DEVICE/CUDA_IPC: Fix proto selection for cuda_ipc #10943
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
michal-shalev
requested review from
Sergei-Lebedev,
brminich,
iyastreb,
ofirfarjun7 and
tvegas1
iyastreb
reviewed
Oct 10, 2025
| iface_attr->bandwidth.dedicated = 0; | ||
| iface_attr->bandwidth.shared = iface->config.bandwidth; | ||
| iface_attr->overhead = 7.0e-6; | ||
| iface_attr->overhead = 0.05e-6; |
There was a problem hiding this comment.
- How do we know the exact overhead of cuda_ipc on device?
- So previous overhead of 7us is still accurate for host operations?
Then this change might break protocol selection for host operations.
Maybe we need a separate iface cuda_ipc_device_iface that would override this latency estimate?
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 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Outdated suggestions cannot be applied.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.
THIS PULL REQUEST IS A TEMPORARY WORKAROUND FOR INTEGRATION, WE ARE NOT PLANNING TO MERGE IT
What?
Fix
cuda_ipcoverhead for device operations to enable proto selection on EOS.Why?
When running on EOS, cuda_ipc was failing to be selected during proto selection for device lanes. Despite having better bandwidth than
rc_gda,cuda_ipchad a hardcoded overhead of 7.0µs appropriate for host operations but incorrect for device operations, causing it to score poorly.Before
After
How?
Changed hardcoded overhead from 7.0µs to 0.05µs in
uct_cuda_ipc_iface_query(). This increasescuda_ipc's score, making it automatically selected for intra-node device lanes.