <rmccormick@nvidia.com>
1f4dab0
* Add documentation for PA metrics feature * Add missing documentation * Addressed comments * Update docs/user_guide/perf_analyzer.md Co-authored-by: Ryan McCormick <rmccormick@nvidia.com> Co-authored-by: Ryan McCormick <rmccormick@nvidia.com>
* Add documentation for PA metrics feature * Add missing documentation * Addressed comments * Update docs/user_guide/perf_analyzer.md Co-authored-by: Ryan McCormick <rmccormick@nvidia.com> Co-authored-by: Ryan McCormick <rmccormick@nvidia.com>
* Add documentation for PA metrics feature * Add missing documentation * Addressed comments * Update docs/user_guide/perf_analyzer.md Co-authored-by: Ryan McCormick <rmccormick@nvidia.com> Co-authored-by: Ryan McCormick <rmccormick@nvidia.com>
No description provided.