Home › GPUs & Macs › NVIDIA RTX PRO 6000 Blackwell vs NVIDIA RTX 5090
NVIDIA RTX PRO 6000 Blackwell vs NVIDIA RTX 5090 for local LLMs
Across 6 tests measured on both, the NVIDIA RTX PRO 6000 Blackwell is faster in 4 and generates a median 1.15× the tokens/s of the NVIDIA RTX 5090.
On Qwen3.8 27B (Q4_K_M, llama-server) the NVIDIA RTX PRO 6000 Blackwell generates 63.8 tokens/s and the NVIDIA RTX 5090 76.9 tokens/s (measured).
Same model, same quant, same test
| Model | Quant / test | RTX PRO 6000 Blackwell tok/s | RTX 5090 tok/s | Faster | Sources and run details |
|---|---|---|---|---|---|
| Qwen3.8 27B | Q4_K_M llama-server | 63.8 CUDA | 76.9 CUDA | RTX 5090 +21% | qwen38-mtp RTX PRO 6000 Blackwell · qwen38-mtp RTX 5090 |
| Llama 2 7B | Q4_0 llama-bench | 281 CUDA | 300 CUDA | RTX 5090 +7% | llama.cpp #15013 RTX PRO 6000 Blackwell · llama.cpp #15013 RTX 5090 |
| gpt-oss 20B | MXFP4 llama-bench | 287 CUDA | 283 CUDA | even | llama.cpp #15396 RTX PRO 6000 Blackwell · llama.cpp #15396 RTX 5090 |
| Llama 3.1 8B Instruct | Q4_K_M LocalScore | 85.8 llamafile | 66.3 llamafile | RTX PRO 6000 Blackwell +29% | LocalScore RTX PRO 6000 Blackwell · LocalScore RTX 5090 |
| Llama 3.2 1B Instruct | Q4_K_M LocalScore | 191 llamafile | 120 llamafile | RTX PRO 6000 Blackwell +60% | LocalScore RTX PRO 6000 Blackwell · LocalScore RTX 5090 |
| Qwen2.5 14B Instruct | Q4_K_M LocalScore | 64.0 llamafile | 41.8 llamafile | RTX PRO 6000 Blackwell +53% | LocalScore RTX PRO 6000 Blackwell · 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 PRO 6000 Blackwell | NVIDIA RTX 5090 | |
|---|---|---|
| Memory | 96 GB | 32 GB |
| Memory bandwidth | 1792 GB/s | 1792 GB/s |
All results for the NVIDIA RTX PRO 6000 Blackwell · All results for the NVIDIA RTX 5090 · More comparisons