Kimi K3 deployment estimate points to 2 B200 nodes or 1 B300 class node
HarveenChadha · x · 2026-07-28
- The post estimates the minimum hardware needed to run Kimi K3 at its original MXFP4 weight / MXFP8 activation setup.
- The rough utilization estimates are:
- NVIDIA B200: 2 nodes (50% effective)
- NVIDIA B300: 1 node (67%)
- NVIDIA Rubin VR200: 1 node (67%)
- AMD MI335X: 1 node (67%)
- AMD MI455X: 1 node (45%)
- It is specifically a weight-utilization view; KV-cache requirements are said to be coming separately.
Related event: Kimi K3 Deployment: Single-Node B300 vs Dual-Node B200(2 posts)→
More from Infra
- Compute, not algorithms, is the real moat in frontier AI — GavinSBaker · 2026-07-28
- Renting GPUs and open-weight models cut one AI bill from $1.2M to $100K — kimmonismus · 2026-07-28
- Running Kimi-K3 Locally Costs Up to 1 Million Euros, Jokes German Tech Blogger — FlorianGallwitz · 2026-07-28
- Bittensor updates $TAO emissions with a 61% gate that favors top-valued subnets — markjeffrey · 2026-07-28
- Qualcomm says NPU latency beats GPU and CPU for on-device AI agents — qdrant_engine · 2026-07-28
- MLA-based KV cache costs 12 GB per million tokens, with KDA state at 230 MB BF16 — zephyr_z9 · 2026-07-28