[GH-ISSUE #43] Questions about accumulator precision #29

New issue

Closed

opened 2026-10-26 21:10:52 +00:00 by bsakaguchi · 1 comment

bsakaguchi commented

2026-10-26 21:10:52 +00:00

Owner

Originally created by @rchardx on GitHub (Mar 6, 2025).
Original GitHub issue: https://github.com/deepseek-ai/DeepGEMM/issues/43

However, for example, to achieve precise FP32 results from the accumulation of 32 FP8×FP8 multiplications, at least 34-bit precision is required.

Could you tell me why 34-bit precision is required?

Originally created by @rchardx on GitHub (Mar 6, 2025). Original GitHub issue: https://github.com/deepseek-ai/DeepGEMM/issues/43 > However, for example, to achieve precise FP32 results from the accumulation of 32 FP8×FP8 multiplications, at least 34-bit precision is required. Could you tell me why 34-bit precision is required?

bsakaguchi closed this issue

2026-10-26 21:10:52 +00:00

bsakaguchi commented

2026-10-26 21:10:54 +00:00

Author

Owner

@soundOfDestiny commented on GitHub (Mar 6, 2025):

It's irrelevant to DeepGEMM.
You can report this in our DeepSeek-V3 repo. (e.g. https://github.com/deepseek-ai/DeepSeek-V3/issues/197)

@soundOfDestiny commented on GitHub (Mar 6, 2025): It's irrelevant to DeepGEMM. You can report this in our [DeepSeek-V3](https://github.com/deepseek-ai/DeepSeek-V3) repo. (e.g. https://github.com/deepseek-ai/DeepSeek-V3/issues/197)

bsakaguchi referenced this issue

2026-10-26 21:14:55 +00:00

[PR #29] [CLOSED] Remove if else check in the loop #212