nvidia: don't report fake PCIe link on unified-memory SoC GPUs - #499
Open
Code4me2 wants to merge 1 commit into
Open
nvidia: don't report fake PCIe link on unified-memory SoC GPUs#499Code4me2 wants to merge 1 commit into
Code4me2 wants to merge 1 commit into
Conversation
On unified-memory SoC platforms such as the DGX Spark (GB10), the GPU is attached via an on-die interconnect (NVLink-C2C) rather than a PCIe link. NVML nevertheless returns NVML_SUCCESS for the current PCIe link generation/width, but with placeholder values (GEN 1 @ 1x), which nvtop surfaces as if they were real. Since unified-memory devices have no PCIe framebuffer link, only report PCIe link generation/width on discrete (non-unified) GPUs so the header falls back to N/A instead of showing a misleading link. Pcie RX/TX throughput already shows N/A on these devices because NVML returns NVML_ERROR_NOT_SUPPORTED there. Closes the PCIe misreporting part of Syllo#426.
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
Fixes the PCIe misreporting portion of #426 on unified-memory SoC GPUs such as the DGX Spark (GB10).
On the GB10, the GPU is attached via NVLink-C2C (on-die interconnect), not a PCIe slot. However, NVML's
nvmlDeviceGetCurrPcieLinkGeneration/nvmlDeviceGetCurrPcieLinkWidthreturnNVML_SUCCESSwith placeholder values (GEN 1 @ 1x), which nvtop displayed as if they were a real PCIe link.These devices are already detected (unified memory →
has_unified_memory), so this PR gates PCIe link gen/width reporting on not being unified memory, making the header fall back toN/A. PCIe RX/TX throughput was alreadyN/Ahere because NVML returnsNVML_ERROR_NOT_SUPPORTEDon this platform.Before (master)
After
Memory (
MEM[...Gi]) is already reported correctly on GB10 via the #466 unified-memory fallback; fan and memory-clock remain correctlyN/Asince the SoC does not expose them through NVML.Validation
has_unified_memoryis only set when NVML reports unified/SoC memory.