Models · chat and writing
Llama 3.1 8B: VRAM requirements and GPUs that run it
The default 8B for years. Conversational, well supported everywhere. At Q4 with 8K context it needs about 6.2 GB, so you want a GPU with at least 8 GB of VRAM or a Mac with 8 GB of unified memory.
Memory needed by quantization
| Quant | Weights | Total at 8K | Total at 32K |
|---|---|---|---|
| Q4_K_M | 5.2 GB | 6.2 GB | 9.3 GB |
| Q5_K_M | 6.2 GB | 7.3 GB | 10 GB |
| Q8_0 | 9.3 GB | 10 GB | 13 GB |
| FP16 | 18 GB | 19 GB | 22 GB |
Totals include about 10% runtime overhead and the KV cache. Longer conversations cost real memory: going from 8K to 32K context adds about 3.1 GB at any quantization. Most people should download Q4_K_M; Q5 or Q8 only make sense if you have spare memory. Read which quantization to pick.
Which GPUs can run Llama 3.1 8B?
36 of the 40 cards we track run it fully in VRAM at 8K context. The cheapest entry point is a 8 GB card such as the RTX 3050, and the fastest is the RTX 5090 at about 140 tokens per second. 4 more can run it with part of the model in system RAM, which is noticeably slower.
| GPU | VRAM | Status | Quant | Speed |
|---|---|---|---|---|
| RTX 5090 | 32 GB | Runs well | Q8_0 | ~140 tok/s |
| RTX 4090 | 24 GB | Runs well | Q8_0 | ~77 tok/s |
| RTX 3080 | 10 GB | Runs well | Q6_K | ~75 tok/s |
| RTX 5080 | 16 GB | Runs well | Q8_0 | ~74 tok/s |
| RX 7900 XTX | 24 GB | Runs well | Q8_0 | ~74 tok/s |
| RTX 3090 / 3090 Ti | 24 GB | Runs well | Q8_0 | ~72 tok/s |
| Arc A750 | 8 GB | Runs well | Q4_K_M | ~71 tok/s |
| RTX 5070 Ti | 16 GB | Runs well | Q8_0 | ~69 tok/s |
| RTX 5060 Ti | 8 GB | Runs well | Q4_K_M | ~62 tok/s |
| RTX 5060 | 8 GB | Runs well | Q4_K_M | ~62 tok/s |
| RTX 3070 / 3070 Ti | 8 GB | Runs well | Q4_K_M | ~62 tok/s |
| RTX 3060 Ti | 8 GB | Runs well | Q4_K_M | ~62 tok/s |
| RX 7900 XT | 20 GB | Runs well | Q8_0 | ~61 tok/s |
| RTX 4080 / 4080 Super | 16 GB | Runs well | Q8_0 | ~56 tok/s |
| RTX 5070 | 12 GB | Runs well | Q8_0 | ~52 tok/s |
| RTX 4070 Ti Super | 16 GB | Runs well | Q8_0 | ~52 tok/s |
| RX 9070 XT | 16 GB | Runs well | Q8_0 | ~49 tok/s |
| GTX 1080 Ti | 11 GB | Runs well | Q6_K | ~48 tok/s |
| RX 7800 XT | 16 GB | Runs well | Q8_0 | ~48 tok/s |
| RTX 4090 Laptop | 16 GB | Runs well | Q8_0 | ~44 tok/s |
| Arc A770 | 16 GB | Runs well | Q8_0 | ~43 tok/s |
| RTX 4060 Ti | 8 GB | Runs well | Q4_K_M | ~40 tok/s |
| RX 7600 | 8 GB | Runs well | Q4_K_M | ~40 tok/s |
| RX 6800 XT | 16 GB | Runs well | Q8_0 | ~39 tok/s |
| RTX 4070 / 4070 Super | 12 GB | Runs well | Q8_0 | ~39 tok/s |
| RTX 4060 | 8 GB | Runs well | Q4_K_M | ~37 tok/s |
| RTX 4070 Laptop | 8 GB | Runs well | Q4_K_M | ~35 tok/s |
| RTX 4060 Laptop | 8 GB | Runs well | Q4_K_M | ~35 tok/s |
| Arc B580 | 12 GB | Runs well | Q8_0 | ~35 tok/s |
| RTX 5060 Ti | 16 GB | Runs well | Q8_0 | ~34 tok/s |
| RTX 4080 Laptop | 12 GB | Runs well | Q8_0 | ~33 tok/s |
| RX 7700 XT | 12 GB | Runs well | Q8_0 | ~33 tok/s |
| RTX 3050 | 8 GB | Runs well | Q4_K_M | ~31 tok/s |
| RX 6700 XT | 12 GB | Runs well | Q8_0 | ~29 tok/s |
| RTX 3060 | 12 GB | Runs well | Q8_0 | ~28 tok/s |
| RTX 4060 Ti | 16 GB | Runs well | Q8_0 | ~22 tok/s |
| GTX 1660 / 1660 Super | 6 GB | Offload | Q4_K_M | ~7.0 tok/s |
| RTX 4050 Laptop | 6 GB | Offload | Q4_K_M | ~4.4 tok/s |
| RTX 3060 Laptop | 6 GB | Offload | Q4_K_M | ~7.0 tok/s |
| GTX 1650 | 4 GB | Offload | Q4_K_M | ~4.7 tok/s |
Running it on a Mac
Apple Silicon Macs can give a model about 72% of their unified memory. Llama 3.1 8B fits from 8 GB (Apple M1, about 12 tokens per second). With a faster chip like the Apple M3 Ultra (96 GB) expect around 63 tokens per second. Compare the options in our Mac vs NVIDIA guide.
Similar size models: Qwen2.5-Coder 7B, Mistral 7B v0.3, Qwen3 8B, DeepSeek-R1 Distill 8B, Qwen3.5 4B.
Compare Llama 3.1 8B
Questions
How much VRAM does Llama 3.1 8B need?
About 6.2 GB at Q4_K_M with 8K context, 10 GB at Q8_0, and 19 GB at full FP16. In practice that means a card with at least 8 GB of VRAM for Q4.
Can Llama 3.1 8B run on an 8 GB GPU?
Yes, at Q4_K_M, using about 6.2 GB.
Can I run Llama 3.1 8B on a Mac?
Yes. The smallest configuration that runs it fully is Apple M1 with 8 GB of unified memory, at about 12 tokens per second. More memory bandwidth (Pro, Max, Ultra chips) means faster replies.
Estimates based on model size, architecture and memory bandwidth. Real numbers vary with software, drivers and context length.