@GPU_MODE For disaggregated upstream vLLM performance, AMD is still behind on Kimi models, but we are looking forward to…

SemiAnalysis Twitter · SemiAnalysis (@SemiAnalysis_) · 2026-08-01

SemiAnalysis notes that AMD MI355X still trails NVIDIA on disaggregated upstream vLLM performance for Kimi models, while anticipating further improvements and upcoming benchmarks on Kimi K3.

Open original ↗

Appears in

AMD MI355X Beats NVIDIA B200 on Kimi K2.5 Inference via Community Hackathon

Extraction

Topics: gpu-inference-performanceamd-mi355xvllmkimi-models

Claims

AMD MI355X still lags behind NVIDIA on disaggregated upstream vLLM performance for Kimi models.
Further performance improvements to AMD's vLLM stack are anticipated.
AMD vLLM performance on the next-generation Kimi K3 model is expected to be evaluated and published.

Key quotes

For disaggregated upstream vLLM performance, AMD is still behind on Kimi models, but we are looking forward to improvements there as well, in addition to seeing AMD vLLM performance on Kimi K3.