What GPU do you need to run Llama 3.1 8B?
8.03B params 128K max context Tool use
The smallest GPU memory that runs Llama 3.1 8B well is 8 GB (Q4_K_M, e.g. GeForce RTX 5060 Ti). 120 of 132 GPUs run it well and 8 more are a tight fit.
MMLU 69.4 SWE-bench Verified 6.2
VRAM needed per quantization
| Quantization | Weights | 16K context | 32K context | Full 128K |
|---|---|---|---|---|
| Q4_K_M | 4.9 GB | 6.9 GB | 8.8 GB | 20.5 GB |
| Q8_0 | 8.5 GB | 10.5 GB | 12.4 GB | 24.2 GB |
| fp16 | 16.1 GB | 18.0 GB | 20.0 GB | 31.7 GB |
Weights plus fp16 KV cache. Leave ~1 GB headroom for the inference engine.
Cheapest GPUs that can run Llama 3.1 8B
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 3.1 8B
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
- GTX 1060 6GB 6 GB Q4_K_M 8K context · ~32 tok/s tight fit
- GTX 1660 6 GB Q4_K_M 8K context · ~32 tok/s tight fit
- GTX 1660 SUPER 6 GB Q4_K_M 8K context · ~57 tok/s tight fit
- GTX 1660 Ti 6 GB Q4_K_M 8K context · ~49 tok/s tight fit
- RTX 2060 6 GB Q4_K_M 8K context · ~57 tok/s tight fit
- RTX 3050 6GB 6 GB Q4_K_M 8K context · ~28 tok/s tight fit
- RTX A2000 6 GB Q4_K_M 8K context · ~49 tok/s tight fit
- GTX 1070 8 GB Q4_K_M 25K context · ~43 tok/s
- GTX 1070 Ti 8 GB Q4_K_M 25K context · ~43 tok/s
- GTX 1080 8 GB Q4_K_M 25K context · ~54 tok/s
- RTX 2060 SUPER 8 GB Q4_K_M 25K context · ~76 tok/s
- RTX 2070 8 GB Q4_K_M 25K context · ~76 tok/s
- RTX 2070 SUPER 8 GB Q4_K_M 25K context · ~76 tok/s
- RTX 2080 8 GB Q4_K_M 25K context · ~76 tok/s
- RTX 2080 SUPER 8 GB Q4_K_M 25K context · ~84 tok/s
- RTX 3050 8 GB Q4_K_M 25K context · ~38 tok/s
- RTX 3060 Ti 8 GB Q4_K_M 25K context · ~76 tok/s
- RTX 3060 Ti GDDR6X 8 GB Q4_K_M 25K context · ~103 tok/s
- RTX 3070 8 GB Q4_K_M 25K context · ~76 tok/s
- RTX 3070 Ti 8 GB Q4_K_M 25K context · ~103 tok/s
- RTX 4060 8 GB Q4_K_M 25K context · ~46 tok/s
- RTX 4060 Ti 8 GB Q4_K_M 25K context · ~49 tok/s
- RTX 5050 8 GB Q4_K_M 25K context · ~54 tok/s
- RTX 5060 8 GB Q4_K_M 25K context · ~76 tok/s
- RTX 5060 Ti 8 GB Q4_K_M 25K context · ~76 tok/s
- T1000 8GB 8 GB Q4_K_M 25K context · ~27 tok/s
- RTX 3080 10 GB Q4_K_M 41K context · ~128 tok/s
- GTX 1080 Ti 11 GB Q8_0 20K context · ~51 tok/s
- RTX 2080 Ti 11 GB Q8_0 20K context · ~65 tok/s
- RTX 2060 12GB 12 GB Q8_0 28K context · ~35 tok/s
- RTX 3060 12 GB Q8_0 28K context · ~38 tok/s
- RTX 3080 12GB 12 GB Q8_0 28K context · ~96 tok/s
- RTX 3080 Ti 12 GB Q8_0 28K context · ~96 tok/s
