mirror of
https://github.com/deepseek-ai/DeepEP.git
synced 2026-11-05 00:47:32 +00:00
[GH-ISSUE #355] test_internode.py runs failed due to "nvshmem detect topo failed" #246
Labels
No labels
pull-request
No milestone
No project
No assignees
1 participant
Notifications
Due date
No due date set.
Dependencies
No dependencies set.
Reference
deepseek-ai/DeepEP#246
Loading…
Add table
Add a link
Reference in a new issue
No description provided.
Delete branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
Originally created by @new-TonyWang on GitHub (Aug 8, 2025).
Original GitHub issue: https://github.com/deepseek-ai/DeepEP/issues/355
I'm using 2 nodes with 16 H200 GPU to run this test in docker container. And I can run "test_intranode.py" successfuly ,but failed run "test_internode.py".
these are my card status
here are the script to run test_internode.py
test_result_internodes_0.log
test_result_internodes_1.log
running shmem_put_bw on two node with the following scripts:
get these log:
put_bw_twonodes.log
nvdebug.txt
@sphish commented on GitHub (Aug 8, 2025):
Your NIC IP address may not be configured correctly.
@new-TonyWang commented on GitHub (Aug 8, 2025):
Thanks! Could you please tell me how to check if the IP address is correct? Both node can access to each other using http and 16 gpu allreduce can also be done successfully.
@sphish commented on GitHub (Aug 11, 2025):
I’m not an expert in this area, but I think you could try using
ib_write_bwto test connectivity.@new-TonyWang commented on GitHub (Aug 12, 2025):
I tried 'ib_write_bw' on both nodes and I got these:
@new-TonyWang commented on GitHub (Aug 13, 2025):
I solved this problem by seting env:
but I don't know why this could work.
@Skylion007 commented on GitHub (Aug 15, 2025):
Memory fragmentation of NVSHMEM, did you try increasing the NVSHMEM_SYMMETRIC_SIZE?
@new-TonyWang commented on GitHub (Sep 4, 2025):
No, I haven't . But I can give a try. Thank you !