Can Kimi K2.5 run on the H100 SXM?
No. Even at 2-bit with 8K context, Kimi K2.5 needs about 360 GB, more than the 76.0 GB usable on the H100 SXM.
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) | 360 GB No | 360 GB No | 361 GB No | 368 GB No | — |
| 3-bit (Q3_K_M) | 468 GB No | 469 GB No | 470 GB No | 477 GB No | — |
| 4-bit (Q4_K_M) | 586 GB No | 586 GB No | 587 GB No | 594 GB No | — |
| 5-bit (Q5_K_M) | 682 GB No | 683 GB No | 684 GB No | 691 GB No | — |
| 6-bit (Q6_K) | 785 GB No | 785 GB No | 787 GB No | 794 GB No | — |
| 8-bit (Q8_0) | 1017 GB No | 1017 GB No | 1019 GB No | 1025 GB No | — |
| 16-bit (F16/BF16) | 1914 GB No | 1914 GB No | 1916 GB No | 1922 GB No | — |
• real GGUF file size. Speed is a bandwidth ceiling (3350 GB/s), not a benchmark. Methodology.
Hardware that does fit it
What fits instead
- No listed GPU with a list price runs Kimi K2.5 fully in memory. The smallest listed hardware that does is the Mac M3 Ultra 512GB (512 GB).
- Largest popular model that fits the H100 SXM: gpt-oss 120B (117B, MoE) at F16, about 61.9 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 K2.5 on every GPU and Mac · Every model that runs on the H100 SXM · 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-K2.5), NVIDIA specs. Model data last checked 2026-11-17; refreshed hourly.