Can flint run on the RTX 5090?
Yes. At 8K context the best fit is Q5_K_M, using about 27.3 GB of the 30.4 GB usable on the RTX 5090. Estimated speed ceiling: ~2012 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) | 14.7 GB Fits | 14.8 GB Fits | 15.5 GB Fits | 18.1 GB Fits | 3823 |
| 3-bit (Q3_K_M) | 18.9 GB Fits | 19.0 GB Fits | 19.7 GB Fits | 22.3 GB Fits | 2933 |
| 4-bit (Q4_K_M) | 23.5 GB Fits | 23.6 GB Fits | 24.2 GB Fits | 26.9 GB Fits | 2345 |
| 5-bit (Q5_K_M) | 27.2 GB Fits | 27.3 GB Fits | 28.0 GB Tight | 30.6 GB No | 2012 |
| 6-bit (Q6_K) | 31.2 GB No | 31.3 GB No | 32.0 GB No | 34.6 GB No | — |
| 8-bit (Q8_0) • | 40.2 GB No | 40.3 GB No | 41.0 GB No | 43.6 GB No | — |
| 16-bit (F16/BF16) • | 74.9 GB No | 75.1 GB No | 75.7 GB No | 78.3 GB No | — |
• real GGUF file size from unsloth/flint-GGUF. Speed is a bandwidth ceiling (1792 GB/s), not a benchmark. Methodology.
More
flint 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 (RecursiveSI/flint), NVIDIA specs. Data as of 2026-11-17.