Can Llama 2 7b hf run on the RTX 5090?
Yes. At 8K context the best fit is F16, using about 17.0 GB of the 30.4 GB usable on the RTX 5090. Estimated speed ceiling: ~133 tokens/sec.
Memory by quant and context
| Quant | 4K ctx | Max tok/s (est., 8K) |
|---|---|---|
| 2-bit (Q2_K) • | 4.8 GB Fits | 707 |
| 3-bit (Q3_K_M) • | 5.5 GB Fits | 543 |
| 4-bit (Q4_K_M) • | 6.3 GB Fits | 439 |
| 5-bit (Q5_K_M) • | 6.9 GB Fits | 375 |
| 6-bit (Q6_K) • | 7.6 GB Fits | 324 |
| 8-bit (Q8_0) | 9.1 GB Fits | 250 |
| 16-bit (F16/BF16) | 15.0 GB Fits | 133 |
• real GGUF file size from mradermacher/Llama-2-7b-hf-i1-GGUF. Speed is a bandwidth ceiling (1792 GB/s), not a benchmark. Methodology.
More
Llama 2 7b hf on every GPU and Mac · Every model that runs on the RTX 5090 · Other models that fit 32 GB VRAM · Try other settings in the calculator
Data: Hugging Face Hub (NousResearch/Llama-2-7b-hf), NVIDIA specs. Data as of 2026-11-17.