- RTX 4070 12 GB Q8_0 28K context · ~53 tok/s
- RTX 4070 SUPER 12 GB Q8_0 28K context · ~53 tok/s
- RTX 4070 Ti 12 GB Q8_0 28K context · ~53 tok/s
- RTX 5070 12 GB Q8_0 28K context · ~70 tok/s
- RTX A2000 12GB 12 GB Q8_0 28K context · ~30 tok/s
- RTX 4060 Ti 16GB 16 GB Q8_0 61K context · ~30 tok/s
- RTX 4070 Ti SUPER 16 GB Q8_0 61K context · ~70 tok/s
- RTX 4080 16 GB Q8_0 61K context · ~75 tok/s
- RTX 4080 SUPER 16 GB Q8_0 61K context · ~77 tok/s
- RTX 5060 Ti 16GB 16 GB Q8_0 61K context · ~47 tok/s
- RTX 5070 Ti 16 GB Q8_0 61K context · ~94 tok/s
- RTX 5080 16 GB Q8_0 61K context · ~101 tok/s
- RTX 2000 Ada 16 GB Q8_0 61K context · ~23 tok/s
- RTX A4000 16 GB Q8_0 61K context · ~47 tok/s
- RTX PRO 2000 Blackwell 16 GB Q8_0 61K context · ~30 tok/s
- Tesla T4 16 GB Q8_0 61K context · ~34 tok/s
- Tesla V100 16GB 16 GB Q8_0 61K context · ~94 tok/s
- RTX 4000 Ada 20 GB fp16 32K context · ~21 tok/s
- RTX 3090 24 GB fp16 65K context · ~55 tok/s
- RTX 3090 Ti 24 GB fp16 65K context · ~59 tok/s
- RTX 4090 24 GB fp16 65K context · ~59 tok/s
- RTX 4500 Ada 24 GB fp16 65K context · ~25 tok/s
- RTX A5000 24 GB fp16 65K context · ~45 tok/s
- RTX PRO 4000 Blackwell 24 GB fp16 65K context · ~39 tok/s
- RTX Spark 24 GB fp16 65K context · ~18 tok/s
- Tesla P40 24 GB fp16 65K context · ~20 tok/s
- RTX 5090 32 GB fp16 128K context · ~105 tok/s
- RTX 5000 Ada 32 GB fp16 128K context · ~34 tok/s
- RTX PRO 4500 Blackwell 32 GB fp16 128K context · ~53 tok/s
- Tesla V100 32GB 32 GB fp16 128K context · ~53 tok/s
- L40S 48 GB fp16 128K context · ~51 tok/s
- RTX 6000 Ada 48 GB fp16 128K context · ~56 tok/s
- RTX A6000 48 GB fp16 128K context · ~45 tok/s
- RTX PRO 5000 Blackwell 48 GB fp16 128K context · ~79 tok/s
- RTX PRO 5000 Blackwell 72GB 72 GB fp16 128K context · ~79 tok/s
- RTX PRO 5500 Blackwell 84 GB fp16 128K context · ~82 tok/s
- RTX PRO 6000 Blackwell 96 GB fp16 128K context · ~105 tok/s
- DGX Spark (GB 10) 128 GB fp16 128K context · ~16 tok/s
AMD
- Radeon RX 5600 XT 6 GB Q4_K_M 8K context · ~49 tok/s tight fit
- Radeon RX 5700 8 GB Q4_K_M 25K context · ~76 tok/s
- Radeon RX 5700 XT 8 GB Q4_K_M 25K context · ~76 tok/s
- Radeon RX 6600 8 GB Q4_K_M 25K context · ~38 tok/s
- Radeon RX 6600 XT 8 GB Q4_K_M 25K context · ~43 tok/s
- Radeon RX 6650 XT 8 GB Q4_K_M 25K context · ~47 tok/s
- Radeon RX 7600 8 GB Q4_K_M 25K context · ~49 tok/s
- Radeon RX 9060 8 GB Q4_K_M 25K context · ~49 tok/s
