Can Kimi K3.1 Pro run on the H100 PCIe?
No. Even at 2-bit with 8K context, Kimi K3.1 Pro needs about 972 GB, more than the 76.0 GB usable on the H100 PCIe.
Even with llama.cpp CPU offload (-ngl, --n-cpu-moe) and 128 GB of system RAM, this model does not fit at 2-bit, or the card cannot hold a useful share of it. See the offload method.
Memory by quant and context
| Quant | 4K ctx | 8K ctx | 32K ctx | 128K ctx | Max tok/s (est., 8K) |
|---|---|---|---|---|---|
| 2-bit (Q2_K) | 972 GB No | 972 GB No | 975 GB No | 985 GB No | — |
| 3-bit (Q3_K_M) | 1267 GB No | 1267 GB No | 1269 GB No | 1279 GB No | — |
| 4-bit (Q4_K_M) | 1584 GB No | 1584 GB No | 1587 GB No | 1596 GB No | — |
| 5-bit (Q5_K_M) | 1846 GB No | 1846 GB No | 1849 GB No | 1859 GB No | — |
| 6-bit (Q6_K) | 2124 GB No | 2125 GB No | 2127 GB No | 2137 GB No | — |
| 8-bit (Q8_0) | 2752 GB No | 2753 GB No | 2755 GB No | 2765 GB No | — |
| 16-bit (F16/BF16) | 5179 GB No | 5180 GB No | 5182 GB No | 5192 GB No | — |
• real GGUF file size. Speed is a bandwidth ceiling (2000 GB/s), not a benchmark. Methodology.
What fits instead
- Nothing we list runs Kimi K3.1 Pro fully in memory at 8K context. It needs a multi-GPU server.
- Largest popular model that fits the H100 PCIe: DeepSeek V4 Flash 0731 (304B, MoE) at F16, about 11.9 GB at 8K context.
"Runs" and "fits" mean the whole model in GPU memory at 8K context (4-bit or better for the model pick). Prices are US launch MSRPs, not today's street prices. "Popular" means the 60 most-downloaded models we track.
More
Kimi K3.1 Pro on every GPU and Mac · Every model that runs on the H100 PCIe · Other models that fit 64 GB VRAM · Try other settings in the calculator
Guides: Best GPU for local LLMs by budget · Can I run gpt-oss, Llama 70B or Qwen3? · Mac vs PC for local LLMs
Data: Hugging Face Hub (moonshotai/Kimi-K3.1-Pro), NVIDIA specs. Model data last checked 2026-11-17; refreshed hourly.