Home › GPUs & Macs › NVIDIA RTX 3060 vs NVIDIA RTX 4000 Ada
NVIDIA RTX 3060 vs NVIDIA RTX 4000 Ada for local LLMs
Across 4 tests measured on both, the NVIDIA RTX 4000 Ada is faster in 3 and generates a median 1.13× the tokens/s of the NVIDIA RTX 3060.
On Llama 2 7B (Q4_0, llama-bench) the NVIDIA RTX 3060 generates 80.6 tokens/s and the NVIDIA RTX 4000 Ada 61.8 tokens/s (measured).
Same model, same quant, same test
| Model | Quant / test | RTX 3060 tok/s | RTX 4000 Ada tok/s | Faster | Sources and run details |
|---|---|---|---|---|---|
| Llama 2 7B | Q4_0 llama-bench | 80.6 Vulkan | 61.8 CUDA | RTX 3060 +30% | llama.cpp #10879 RTX 3060 · llama.cpp #15013 RTX 4000 Ada |
| Llama 3.1 8B Instruct | Q4_K_M LocalScore | 44.1 llamafile | 53.5 llamafile | RTX 4000 Ada +21% | LocalScore RTX 3060 · LocalScore RTX 4000 Ada |
| Llama 3.2 1B Instruct | Q4_K_M LocalScore | 185 llamafile | 210 llamafile | RTX 4000 Ada +14% | LocalScore RTX 3060 · LocalScore RTX 4000 Ada |
| Qwen2.5 14B Instruct | Q4_K_M LocalScore | 26.6 llamafile | 30.0 llamafile | RTX 4000 Ada +13% | LocalScore RTX 3060 · LocalScore RTX 4000 Ada |
Generation speed in tokens/s, best measured backend for each chip (shown under the number). Backends can differ between the two chips; check them before drawing conclusions. Measured results only, nothing estimated.
Specs
| NVIDIA RTX 3060 | NVIDIA RTX 4000 Ada | |
|---|---|---|
| Memory | 12 GB | 20 GB |
| Memory bandwidth | 360 GB/s | 360 GB/s |
All results for the NVIDIA RTX 3060 · All results for the NVIDIA RTX 4000 Ada · More comparisons