[GH-ISSUE #530] What happen if there is an ACK loss when nvshmemi_ibgda_amo_nonfetch_add? #339

New issue

Open

opened 2026-10-26 09:58:48 +00:00 by bsakaguchi · 0 comments

bsakaguchi commented

2026-10-26 09:58:48 +00:00

Owner

Originally created by @gangxie112 on GitHub (Dec 24, 2025).
Original GitHub issue: https://github.com/deepseek-ai/DeepEP/issues/530

Hello,

Could the atomic-add be done twice if there is there is an ack loss? Recently, we hit some issues of deepep hung and data corruption with release 1.2.1. Little worried about this part.

Thanks

Originally created by @gangxie112 on GitHub (Dec 24, 2025). Original GitHub issue: https://github.com/deepseek-ai/DeepEP/issues/530 Hello, Could the atomic-add be done twice if there is there is an ack loss? Recently, we hit some issues of deepep hung and data corruption with release 1.2.1. Little worried about this part. Thanks

bsakaguchi referenced this issue

2026-10-27 21:03:39 +00:00

[PR #339] [MERGED] Support nvfp4 intranode dispatch #524

bsakaguchi referenced this issue

2026-10-27 21:04:11 +00:00

[PR #447] [CLOSED] Fix bug on hybrid-ep #575

bsakaguchi referenced this issue

2026-10-27 21:05:58 +00:00

[PR #748] [CLOSED] [HybridEP] Report RDMA memory registration failures #755