NVSwitch Power Through DCGM For HGX Platforms

Open
  • Is something wrong?
#324 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Assessment

Difficulty
5/5
Estimated time
Over a week
Newbie friendliness
35/100
Issue type
Feature
Clarity
Mostly clear
Activity status
Active
Tech stack
cpp
Domain
observability

Research direction

Start by reading the DCGM telemetry collection entry points and how existing power metrics are exposed; the issue does not name specific files or tests. Determine whether HGX NVSwitch power can be sampled across the listed platforms and exposed through DCGM, then validate the sampling and platform coverage with maintainers, since no completion criteria or test path is specified.

Written by the indexing model from the issue text.

Description

human

Currently NVIDIA only provides NVSwitch power on Oberon platforms (GB, VR NVL72) through NVOS & OOO RedFish on the NVSwitch Tray, but does not provide NVSwitch power on the equivalent HGX platforms like Vulcan hgx h100/h200, Umbriel hgx b200/b300, etc.

Can this be implemented into dcgm such that we can get the sampling of the NVSwitch on HGX platforms like Vulcan hgx h100/h200, Umbriel hgx b200/b300, etc.? Can the ETA be provided for this feature?

+viz @kedarpotdar-nv

AI research
For GB300 NVLink switch systems, use nv show platform asic to inspect ASIC power, or nv show platform asic ASIC1 power for an individual ASIC; NVIDIA documents short-term averages over 50 ms and long-term averages over one second, in watts ([NVOS manual, pp. 342–345](https://docs.nvidia.com/networking/display/nvidia-nvos-user-manual-for-nvlink-switches-v25-02-4282.pdf)). The corresponding read-only REST endpoints are GET /nvue_v1/platform/asic and GET /nvue_v1/platform/asic/ASIC1/power, which provide an alternative for automated collection ([NVOS API documentation](https://docs.nvidia.com/networking/display/nvidia-nvos-user-manual-for-nvlink-switches-v25-02-4282.pdf)).For more detail, nv show platform asic ASIC1 power counters exposes power-average and power-shaper histograms; keep these consumption readings separate from power-profile allocations and caps ([NVOS manual](https://docs.nvidia.com/networking/display/nvidia-nvos-user-manual-for-nvlink-switches-v25-02-4282.pdf)). The cited manual explicitly limits these ASIC power commands to GB300 systems and describes ASIC-level usage, so they should not be presented as universal GB200 support or whole-switch-tray input power ([platform restrictions and scope](https://docs.nvidia.com/networking/display/nvidia-nvos-user-manual-for-nvlink-switches-v25-02-4282.pdf)).
Dominant language
C++
Stars
813
Forks
115
Avg merge
6d 10h
Merged PRs (30d)
1

Getting set up

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

More from NVIDIA/DCGM

All issues in NVIDIA/DCGM

Similar issues

More C++ issues

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.