Running 2.8T-param Kimi K3 on a 16x GB10 Cluster: 30 tok/s Coding Decode
ciprianveg · reddit · 2026-09-21
A developer shares a hands-on deployment of Moonshot AI's Kimi K3 (2.8T params) on a self-built 16x GB10 cluster:
- Decode: sustained 30 tok/s (peaking 38) during heavy code generation and agentic tasks
- Prefill: 750–910 tok/s, optimized via a modified NCCL topology and dual-switch layout
- Concurrency: multiple concurrent users without dropping generation rates or starving KV cache
- Context: stable multi-hundred-thousand-token runs with repeated 500k compaction
- Network: dual MikroTik CRS804-4DDQ switches with 4x 400G-to-4x100G breakouts
- Runtime: customized gb10-vllm stack with dspark / Inferact Kimi-K3-DSPark wrappers and custom MLA/KV kernels
All runtime patches, configs and build scripts are open-sourced on GitHub (gb10-vllm).
More from Infra
- Agent Substrate Calculator Estimates Session Resume Speed Under Load — pjausovec · 2026-11-12
- Dell expands AI Data Platform: agents have a data-context problem, not a model problem — DavidLinthicum · 2026-11-12
- SGLang Megakernel Pushes Muse Spark to 1,109 tok/s, 2.1x Over Baseline — hsu_byron · 2026-11-12
- Potion UI's local login system turns your PC into a multi-user AI server — cocktailpeanut · 2026-11-12
- AWS Finally Built Its Own AWS Emulator — super_linear · 2026-11-12
- Megakernel is multiprogramming for GPUs: 'GPUs are entering the CPU age' — pranjalssh · 2026-11-12