TokenSpeed Hits New SOTA Running Kimi K3 on AMD MI355X

TokenSpeed achieved a new SOTA for Kimi K3 inference on 8x AMD Instinct MI355X GPUs, reaching 216 tokens/s per user at C1 precision. The decoding speed is 1.4x faster than the previous ATOM approach.

2026-10-10 ~ 2026-10-10 · 2 related posts

• TokenSpeed hits SOTA for Kimi K3 on AMD MI355X: 1.4x faster than ATOM at 216 tok/s — zhyncs42 · 2026-10-10
• Kimi K3 inference on 8× AMD MI355X hits 216 tok/s per user with TokenSpeed — zhyncs42 · 2026-10-10