deepseek-ai/DeepSeek-V3
1
0
Fork 0
mirror of https://github.com/deepseek-ai/DeepSeek-V3.git synced 2026-11-05 00:47:48 +00:00
Code Issues 25 Projects Packages Wiki Activity

[GH-ISSUE #659] [Question] about suggestions on hardware design in DeepSeek v3 paper #369

New issue
Closed
opened 2026-10-26 21:20:14 +00:00 by bsakaguchi · 1 comment
bsakaguchi commented 2026-10-26 21:20:14 +00:00
Owner
Copy link

Originally created by @RongHQ on GitHub (Feb 14, 2025).
Original GitHub issue: https://github.com/deepseek-ai/DeepSeek-V3/issues/659

I am a hardware designers and really appreciate that the DeepSeek v3 paper included suggestions on hardware design!

I have a few questions:

  1. Could you please share the methodology of the experiment where you measured Hopper Tensor Core FP8 GEMM accuracy? Paper states that only the highest 14bits are calculated in tensor core when doing FP8 GEMM.
  2. Do you have any suggestions on inference focused hardware? For example the FP8 accumulation precision needed and generally any other suggestions?

Thank you!

Originally created by @RongHQ on GitHub (Feb 14, 2025). Original GitHub issue: https://github.com/deepseek-ai/DeepSeek-V3/issues/659 I am a hardware designers and really appreciate that the DeepSeek v3 paper included suggestions on hardware design! I have a few questions: 1. Could you please share the methodology of the experiment where you measured Hopper Tensor Core FP8 GEMM accuracy? Paper states that only the highest 14bits are calculated in tensor core when doing FP8 GEMM. 2. Do you have any suggestions on inference focused hardware? For example the FP8 accumulation precision needed and generally any other suggestions? Thank you!
bsakaguchi 2026-10-26 21:20:14 +00:00
  • closed this issue
  • added the
    question
         label
bsakaguchi commented 2026-10-26 21:20:16 +00:00
Author
Owner
Copy link

@mowentian commented on GitHub (Feb 21, 2025):

We will elaborate on this in detail in our forthcoming paper.

<!-- gh-comment-id:2673038681 --> @mowentian commented on GitHub (Feb 21, 2025): We will elaborate on this in detail in our forthcoming paper.
Sign in to join this conversation.
No Branch/Tag specified
Branches Tags
main
v1.0.0
Labels
Clear labels   
closed-as-stale

   
enhancement

   
question

   
spam

   
stale
No labels
closed-as-stale
enhancement
question
spam
stale
Milestone
Clear milestone
No items
No milestone
Projects
Clear projects
No items
No project
Assignees
Clear assignees
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference
deepseek-ai/DeepSeek-V3#369
No description provided.