Can phi 4 run on the H200?
Yes. At 8K context the best fit is F16, using about 29.5 GB of the 134 GB usable on the H200. Estimated speed ceiling: ~164 tokens/sec.
Memory by quant and context
| Quant | 4K ctx | 8K ctx | Max tok/s (est., 8K) |
|---|---|---|---|
| 2-bit (Q2_K) • | 6.6 GB Fits | 7.4 GB Fits | 856 |
| 3-bit (Q3_K_M) • | 8.1 GB Fits | 8.9 GB Fits | 668 |
| 4-bit (Q4_K_M) • | 9.7 GB Fits | 10.4 GB Fits | 540 |
| 5-bit (Q5_K_M) • | 11.1 GB Fits | 11.9 GB Fits | 461 |
| 6-bit (Q6_K) • | 12.6 GB Fits | 13.4 GB Fits | 399 |
| 8-bit (Q8_0) • | 15.9 GB Fits | 16.7 GB Fits | 308 |
| 16-bit (F16/BF16) • | 28.7 GB Fits | 29.5 GB Fits | 164 |
• real GGUF file size from unsloth/phi-4-GGUF. Speed is a bandwidth ceiling (4800 GB/s), not a benchmark. Methodology.
More
phi 4 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 (microsoft/phi-4), NVIDIA specs. Data as of 2026-11-17.