Wafer serves Kimi K3 on AMD MI355X with 3.8x throughput and 71% lower cost vs B200
SumitGup · x · 2026-08-03
Wafer announced serving Kimi K3 on AMD MI355X, achieving 3.8x higher throughput and 71% lower cost compared to B200. Kimi K3 has 2.8T parameters, requiring over 1.5TB VRAM, which cannot fit on a single B200 node. MI355X offers 288GB VRAM per GPU, priced 2.4x cheaper than B300 and 1.7x cheaper than B200. On a 1024-token input/400-token output benchmark, MI355X reaches 952 tok/s/node and 118 tok/s single stream, over 3.8x aggregate throughput per node. AMD provided day-0 support for Kimi K3, reducing engineering effort.
Related event: AMD MI355X Boosts Kimi K3 Throughput and Cuts Costs(3 posts)→
More from Infra
- Nemotron 3 Nano Omni Hits 264 tok/s Native on DGX Spark — ivan_bezdomny · 2026-08-03
- Global AI compute to hit 200M H100-equivalents by 2028, fueling agentic loop toward ASI — 新智元 · 2026-08-03
- tinybox Dual-GPU Edition Hits 245 tok/s Running DeepSeek — AccBalanced · 2026-08-03
- Troubleshooting KV Cache Misses Caused by Multiple Agent Tool Calls — CentrifugalMalaise · 2026-08-03
- Amazon Graviton Revenue Commitments Surge Nearly 3x QoQ — Beth_Kindig · 2026-08-03
- Epoch AI Report: A Single AI Chat Uses Less Energy Than a 10-Sec Microwave Run — Jsevillamol · 2026-08-03