Can LFM2.5 24B A2B run on the RTX 5090?
Yes. At 8K context the best fit is Q8_0, using about 24.5 GB of the 30.4 GB usable on the RTX 5090. Estimated speed ceiling: ~843 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) | 9.1 GB Fits | 9.2 GB Fits | 9.6 GB Fits | 11.5 GB Fits | 2389 |
| 3-bit (Q3_K_M) | 11.6 GB Fits | 11.7 GB Fits | 12.2 GB Fits | 14.1 GB Fits | 1833 |
| 4-bit (Q4_K_M) • | 14.3 GB Fits | 14.3 GB Fits | 14.8 GB Fits | 16.7 GB Fits | 1479 |
| 5-bit (Q5_K_M) • | 16.6 GB Fits | 16.7 GB Fits | 17.1 GB Fits | 19.0 GB Fits | 1261 |
| 6-bit (Q6_K) • | 19.1 GB Fits | 19.1 GB Fits | 19.6 GB Fits | 21.5 GB Fits | 1091 |
| 8-bit (Q8_0) • | 24.5 GB Fits | 24.5 GB Fits | 25.0 GB Fits | 26.9 GB Fits | 843 |
| 16-bit (F16/BF16) • | 90.1 GB No | 90.1 GB No | 90.6 GB No | 92.5 GB No | — |
• real GGUF file size from LiquidAI/LFM2.5-24B-A2B-GGUF. Speed is a bandwidth ceiling (1792 GB/s), not a benchmark. Methodology.
More
LFM2.5 24B A2B 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 (LiquidAI/LFM2.5-24B-A2B), NVIDIA specs. Data as of 2026-11-17.