run it locally
Run Llama 3.1 8B on RTX 3060 12GB
8B parameters on a 12 GB card. It fits at Q8_0 — here's the VRAM math at every quantization, the speed to expect, and the quality you trade.
| quantization | vram needed | fits 12gb? | tokens/sec | quality |
|---|---|---|---|---|
| FP16 (full) | 19.2 GB | ✗ no | — | Reference quality — no quantization loss. |
| Q8_0recommended | 10.2 GB | ✓ yes | ~36 | Near-lossless; rarely worth the extra space over Q6. |
| Q6_K | 7.9 GB | ✓ yes | ~47 | Virtually indistinguishable from full precision. |
| Q5_K_M | 6.6 GB | ✓ yes | ~55 | Minor loss; an excellent quality-vs-size balance. |
| Q4_K_M | 5.4 GB | ✓ yes | ~68 | Small but measurable loss; the popular default. |
| Q3_K_M | 4.1 GB | ✓ yes | ~89 | Noticeable degradation; only when you're tight on VRAM. |
Weights = params × bytes/weight, +20% for KV cache & runtime; usable VRAM is 95% of nameplate. Tokens/sec is a bandwidth ceiling (360 GB/s) — real throughput is lower with long context. Try other combinations →
Running Llama 3.1 8B on a RTX 3060 12GB
At Q8_0, Llama 3.1 8B's weights take about 10 GB, inside the RTX 3060 12GB's 11 GB of usable memory. Near-lossless; rarely worth the extra space over Q6. Generation speed is bound by memory bandwidth — the GPU reads the whole model once per token — so expect on the order of 36 tokens/sec before context overhead.
Quantization is the lever
Each step down in precision shrinks the model: FP16 needs about 19 GB, Q4_K_M about 5 GB — a 72% reduction for a small, usually acceptable quality cost. Q5_K_M and Q6_K are near-lossless if you have the headroom; drop to Q3 only when you're genuinely out of VRAM. The quantization guide covers the tradeoffs in detail.
Frequently asked questions
Can a RTX 3060 12GB run Llama 3.1 8B?
Yes — Llama 3.1 8B fits on a RTX 3060 12GB at Q8_0 (10 GB of the 11 GB usable), with roughly 36 tokens/sec. Higher-precision quants need more VRAM; the table above shows each option.
How much VRAM does Llama 3.1 8B need?
Llama 3.1 8B is a 8B-parameter model. At FP16 that's about 19 GB; at Q4_K_M (the popular default) about 5 GB, including ~20% for the KV cache and runtime. Quantization is the main lever — see the per-quant table above.
Other combinations
Llama 3.1 8B on other GPUs: RTX 4070 Ti, RTX 4080
Other models on the RTX 3060 12GB: Gemma 2 9B, Mistral 7B, Qwen2.5 7B, Phi-3 Medium 14B, Gemma 2 27B, Qwen2.5 32B
Related
- VRAM calculator — any model, quant and GPU.
- Running models locally — the hardware reality.
- All run-locally combinations →