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 Runs Kimi K3 at 952 tok/s, Outperforming B200(6 posts)→
More from Infra
- China's AI boom drives up to 660% profit surges for domestic chip firms; GPU maker MetaX turns profitable — pstAsiatech · 2026-10-16
- When Does an Agent Actually Need an API Gateway? Reddit Weighs Direct Calls vs. Unified Endpoints — TurbulentStiffness · 2026-10-16
- Taiwan to build $4B AI manufacturing complex in Texas to back Nvidia and TSMC — Polymarket · 2026-10-16
- When the KV Cache Is Slower Than the GPU: Remote KV Tiers Backfire in Production — MultiverseComputingCAI · 2026-10-16
- GitHub rewrote the 800k-line Copilot runtime in Rust using AI agents, cutting CPU and memory by over 90% — bibryam · 2026-10-16
- NVIDIA quietly delists the entry-level RTX 5050 from its marketplace, following the RTX 5090 — chemist_slime · 2026-10-16