# vLLM-ATOM is a purpose-built plugin that aims to improve inference performance across various AI LLMs.

*genai · news · 2026-05-11 · Wccftech*

## Key points

- AMD's vLLM-ATOM plugin enables native kernel optimizations for MI350 and MI400 GPUs without vLLM code changes.
- vLLM-ATOM grants instant access to features like FP4 precision and rack-scale inference on AMD's newest GPUs.
- The plugin validates new hardware and kernel features, then upstreams mature optimizations to vLLM's ROCm backend.
- Users can run vLLM-ATOM as either a standalone server or as a plugin backend within vLLM workflows.
- vLLM-ATOM supports both LLMs and VLMs through a unified inference pipeline on AMD hardware.

**Companies:** AMD
**Countries:** United States

[Read the full story on Wccftech](https://wccftech.com/amd-vllm-atom-plugin-supercharges-deepseek-r1-kimi-k2-gpt-oss-120b-ai-llm-inference-on-instinct-mi350-mi400/)

---

Canonical: https://newsio.io/n/54745014-bde4-4d15-8799-e574d11f81cd/vllm-atom-is-a-purpose-built-plugin-that-aims-to-improve-inference-performance-a
Summarized by Newsio from Wccftech. https://newsio.io/how-it-works