# Wafer runs Kimi K3 on AMD MI355X — 48 tok/s/$, 45% more per dollar than B300

> Kimi K3's 2.8T weights fit on eight MI355X cards and serve at 48 tokens per dollar — 1.45x the B300, 6.9x the B200.

- Org: Wafer
- Date: 2026-07-31
- Category: showcase
- Source: https://wafer.ai/blog/kimi-k3-mi355x
- Canonical page: https://ai-tldr.dev/releases/wafer-kimi-k3-mi355x-jul31/

## In short

Wafer engineer Ian Ye benchmarks Moonshot's 2.8T Kimi K3 on AMD MI355X with sglang and block-diffusion speculative decoding. MI355X hits 952 tok/s/node and 48 tok/s/$ vs 33 tok/s/$ on NVIDIA B300.

## What it is

Wafer, a Y Combinator-backed inference company, has published a benchmark of Moonshot's 2.8T Kimi K3 on AMD MI355X GPUs. The team compares eight-GPU MI355X (TP8), eight-GPU B300 (TP8+DCP8), and two-node sixteen-GPU B200 (TP16) setups on the same 1M-context model.

## How it works

Kimi K3 needs over 1.5TB of VRAM before its 1M-token KV cache, which just fits on eight MI355X cards (288GB each). Wafer served it through sglang with RadixArk's Kimi-K3-DSpark block-diffusion draft model for speculative decoding, plus AITER MLA kernel padding fixes that lift cold-prefill by 2 to 3 times.

## Why it matters

At AMD's typical $2.50/GPU-hr, MI355X delivers 48 tok/s/$ against 33 tok/s/$ on B300 and 7 tok/s/$ on B200. For teams already running open weights at trillion-plus scale, this is the second Wafer post in a month showing frontier inference on AMD at a real cost advantage — CUDA lock-in on the biggest MoEs keeps loosening.

## Who it's for

inference-cost-sensitive teams, AMD hardware evaluators, Kimi K3 deployers

## Links

- [Blog Post](https://wafer.ai/blog/kimi-k3-mi355x)
- [Prior GLM-5.2 Post](https://www.wafer.ai/blog/glm52-amd)
- [Kimi K3 Model Card](https://huggingface.co/moonshotai/Kimi-K3)
- [HN Discussion](https://news.ycombinator.com/item?id=49141073)