Llama · Dense · Llama 3.2 Community
Can I run Llama 3.2 3B locally?
Pick this if you want a snappy assistant on a laptop or 8 GB card and can live with the odd mistake.
- Parameters
- 3.2B
- VRAM at Q4
- 3.5 GB
- Max context
- 131K
- Released
- 2024-09
Memory needed by quantisation
Total includes weights, an 8K-token KV cache and runtime overhead. Lower quants trade quality for size.
| Quant | Bits/weight | Weights | Total needed | Quality |
|---|---|---|---|---|
| FP16 | 16 | 6.0 GB | 7.8 GB | Full precision. Reference quality, 2× the size of 8-bit. |
| Q8_0 | 8.5 | 3.2 GB | 4.9 GB | Effectively lossless. The safe choice when it fits. |
| Q6_K | 6.56 | 2.5 GB | 4.2 GB | Near-lossless; quality loss is hard to measure. |
| Q5_K_M | 5.67 | 2.1 GB | 3.8 GB | Very good. A common sweet spot above Q4. |
| Q4_K_M | 4.83 | 1.8 GB | 3.5 GB | The default. Best size/quality trade-off for local use. |
| Q3_K_M | 3.91 | 1.5 GB | 3.1 GB | Noticeable degradation; useful to squeeze a size up. |
| Q2_K | 3.35 | 1.3 GB | 2.9 GB | Aggressive. Quality drops a lot — last resort to fit. |
Which hardware runs Llama 3.2 3B?
Best quantisation that fits each device at 8K context, with a rough speed estimate. Try your exact setup →
| Hardware | Memory | Fit | Best quant | Speed |
|---|---|---|---|---|
| RTX 3060 12 GB | 12 GB | Runs well | FP16 | ~40 tok/s |
| RTX 4060 Ti 8 GB | 8 GB | Runs well | Q8 | ~61 tok/s |
| RTX 4060 Ti 16 GB | 16 GB | Runs well | FP16 | ~32 tok/s |
| RTX 3080 10 GB | 10 GB | Runs well | FP16 | ~85 tok/s |
| RTX 5070 | 12 GB | Runs well | FP16 | ~75 tok/s |
| RTX 4070 Super | 12 GB | Runs well | FP16 | ~57 tok/s |
| Radeon RX 7900 XT | 20 GB | Runs well | FP16 | ~90 tok/s |
| RTX 5070 Ti | 16 GB | Runs well | FP16 | ~100 tok/s |
| RTX 4070 Ti Super | 16 GB | Runs well | FP16 | ~75 tok/s |
| RTX 3090 | 24 GB | Runs well | FP16 | ~105 tok/s |
| Radeon RX 7900 XTX | 24 GB | Runs well | FP16 | ~108 tok/s |
| RTX 5080 | 16 GB | Runs well | FP16 | ~108 tok/s |
| RTX 4080 Super | 16 GB | Runs well | FP16 | ~83 tok/s |
| RTX 4090 | 24 GB | Runs well | FP16 | ~113 tok/s |
| RTX 5090 | 32 GB | Runs well | FP16 | ~201 tok/s |
| RTX A6000 | 48 GB | Runs well | FP16 | ~86 tok/s |
| RTX 6000 Ada | 48 GB | Runs well | FP16 | ~108 tok/s |
| A100 80 GB | 80 GB | Runs well | FP16 | ~229 tok/s |
| H100 80 GB | 80 GB | Runs well | FP16 | ~376 tok/s |
| Mac · M1/M2/M3 (base), 8 GB | 8 GB | Runs well | Q8 | ~21 tok/s |
| Mac · M1/M2/M3 (base), 16 GB | 16 GB | Runs well | FP16 | ~11 tok/s |
| Mac · M4 (base), 24 GB | 24 GB | Runs well | FP16 | ~13 tok/s |
| Mac · M4 Pro, 48 GB | 48 GB | Runs well | FP16 | ~31 tok/s |
| Mac · M1/M2/M3 Max, 32 GB | 32 GB | Runs well | FP16 | ~45 tok/s |
| Mac · M4 Max, 64 GB | 64 GB | Runs well | FP16 | ~61 tok/s |
| Mac · M1/M2/M3 Max, 64 GB | 64 GB | Runs well | FP16 | ~45 tok/s |
| Mac · M3/M4 Max, 128 GB | 128 GB | Runs well | FP16 | ~61 tok/s |
| Mac Studio · M1/M2 Ultra, 128 GB | 128 GB | Runs well | FP16 | ~90 tok/s |
| Mac Studio · M3 Ultra, 256 GB | 256 GB | Runs well | FP16 | ~92 tok/s |
| Mac Studio · M3 Ultra, 512 GB | 512 GB | Runs well | FP16 | ~92 tok/s |
| CPU only · 8 GB RAM | 8 GB | Runs well | Q8 | ~11 tok/s |
| CPU only · 16 GB RAM | 16 GB | Runs well | FP16 | ~6.7 tok/s |
| CPU only · 32 GB RAM | 32 GB | Runs well | FP16 | ~7.9 tok/s |
| CPU only · 64 GB RAM | 64 GB | Runs well | FP16 | ~9.0 tok/s |
| CPU only · 128 GB RAM | 128 GB | Runs well | FP16 | ~10 tok/s |
GPUs that run Llama 3.2 3B well
The most affordable cards in our list that run it at a good quantisation.
RTX 3060 12 GB
12 GB · ~$279
Check price →RTX 4060 Ti 8 GB
8 GB · ~$379
Check price →RTX 4060 Ti 16 GB
16 GB · ~$449
Check price →Hardware links are affiliate links — they don't change the recommendation.
What Llama 3.2 3B is good for
Related models
FAQ
How much VRAM does Llama 3.2 3B need?
At Q4_K_M, Llama 3.2 3B needs about 3.5 GB including a 8K-token context and overhead (1.8 GB for the weights alone). Higher quantisation needs more; see the table for every level.
What is the cheapest way to run Llama 3.2 3B?
The smallest device that runs it well in our list is the RTX 4060 Ti 8 GB (8 GB). Anything with at least that much memory should handle it at a usable quantisation.
Is Llama 3.2 3B good for low-end & edge hardware?
Pick this if you want a snappy assistant on a laptop or 8 GB card and can live with the odd mistake.
Estimates — see how we compute these. Memory figures assume an 8K context; long-context use needs more.