Feature/dcgm gpu backend #1391
Open
Feature/dcgm gpu backend #1391
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
- Introduce a new DCGM-backed NVIDIA GPU collector on Linux that populates the existing Gpu::gpu_info structures using dcgmUpdateAllFields and dcgmGetLatestValuesForFields. - Prefer DCGM over NVML when built with -DBTOP_DCGM=ON and libdcgm is available, while keeping NVML as a transparent fallback on systems without DCGM. - Track a unified nvidia_device_count so AMD (ROCm SMI) and Intel GPU backends stack correctly after whichever NVIDIA backend is active. - Expose a new CMake option BTOP_DCGM and link libdcgm when enabled, keeping GPU runtime behaviour controlled via existing shown_gpus and show_gpu_info config options. - Update README GPU compatibility and CMake documentation to describe the DCGM backend, including usage on DGX Spark / data center GB-series systems and how to enable it. Tests: - Built with -DBTOP_GPU=ON -DBTOP_DCGM=ON on Linux and verified that btop runs with DCGM present (DGX-style node) and falls back to NVML or no NVIDIA GPUs when DCGM is unavailable.
GPU name retrieval (lines 1290-1305): Added dcgmGetDeviceAttributes() to get proper GPU names, with the same cleanup logic as NVML supported_functions initialization (lines 1482-1496): During collect<1>, supported_functions is now set based on which fields returned valid data, with unsupported features (pwr_state, pcie_txrx, encoder/decoder) explicitly set to false Empty deque fallback (lines 1499-1509): All deques now guaranteed to have at least one value (0) to prevent .back() crashes
GPU name retrieval (lines 1290-1305): Added dcgmGetDeviceAttributes() to get proper GPU names, with the same cleanup logic as NVML supported_functions initialization (lines 1482-1496): During collect<1>, supported_functions is now set based on which fields returned valid data, with unsupported features (pwr_state, pcie_txrx, encoder/decoder) explicitly set to false Empty deque fallback (lines 1499-1509): All deques now guaranteed to have at least one value (0) to prevent .back() crashes
deckstose
reviewed
Dec 4, 2025
Collaborator
deckstose
left a comment
deckstose
left a comment
There was a problem hiding this comment.
This lacks Makefile support.
In my opinion the changes can be split out of btop_collect into their own file.
What's up with the changes to the utility functions? Please undo them.
Owner
|
@deckstose Thinking about adding rules that PR's that are obviously vide coded should be dismissed unless the author has some proof that they actually understand the code in the PR (like for example that they have other repositories in C++ that aren't vibe coded). |
Collaborator
|
I totally agree |
Owner
|
@deckstose
|
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Added Nvidia unified memory architecture based GB (Grace Blackwell) GPU support by using DCGM support. Tested on Nvidia DGX Spark (GB10)