NVIDIA RTX PRO 6000 Blackwell vs NVIDIA GeForce RTX 4090
For running models locally, priced in India. At 32k context the first runs 41 models fully in memory, the second 26.
Specifications
| Spec | NVIDIA RTX PRO 6000 Blackwell | NVIDIA GeForce RTX 4090 |
|---|---|---|
| GPU | RTX PRO 6000 Blackwell | GeForce RTX 4090 |
| Memory | 96 GB GDDR7 | 24 GB GDDR6X |
| Memory bus | 512-bit | 384-bit |
| Memory bandwidth | 1,792 GB/s | 1,008 GB/s |
| FP16 compute | 504 TFLOPS | 165 TFLOPS |
| Software | CUDA | CUDA |
| Board power | 600 W | 450 W |
| PCIe generation | Gen 5 | Gen 4 |
| PCIe lanes | x16 | x16 |
What each runs
Estimated decode speed with the most precise quant that fits each one est.
| Model | NVIDIA RTX PRO 6000 Blackwell | NVIDIA GeForce RTX 4090 |
|---|---|---|
| DeepSeek V4 Flash | 69tok/sUD-Q2_K_XL | won't run |
| Qwen 3.8 Flash Next | 21tok/sUD-Q3_K_XL | won't run |
| MiniMax M2.7 | 56tok/sUD-Q2_K_XL | won't run |
| Qwen 3.8 27B | 35tok/sQ8_0 | 28tok/sUD-Q5_K_M |
| Qwen 3.5 122B-A10B | 81tok/sQ5_K_M | offload |
| Qwen 3.6 27B | 36tok/sQ8_0 | 29tok/sQ5_K_M |
| gpt-oss 120B | 154tok/sMXFP4 | won't run |
| Qwen3 Coder Next | 133tok/sQ8_0 | offload |
| Nemotron 3 Super 120B-A12B | 79tok/sUD-Q4_K_M | won't run |
| Gemma 4 31B | 31tok/sQ8_0 | 29tok/sQ4_K_M |
| Mistral Small 4 119B | 101tok/sUD-Q6_K | offload |
| Qwen 3.6 35B-A3B | 175tok/sQ8_0 | 145tok/sUD-Q4_K_M |
| Qwen 3.5 27B | 36tok/sQ8_0 | 29tok/sQ5_K_M |
| Gemma 4 26B-A4B | 132tok/sQ8_0 | 84tok/sUD-Q6_K |
| GLM 4.5 Air | 35tok/sQ5_K_M | offload |
| Qwen 3.5 35B-A3B | 175tok/sQ8_0 | 146tok/sQ4_K_M |
| Qwen3 Next 80B-A3B | 166tok/sQ8_0 | offload |
| GLM 4.7 Flash | 95tok/sQ8_0 | 70tok/sQ5_K_M |
| Devstral Small 2 24B | 37tok/sQ8_0 | 28tok/sQ5_K_M |
| Nemotron 3.5 Lightning 30B-A3B | 187tok/sUD-Q6_K_XL | offload |
| gpt-oss 20B | 206tok/sMXFP4 | 116tok/sMXFP4 |
| Kimi Linear 48B-A3B | 191tok/sQ8_0 | offload |
| Qwen3 Coder 30B-A3B | 104tok/sQ8_0 | 74tok/sQ4_K_M |
| Qwen3 32B | 26tok/sQ8_0 | offload |
| Gemma 4 12B | 82tok/sQ8_0 | 46tok/sQ8_0 |
| Qwen 3.5 9B | 104tok/sQ8_0 | 59tok/sQ8_0 |
| Mistral Small 3.2 24B | 37tok/sQ8_0 | 28tok/sQ5_K_M |
| Gemma 3 27B | 35tok/sQ8_0 | 28tok/sQ5_K_M |
| DeepSeek R1 Distill Qwen 32B | 26tok/sQ8_0 | offload |
| Llama 3.3 70B | 13tok/sQ8_0 | won't run |
| Ministral 3 14B | 56tok/sQ8_0 | 32tok/sQ8_0 |
| Qwen3 14B | 53tok/sQ8_0 | 30tok/sQ8_0 |
| DeepSeek R1 0528 Qwen3 8B | 82tok/sQ8_0 | 46tok/sQ8_0 |
| Qwen3 8B | 82tok/sQ8_0 | 46tok/sQ8_0 |
| Phi-4 14B | 50tok/sQ8_0 | 28tok/sQ8_0 |
| Qwen 3.5 4B | 198tok/sQ8_0 | 112tok/sQ8_0 |
| Ministral 3 8B | 82tok/sQ8_0 | 46tok/sQ8_0 |
| Granite 4.2 8B | 76tok/sQ8_0 | 43tok/sQ8_0 |
| Gemma 4 E4B | 121tok/sQ8_0 | 68tok/sQ8_0 |
| LFM2.5 8B-A1B | 249tok/sQ8_0 | 140tok/sQ8_0 |
| Llama 3.1 8B | 102tok/sQ6_K | 57tok/sQ6_K |
| GLM 5.3 Flash | offload | won't run |