Can I run Llama 3.1 70B on NVIDIA RTX 3060 12GB?
No: Llama 3.1 70B does not fit in NVIDIA RTX 3060 12GB memory at any common quantisation, even with offload. Try a smaller model.
Memory by quantisation (8,192-token context)
| Quantisation | Memory | Verdict | Est. speed |
|---|---|---|---|
| F16Unquantized reference | 136.0 GiB | Does not fit | — |
| Q8_0Near-lossless | 74.4 GiB | Does not fit | — |
| Q6_KVery close to F16 | 58.5 GiB | Does not fit | — |
| Q5_K_MGreat quality/size | 51.3 GiB | Does not fit | — |
| Q4_K_MBest default for most | 44.4 GiB | Does not fit | — |
| IQ4_XSSmaller 4-bit (imatrix) | 39.5 GiB | Does not fit | — |
| Q3_K_MNoticeable loss | 36.7 GiB | Does not fit | — |
| Q2_KLast resort; big loss | 32.1 GiB | Does not fit | — |
NVIDIA RTX 3060 12GB offers about 12 GiB of memory (92% usable for a model) at 360 GB/s. Speed is a bandwidth-bound estimate of single-stream generation at 65% efficiency; real results vary by runtime, drivers and settings, so measure on your machine.
How this is calculated
Weights ≈ parameters × bits per weight ÷ 8. KV cache ≈ 2 × 80 layers × 8 KV heads × 128 head dim × 2 bytes per token. Decode speed ≈ efficiency × memory bandwidth ÷ bytes read per token. Learn the maths in VRAM math and memory bandwidth, or try the VRAM calculator.
Llama 3.1 70B on other hardware
Similar models on NVIDIA RTX 3060 12GB
Running slowly or out of memory? See CUDA out of memory and the quantisation lesson.