Can DeepSeek V3.2 Exp run on the H100 SXM?
No. Even at 2-bit with 8K context, DeepSeek V3.2 Exp needs about 241 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) | 240 GB No | 241 GB No | 243 GB No | 251 GB No | — |
| 3-bit (Q3_K_M) | 313 GB No | 313 GB No | 315 GB No | 323 GB No | — |
| 4-bit (Q4_K_M) | 391 GB No | 391 GB No | 393 GB No | 401 GB No | — |
| 5-bit (Q5_K_M) | 456 GB No | 456 GB No | 458 GB No | 466 GB No | — |
| 6-bit (Q6_K) | 524 GB No | 525 GB No | 527 GB No | 535 GB No | — |
| 8-bit (Q8_0) | 679 GB No | 680 GB No | 681 GB No | 689 GB No | — |
| 16-bit (F16/BF16) | 1278 GB No | 1278 GB No | 1280 GB No | 1288 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 DeepSeek V3.2 Exp 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
DeepSeek V3.2 Exp 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 (deepseek-ai/DeepSeek-V3.2-Exp), NVIDIA specs. Model data last checked 2026-11-17; refreshed hourly.