Home › GPUs & Macs › NVIDIA RTX 3090 Ti vs NVIDIA RTX 4090
NVIDIA RTX 3090 Ti vs NVIDIA RTX 4090 for local LLMs
Across 5 tests measured on both, the two are about even (median ratio 1.04×).
On Qwen3.8 27B (Q4_K_M, llama-server) the NVIDIA RTX 3090 Ti generates 41.6 tokens/s and the NVIDIA RTX 4090 47.7 tokens/s (measured).
Same model, same quant, same test
| Model | Quant / test | RTX 3090 Ti tok/s | RTX 4090 tok/s | Faster | Sources and run details |
|---|---|---|---|---|---|
| Qwen3.8 27B | Q4_K_M llama-server | 41.6 CUDA | 47.7 | RTX 4090 +15% | qwen38-mtp RTX 3090 Ti · qwen38-mtp RTX 4090 |
| Llama 2 7B | Q4_0 llama-bench | 172 CUDA | 190 Vulkan | RTX 4090 +10% | llama.cpp #15013 RTX 3090 Ti · llama.cpp #10879 RTX 4090 |
| Llama 3.1 8B Instruct | Q4_K_M LocalScore | 110 llamafile | 105 llamafile | RTX 3090 Ti +4% | LocalScore RTX 3090 Ti · LocalScore RTX 4090 |
| Llama 3.2 1B Instruct | Q4_K_M LocalScore | 354 llamafile | 292 llamafile | RTX 3090 Ti +21% | LocalScore RTX 3090 Ti · LocalScore RTX 4090 |
| Qwen2.5 14B Instruct | Q4_K_M LocalScore | 64.2 llamafile | 49.1 llamafile | RTX 3090 Ti +31% | LocalScore RTX 3090 Ti · LocalScore RTX 4090 |
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 3090 Ti | NVIDIA RTX 4090 | |
|---|---|---|
| Memory | 24 GB | 24 GB |
| Memory bandwidth | 1008 GB/s | 1008 GB/s |
All results for the NVIDIA RTX 3090 Ti · All results for the NVIDIA RTX 4090 · More comparisons