Can Llama 3.2 11B Vision run on the H200?
Yes. At 8K context the best fit is F16, using about 21.7 GB of the 134 GB usable on the H200. Estimated speed ceiling: ~226 tokens/sec.
Memory by quant and context
| Quant | 4K ctx | 8K ctx | 32K ctx | 128K ctx | Max tok/s (est., 8K) |
|---|---|---|---|---|---|
| 2-bit (Q2_K) | 5.0 GB Fits | 5.6 GB Fits | 9.4 GB Fits | 24.4 GB Fits | 1203 |
| 3-bit (Q3_K_M) | 6.1 GB Fits | 6.7 GB Fits | 10.5 GB Fits | 25.5 GB Fits | 923 |
| 4-bit (Q4_K_M) | 7.3 GB Fits | 8.0 GB Fits | 11.7 GB Fits | 26.7 GB Fits | 738 |
| 5-bit (Q5_K_M) | 8.3 GB Fits | 9.0 GB Fits | 12.7 GB Fits | 27.7 GB Fits | 633 |
| 6-bit (Q6_K) | 9.4 GB Fits | 10.0 GB Fits | 13.8 GB Fits | 28.8 GB Fits | 550 |
| 8-bit (Q8_0) | 11.8 GB Fits | 12.4 GB Fits | 16.2 GB Fits | 31.2 GB Fits | 424 |
| 16-bit (F16/BF16) | 21.1 GB Fits | 21.7 GB Fits | 25.5 GB Fits | 40.5 GB Fits | 226 |
• real GGUF file size. Speed is a bandwidth ceiling (4800 GB/s), not a benchmark. Methodology.
More
Llama 3.2 11B Vision on every GPU and Mac · Every model that runs on the H200 · Other models that fit 64 GB VRAM · Try other settings in the calculator
Data: Hugging Face Hub (meta-llama/Llama-3.2-11B-Vision), NVIDIA specs. Data as of 2026-11-17.