mirror of
https://github.com/deepseek-ai/DeepSeek-V3.git
synced 2026-11-05 00:47:48 +00:00
[GH-ISSUE #28] [Question] On NVLink bandwidth of H800 #22
Labels
No labels
closed-as-stale
enhancement
question
spam
stale
No milestone
No project
No assignees
1 participant
Notifications
Due date
No due date set.
Dependencies
No dependencies set.
Reference
deepseek-ai/DeepSeek-V3#22
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 @GHGmc2 on GitHub (Dec 30, 2024).
Original GitHub issue: https://github.com/deepseek-ai/DeepSeek-V3/issues/28
May I know why the bandwidth of NVLink on H800 is 160GB/s, instead of 400GB/s?
Thanks.
@mowentian commented on GitHub (Dec 31, 2024):
400GB/s 是双向理论值,160GB/s 是我们的单向实测
@GHGmc2 commented on GitHub (Jan 1, 2025):
谢谢。那IB带宽50GB/s也是单向实测值吗?
@yiakwy-xpu-ml-framework-team commented on GitHub (Jan 1, 2025):
Could I get the nccl test snapshot (both algorithm and bus bandwidth) of all to all test with multiple nodes ? If the bandwidth data (160 GB/s) is from nccl-test, it hardly to be true. H800 should have the same IB/NCCL performance with H100 (450-490 GB/s) (NV SHARP)
@shenh10 commented on GitHub (Feb 6, 2025):
同问。IB connect X-8 单向 400Gb/s (50GB/s)像是用的理论值。 @mowentian
@renwuli commented on GitHub (Mar 5, 2025):
可以分享一下您的单向实测脚本吗?