Home › GPUs & Macs › NVIDIA RTX 4090 vs NVIDIA RTX 5090
NVIDIA RTX 4090 vs NVIDIA RTX 5090 for local LLMs
Across 7 tests measured on both, the NVIDIA RTX 5090 is faster in 4 and generates a median 1.11× the tokens/s of the NVIDIA RTX 4090.
On Qwen3.8 27B (Q4_K_M, llama-server) the NVIDIA RTX 4090 generates 47.7 tokens/s and the NVIDIA RTX 5090 76.9 tokens/s (measured).
Same model, same quant, same test
| Model | Quant / test | RTX 4090 tok/s | RTX 5090 tok/s | Faster | Sources and run details |
|---|---|---|---|---|---|
| Qwen3.8 27B | Q4_K_M llama-server | 47.7 | 76.9 CUDA | RTX 5090 +61% | qwen38-mtp RTX 4090 · qwen38-mtp RTX 5090 |
| Qwen3.8 27B | UD-Q4_K_XL llama-server | 40.4 CUDA | 74.3 CUDA | RTX 5090 +84% | qwen38-mtp RTX 4090 · qwen38-mtp RTX 5090 |
| Llama 2 7B | Q4_0 llama-bench | 190 Vulkan | 300 CUDA | RTX 5090 +58% | llama.cpp #10879 RTX 4090 · llama.cpp #15013 RTX 5090 |
| gpt-oss 20B | MXFP4 llama-bench | 255 CUDA | 283 CUDA | RTX 5090 +11% | llama.cpp #15396 RTX 4090 · llama.cpp #15396 RTX 5090 |
| Llama 3.1 8B Instruct | Q4_K_M LocalScore | 105 llamafile | 66.3 llamafile | RTX 4090 +59% | LocalScore RTX 4090 · LocalScore RTX 5090 |
| Llama 3.2 1B Instruct | Q4_K_M LocalScore | 292 llamafile | 120 llamafile | RTX 4090 +144% | LocalScore RTX 4090 · LocalScore RTX 5090 |
| Qwen2.5 14B Instruct | Q4_K_M LocalScore | 49.1 llamafile | 41.8 llamafile | RTX 4090 +18% | LocalScore RTX 4090 · LocalScore RTX 5090 |
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 4090 | NVIDIA RTX 5090 | |
|---|---|---|
| Memory | 24 GB | 32 GB |
| Memory bandwidth | 1008 GB/s | 1792 GB/s |
All results for the NVIDIA RTX 4090 · All results for the NVIDIA RTX 5090 · More comparisons