What GPU do you need to run LFM2.5 24B A2B?
23.97B params 128K max context Tool useReasoning
The smallest GPU memory that runs LFM2.5 24B A2B well is 16 GB (Q4_K_M, e.g. RTX PRO 2000 Blackwell). 72 of 132 GPUs run it well. It's a mixture-of-experts model, so all weights must fit in memory but speed depends only on the active experts.
SWE-bench Verified 51
VRAM needed per quantization
| Quantization | Weights | 16K context | 32K context | Full 128K |
|---|---|---|---|---|
| Q4_K_M | 13.5 GB | 13.8 GB | 14.1 GB | 16.0 GB |
| Q8_0 | 23.7 GB | 24.0 GB | 24.3 GB | 26.2 GB |
Weights plus fp16 KV cache. Leave ~1 GB headroom for the inference engine.
Cheapest GPUs that can run LFM2.5 24B A2B
Lowest launch MSRP among GPUs that run its smallest quantization (Q4_K_M) well, with at least 16K context.
As an Amazon Associate, I earn from qualifying purchases.
GPUs that can run LFM2.5 24B A2B
Best quantization that fits on each GPU, with the context and speed it gets. For chips with several memory sizes, the smallest that works is shown.
NVIDIA
- RTX 4060 Ti 16GB 16 GB Q4_K_M 128K context · ~121 tok/s
- RTX 4070 Ti SUPER 16 GB Q4_K_M 128K context · ~282 tok/s
- RTX 4080 16 GB Q4_K_M 128K context · ~301 tok/s
- RTX 4080 SUPER 16 GB Q4_K_M 128K context · ~309 tok/s
- RTX 5060 Ti 16GB 16 GB Q4_K_M 128K context · ~188 tok/s
- RTX 5070 Ti 16 GB Q4_K_M 128K context · ~376 tok/s
- RTX 5080 16 GB Q4_K_M 128K context · ~403 tok/s
- RTX 2000 Ada 16 GB Q4_K_M 128K context · ~94 tok/s
- RTX A4000 16 GB Q4_K_M 128K context · ~188 tok/s
- RTX PRO 2000 Blackwell 16 GB Q4_K_M 128K context · ~121 tok/s
- Tesla T4 16 GB Q4_K_M 128K context · ~134 tok/s
- Tesla V100 16GB 16 GB Q4_K_M 128K context · ~378 tok/s
- RTX 4000 Ada 20 GB Q4_K_M 128K context · ~151 tok/s
- RTX 3090 24 GB Q4_K_M 128K context · ~393 tok/s
- RTX 3090 Ti 24 GB Q4_K_M 128K context · ~424 tok/s
- RTX 4090 24 GB Q4_K_M 128K context · ~424 tok/s
- RTX 4500 Ada 24 GB Q4_K_M 128K context · ~182 tok/s
- RTX A5000 24 GB Q4_K_M 128K context · ~323 tok/s
- RTX PRO 4000 Blackwell 24 GB Q4_K_M 128K context · ~282 tok/s
- RTX Spark 24 GB Q4_K_M 128K context · ~126 tok/s
- Tesla P40 24 GB Q4_K_M 128K context · ~145 tok/s
- RTX 5090 32 GB Q8_0 128K context · ~522 tok/s
- RTX 5000 Ada 32 GB Q8_0 128K context · ~168 tok/s
- RTX PRO 4500 Blackwell 32 GB Q8_0 128K context · ~261 tok/s
- Tesla V100 32GB 32 GB Q8_0 128K context · ~262 tok/s
- L40S 48 GB Q8_0 128K context · ~252 tok/s
- RTX 6000 Ada 48 GB Q8_0 128K context · ~280 tok/s
- RTX A6000 48 GB Q8_0 128K context · ~224 tok/s
- RTX PRO 5000 Blackwell 48 GB Q8_0 128K context · ~392 tok/s
- RTX PRO 5000 Blackwell 72GB 72 GB Q8_0 128K context · ~392 tok/s
- RTX PRO 5500 Blackwell 84 GB Q8_0 128K context · ~408 tok/s
- RTX PRO 6000 Blackwell 96 GB Q8_0 128K context · ~522 tok/s
- DGX Spark (GB 10) 128 GB Q8_0 128K context · ~80 tok/s
AMD
- AMD BC-250 16 GB Q4_K_M 128K context · ~188 tok/s
- Radeon RX 6800 16 GB Q4_K_M 128K context · ~215 tok/s
- Radeon RX 6800 XT 16 GB Q4_K_M 128K context · ~215 tok/s
- Radeon RX 6900 XT 16 GB Q4_K_M 128K context · ~215 tok/s
- Radeon RX 6950 XT 16 GB Q4_K_M 128K context · ~242 tok/s
- Radeon RX 7600 XT 16 GB Q4_K_M 128K context · ~121 tok/s
- Radeon RX 7800 XT 16 GB Q4_K_M 128K context · ~262 tok/s
- Radeon RX 7900 GRE 16 GB Q4_K_M 128K context · ~242 tok/s
- Radeon RX 9060 XT 16GB 16 GB Q4_K_M 128K context · ~134 tok/s
- Radeon RX 9070 16 GB Q4_K_M 128K context · ~269 tok/s
- Radeon RX 9070 XT 16 GB Q4_K_M 128K context · ~269 tok/s
- Radeon RX 7900 XT 20 GB Q4_K_M 128K context · ~336 tok/s
- Radeon RX 7900 XTX 24 GB Q4_K_M 128K context · ~403 tok/s
- Radeon AI PRO R9700 32 GB Q8_0 128K context · ~187 tok/s
- Ryzen AI Max (Strix Halo) 32 GB Q8_0 128K context · ~75 tok/s
- Ryzen AI Max 400 (Gorgon Halo) 64 GB Q8_0 128K context · ~80 tok/s
Intel
- Arc A770 16GB 16 GB Q4_K_M 128K context · ~235 tok/s
- Arc Pro B65 32 GB Q8_0 128K context · ~177 tok/s
- Arc Pro B70 32 GB Q8_0 128K context · ~177 tok/s
Apple
- M1 16 GB Q4_K_M 128K context · ~29 tok/s
- M1 Pro 16 GB Q4_K_M 128K context · ~84 tok/s
- M2 16 GB Q4_K_M 128K context · ~42 tok/s
- M2 Pro 16 GB Q4_K_M 128K context · ~84 tok/s
- M3 16 GB Q4_K_M 128K context · ~42 tok/s
- M4 16 GB Q4_K_M 128K context · ~50 tok/s
- M5 16 GB Q4_K_M 128K context · ~64 tok/s
- M6 16 GB Q4_K_M 128K context · ~64 tok/s
- M3 Pro 18 GB Q4_K_M 128K context · ~63 tok/s
- M4 Pro 24 GB Q4_K_M 128K context · ~115 tok/s
- M5 Pro 24 GB Q4_K_M 128K context · ~129 tok/s
- M1 Max 32 GB Q8_0 128K context · ~117 tok/s
- M2 Max 32 GB Q8_0 128K context · ~117 tok/s
- M3 Max 36 GB Q8_0 128K context · ~87 tok/s
- M4 Max 36 GB Q8_0 128K context · ~120 tok/s
- M5 Max 36 GB Q8_0 128K context · ~134 tok/s
- M1 Ultra 64 GB Q8_0 128K context · ~233 tok/s
- M2 Ultra 64 GB Q8_0 128K context · ~233 tok/s
- M3 Ultra 96 GB Q8_0 128K context · ~239 tok/s
- M5 Ultra 96 GB Q8_0 128K context · ~350 tok/s