What GPU do you need to run Llama 4 Scout?
109B params 128K max context VisionTool use
The smallest GPU memory that runs Llama 4 Scout well is 72 GB (Q4_K_M, e.g. RTX PRO 5000 Blackwell 72GB). 15 of 132 GPUs run it well.
MMLU 79.6
VRAM needed per quantization
| Quantization | Weights | 16K context | 32K context | Full 128K |
|---|---|---|---|---|
| Q4_K_M | 65.4 GB | 68.3 GB | 71.3 GB | 88.8 GB |
| Q8_0 | 115.0 GB | 117.9 GB | 120.9 GB | 138.4 GB |
Weights plus fp16 KV cache. Leave ~1 GB headroom for the inference engine.
Cheapest GPUs that can run Llama 4 Scout
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 Llama 4 Scout
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 PRO 5000 Blackwell 72GB 72 GB Q4_K_M 36K context · ~20 tok/s
- RTX PRO 5500 Blackwell 84 GB Q4_K_M 101K context · ~21 tok/s
- RTX PRO 6000 Blackwell 96 GB Q4_K_M 128K context · ~27 tok/s
- DGX Spark (GB 10) 128 GB Q8_0 71K context · ~2 tok/s
- RTX Spark 128 GB Q8_0 71K context · ~3 tok/s
AMD
- Ryzen AI Max (Strix Halo) 96 GB Q4_K_M 128K context · ~4 tok/s
- Ryzen AI Max 400 (Gorgon Halo) 128 GB Q8_0 71K context · ~2 tok/s
Apple
- M2 Max 96 GB Q4_K_M 128K context · ~6 tok/s
- M3 Max 96 GB Q4_K_M 128K context · ~5 tok/s
- M3 Ultra 96 GB Q4_K_M 128K context · ~12 tok/s
- M5 Ultra 96 GB Q4_K_M 128K context · ~18 tok/s
- M1 Ultra 128 GB Q8_0 71K context · ~7 tok/s
- M2 Ultra 128 GB Q8_0 71K context · ~7 tok/s
- M4 Max 128 GB Q8_0 71K context · ~5 tok/s
- M5 Max 128 GB Q8_0 71K context · ~5 tok/s