Can Olmo 3 7B Instruct run on the H200?
Yes. At 8K context the best fit is F16, using about 16.7 GB of the 134 GB usable on the H200. Estimated speed ceiling: ~329 tokens/sec.
Memory by quant and context
| Quant | 4K ctx | 8K ctx | 32K ctx | Max tok/s (est., 8K) |
|---|---|---|---|---|
| 2-bit (Q2_K) • | 5.3 GB Fits | 5.8 GB Fits | 8.8 GB Fits | 1680 |
| 3-bit (Q3_K_M) • | 6.0 GB Fits | 6.5 GB Fits | 9.5 GB Fits | 1315 |
| 4-bit (Q4_K_M) • | 6.8 GB Fits | 7.3 GB Fits | 10.3 GB Fits | 1073 |
| 5-bit (Q5_K_M) • | 7.4 GB Fits | 7.9 GB Fits | 10.9 GB Fits | 922 |
| 6-bit (Q6_K) • | 8.2 GB Fits | 8.7 GB Fits | 11.7 GB Fits | 801 |
| 8-bit (Q8_0) • | 9.8 GB Fits | 10.3 GB Fits | 13.3 GB Fits | 619 |
| 16-bit (F16/BF16) • | 16.2 GB Fits | 16.7 GB Fits | 19.7 GB Fits | 329 |
• real GGUF file size from unsloth/Olmo-3-7B-Instruct-GGUF. Speed is a bandwidth ceiling (4800 GB/s), not a benchmark. Methodology.
More
Olmo 3 7B Instruct 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 (allenai/Olmo-3-7B-Instruct), NVIDIA specs. Data as of 2026-11-17.