Can Qwen3.8 27B run on the 2x RTX 5090?
Yes. At 8K context the best fit is F16, using about 52.3 GB of the 60.8 GB usable on the 2x RTX 5090. Estimated speed ceiling: ~33 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) | 10.8 GB Fits | 11.1 GB Fits | 12.6 GB Fits | 18.6 GB Fits | 172 |
| 3-bit (Q3_K_M) | 13.8 GB Fits | 14.0 GB Fits | 15.5 GB Fits | 21.5 GB Fits | 132 |
| 4-bit (Q4_K_M) • | 16.6 GB Fits | 16.8 GB Fits | 18.3 GB Fits | 24.3 GB Fits | 108 |
| 5-bit (Q5_K_M) • | 19.6 GB Fits | 19.9 GB Fits | 21.4 GB Fits | 27.4 GB Fits | 90 |
| 6-bit (Q6_K) • | 21.6 GB Fits | 21.8 GB Fits | 23.3 GB Fits | 29.3 GB Fits | 82 |
| 8-bit (Q8_0) • | 28.2 GB Fits | 28.4 GB Fits | 29.9 GB Fits | 35.9 GB Fits | 62 |
| 16-bit (F16/BF16) • | 52.0 GB Fits | 52.3 GB Fits | 53.8 GB Fits | 59.8 GB Tight | 33 |
• real GGUF file size from unsloth/Qwen3.8-27B-GGUF. Speed is a bandwidth ceiling (1792 GB/s), not a benchmark. Methodology.
More
Qwen3.8 27B on every GPU and Mac · Every model that runs on the 2x RTX 5090 · Try other settings in the calculator
Data: Hugging Face Hub (Qwen/Qwen3.8-27B), NVIDIA specs. Data as of 2026-11-17.