Skip to content
Trending storyDeveloping

DeepSeek avoids tensor parallelism in inference, even on Blackwell GPUs

1 article1 sourcesince Oct 20Last article 1h ago ·

Overview

AISummary of 1 article

DeepSeek still does not use tensor parallelism in its inference setup, even on Blackwell GPUs with standard interconnect, according to a post by the X user wh (@nrehiew_). wh says the choice traces back to DeepSeek V3, which ran on H800 GPUs with limited interconnect. wh notes that the default recipes in vLLM and SGLang both recommend tensor parallelism.

Written by AI from the articles below · updated Oct 20, 10:52 AM ET

Check the sources:

Latest developmentOct 20, 10:50 AM ET
DeepSeek still avoids tensor parallelism even on Blackwell GPUs

Article timeline

The articles in this story. Times are ET.

Oct 20
  1. whX
    DeepSeek still avoids tensor parallelism even on Blackwell GPUs

    AIwh says DeepSeek still does not use tensor parallelism in its inference setup, even on Blackwell GPUs with standard interconnect, a choice that traces back to DeepSeek V3 on H800s with limited interconnect. wh notes that the default recipes in vLLM and SGLang both recommend tensor parallelism.

    Image from @nrehiew_'s post

Heat Trend

Not enough continuous observations to show a trend yet.