Kimi-K2-Instruct-0905 — which GPUs fit

Moonshot · 1026.5B (active 32B) · MoE · ctx 256K · MIT · HuggingFace

Kimi-K2-Instruct-0905 (Moonshot) is a 1026.5B parameters MoE (32B active per token) open-weight model tracked on AI Hashrate. At 8K context we estimate roughly 571.98 GB VRAM for Q4 (weights 564.58 GB + KV 6.4 GB + 1 GB overhead) and 2060.4 GB for FP16. About 0 GPUs/accelerators in our catalog fully fit this model at Q4 under that context assumption. Fastest Q4 configs: B200 192GB SXM ≈ 51.2 tok/s (estimated); MI300X 192GB ≈ 33.9 tok/s (estimated); Mac Studio M2 Ultra 192GB ≈ 5.1 tok/s (estimated). Use the table below for VRAM fit and tok/s per dollar (list MSRP). Estimates follow memory-bandwidth math; measured rows override when present. See methodology for details. Methodology.

3 hardware configs — sorted by tok/s. Default fit context 8K.

Hardware	VRAM	Quant	tok/s	Fits?	tok/s/$
B200 192GB SXM	192	Q4	51.2 est.	No	0.001
MI300X 192GB	192	Q4	33.9 est.	No	0.002
Mac Studio M2 Ultra 192GB	192	Q4	5.1 est.	No	0.001