AMD MI355X vLLM HAS BEATEN B200 vLLM ON KIMI K2.5 (THE SAME MODEL ARCH AS XAI CURSOR COMPOSER 2.5)🚀🚨
SemiAnalysis Twitter · SemiAnalysis (@SemiAnalysis_) · 2026-08-01
SemiAnalysis announces that AMD MI355X running vLLM has surpassed NVIDIA B200 vLLM on Kimi K2.5 inference throughput, using community-contributed upstream kernels from the GPU_MODE hackathon.
Appears in
Extraction
Topics: gpu-inference-performanceamd-mi355xvllmkimi-models
Claims
- AMD MI355X vLLM has beaten NVIDIA B200 vLLM on Kimi K2.5 inference performance.
- Kimi K2.5 shares the same model architecture as xAI Cursor Composer 2.5.
- The performance gains were achieved using upstream AMD kernels contributed by the GPU_MODE community.
Key quotes
AMD MI355X vLLM HAS BEATEN B200 vLLM ON KIMI K2.5 (THE SAME MODEL ARCH AS XAI CURSOR COMPOSER 2.5)🚀🚨
This uses upstream AMD kernels from the @GPU_MODE community.