- Radeon RX 9060 XT 8 GB Q4_K_M 25K context · ~54 tok/s
- Radeon RX 6700 XT 12 GB Q8_0 28K context · ~40 tok/s
- Radeon RX 6750 XT 12 GB Q8_0 28K context · ~45 tok/s
- Radeon RX 7700 XT 12 GB Q8_0 28K context · ~45 tok/s
- Radeon RX 9070 GRE 12 GB Q8_0 28K context · ~45 tok/s
- AMD BC-250 16 GB Q8_0 61K context · ~47 tok/s
- Radeon RX 6800 16 GB Q8_0 61K context · ~54 tok/s
- Radeon RX 6800 XT 16 GB Q8_0 61K context · ~54 tok/s
- Radeon RX 6900 XT 16 GB Q8_0 61K context · ~54 tok/s
- Radeon RX 6950 XT 16 GB Q8_0 61K context · ~60 tok/s
- Radeon RX 7600 XT 16 GB Q8_0 61K context · ~30 tok/s
- Radeon RX 7800 XT 16 GB Q8_0 61K context · ~65 tok/s
- Radeon RX 7900 GRE 16 GB Q8_0 61K context · ~60 tok/s
- Radeon RX 9060 XT 16GB 16 GB Q8_0 61K context · ~34 tok/s
- Radeon RX 9070 16 GB Q8_0 61K context · ~67 tok/s
- Radeon RX 9070 XT 16 GB Q8_0 61K context · ~67 tok/s
- Radeon RX 7900 XT 20 GB fp16 32K context · ~47 tok/s
- Radeon RX 7900 XTX 24 GB fp16 65K context · ~56 tok/s
- Radeon AI PRO R9700 32 GB fp16 128K context · ~38 tok/s
- Ryzen AI Max (Strix Halo) 32 GB fp16 128K context · ~15 tok/s
- Ryzen AI Max 400 (Gorgon Halo) 64 GB fp16 128K context · ~16 tok/s
Intel
- Arc A580 8 GB Q4_K_M 25K context · ~86 tok/s
- Arc A750 8 GB Q4_K_M 25K context · ~86 tok/s
- Arc A770 8GB 8 GB Q4_K_M 25K context · ~86 tok/s
- Arc B570 10 GB Q4_K_M 41K context · ~64 tok/s
- Arc B580 12 GB Q8_0 28K context · ~48 tok/s
- Arc A770 16GB 16 GB Q8_0 61K context · ~59 tok/s
- Arc Pro B65 32 GB fp16 128K context · ~36 tok/s
- Arc Pro B70 32 GB fp16 128K context · ~36 tok/s
Apple
- M1 8 GB Q4_K_M 25K context · ~11 tok/s
- M2 8 GB Q4_K_M 25K context · ~17 tok/s
- M3 8 GB Q4_K_M 25K context · ~17 tok/s
- M1 Pro 16 GB Q8_0 61K context · ~21 tok/s
- M2 Pro 16 GB Q8_0 61K context · ~21 tok/s
- M4 16 GB Q8_0 61K context · ~13 tok/s
- M5 16 GB Q8_0 61K context · ~16 tok/s
- M6 16 GB Q8_0 61K context · ~16 tok/s
- M3 Pro 18 GB Q8_0 77K context · ~16 tok/s
- M4 Pro 24 GB fp16 65K context · ~16 tok/s
- M5 Pro 24 GB fp16 65K context · ~18 tok/s
- M1 Max 32 GB fp16 128K context · ~23 tok/s
- M2 Max 32 GB fp16 128K context · ~23 tok/s
- M3 Max 36 GB fp16 128K context · ~18 tok/s
- M4 Max 36 GB fp16 128K context · ~24 tok/s
- M5 Max 36 GB fp16 128K context · ~27 tok/s
- M1 Ultra 64 GB fp16 128K context · ~47 tok/s
- M2 Ultra 64 GB fp16 128K context · ~47 tok/s
- M3 Ultra 96 GB fp16 128K context · ~48 tok/s
- M5 Ultra 96 GB fp16 128K context · ~70 tok/s