Can Qwen2.5 3B Instruct run on the H100 PCIe?
Yes. At 8K context the best fit is F16, using about 6.7 GB of the 76.0 GB usable on the H100 PCIe. Estimated speed ceiling: ~324 tokens/sec.
Memory by quant and context
| Quant | 4K ctx | 8K ctx | 32K ctx | Max tok/s (est., 8K) |
|---|---|---|---|---|
| 2-bit (Q2_K) • | 2.1 GB Fits | 2.3 GB Fits | 3.1 GB Fits | 1453 |
| 3-bit (Q3_K_M) • | 2.4 GB Fits | 2.6 GB Fits | 3.4 GB Fits | 1160 |
| 4-bit (Q4_K_M) • | 2.8 GB Fits | 2.9 GB Fits | 3.8 GB Fits | 950 |
| 5-bit (Q5_K_M) • | 3.1 GB Fits | 3.2 GB Fits | 4.1 GB Fits | 820 |
| 6-bit (Q6_K) • | 3.4 GB Fits | 3.6 GB Fits | 4.4 GB Fits | 716 |
| 8-bit (Q8_0) • | 4.2 GB Fits | 4.3 GB Fits | 5.2 GB Fits | 553 |
| 16-bit (F16/BF16) | 6.6 GB Fits | 6.7 GB Fits | 7.6 GB Fits | 324 |
• real GGUF file size from Qwen/Qwen2.5-3B-Instruct-GGUF. Speed is a bandwidth ceiling (2000 GB/s), not a benchmark. Methodology.
More
Qwen2.5 3B Instruct on every GPU and Mac · Every model that runs on the H100 PCIe · Other models that fit 64 GB VRAM · Try other settings in the calculator
Data: Hugging Face Hub (Qwen/Qwen2.5-3B-Instruct), NVIDIA specs. Data as of 2026-11-17.