Skip to content
Trending storyDeveloping

vLLM adds Day 0 BF16 and NVFP4 support for a model on NVIDIA Blackwell GPUs

1 article1 sourcesince Oct 15Last article 1h ago ·

Overview

AISummary of 1 article

vLLM announces Day 0 support for the BF16 and NVFP4 versions of a model on NVIDIA Blackwell GPUs.

The project says BF16 serves as the baseline, while NVFP4 gives a smaller weight footprint and low-precision execution.

NVIDIA reports over 600 output tokens/sec at 100K input tokens on Artificial Analysis agentic tasks, using a Baseten endpoint. A separate SWE-bench Pro comparison reports 4× faster task completion on 4× B200 GPUs with NVFP4 at 50 TPS/user interactivity. These performance figures come from NVIDIA and the comparison is not independently verified.

Written by AI from the articles below · updated Oct 15, 12:02 PM ET

Check the sources:

Latest developmentOct 15, 11:50 AM ET
vLLM adds BF16 and NVFP4 support for NVIDIA Blackwell GPUs

Article timeline

The articles in this story. Times are ET.

Oct 15
  1. vLLMOfficial
    vLLM adds BF16 and NVFP4 support for NVIDIA Blackwell GPUs

    AIvLLM adds Day 0 support for BF16 and NVFP4 model versions, with NVFP4 offering a smaller weight footprint and low-precision execution on NVIDIA Blackwell GPUs. NVIDIA reports over 600 output tokens/sec at 100K input tokens on Artificial Analysis agentic tasks via a Baseten endpoint. A separate SWE-bench Pro comparison reports 4× faster task completion on 4× B200 GPUs using NVFP4 at 50 TPS/user interactivity.

    Image from @vllm_project's post

Heat trend

Not enough continuous observations to show a trend yet.