Can the GeForce RTX 5090 run Kimi K2 Instruct?
Does not fit at Q4_K_M.
Kimi K2 Instruct is 1000B parameters (MoE; all experts resident). At Q4_K_M the weights are about 558.8 GiB; with the 1.5 GiB runtime reserve that is 560.3 GiB against the card's 32GB, about 528.3 GiB more than the card has.
All three quants
32GB card| Quant | Weights | With reserve | Verdict | Headroom |
|---|---|---|---|---|
| Q4_K_Mthe everyday quant | 558.8 GiB | 560.3 GiB | Does not fit | short 528.3 GiB |
| Q8_0near-lossless | 987.2 GiB | 988.7 GiB | Does not fit | short 956.7 GiB |
| FP16full weights | 1862.6 GiB | 1864.1 GiB | Does not fit | short 1832.1 GiB |
Verdict rules: fits comfortably when weights plus reserve sit within 85% of VRAM; tight when they fit with little left for context; CPU offload when they exceed VRAM by up to 2x (runs, slowly, with layers in system RAM); no beyond that. The KV cache per token depends on the model's architecture and is not modelled here.
Not this card. These fit it comfortably at Q4.
Nothing in the consumer catalog fits Kimi K2 Instruct at Q4 on a single card; this is multi-GPU or datacenter territory, or a job for hosted inference (see PicksByModel for per-token pricing).
What the GeForce RTX 5090 does run well → · Kimi K2 Instruct on every card →
About the model
Kimi K2 Instruct by Moonshot AI: 131,072-token native context, text, released 2025 under the Modified MIT. Model card → · Hosted pricing on PicksByModel →