NVIDIA RTX PRO 4000 Blackwell
24 GB · 672 GB/s · 145 W
In short
The NVIDIA RTX PRO 4000 Blackwell has 24 GB of GDDR7 memory with 672 GB/s of bandwidth, which sets how fast it generates text. At 64k context it runs 48 of the 125 open models we track fully in memory; the most capable is Qwen 3.8 27B at Q4_K_M, at about 19 tok/s (estimated). We have no current price for the NVIDIA RTX PRO 4000 Blackwell in the US.
Updated . Speeds are estimates; how they are worked out.
Prices
Every retailer we track in the US, cheapest first. Buy links may earn us a commission.
No retailer in this market lists it right now.
What it runs
Every model in the catalogue, with the best quant for this hardware. Estimates, not measurements.
| Model | Quant | Fits | Memory | Max context | Decode @64k | Prefill | Good for |
|---|---|---|---|---|---|---|---|
| Qwen 3.8 27B Alibaba Qwen | Q4_K_M | fits | 21.5 GB | 92k161k with q8 cache | 19tok/sest. | 308 tok/s | chat, coding |
| Qwen 3.6 27B Alibaba Qwen | Q5_K_M | fits | 23.4 GB | 64k112k with q8 cache | 17tok/sest. | 308 tok/s | chat |
| Qwen 3.6 35B-A3B Alibaba Qwen | Q4_K_M | fits | 23.1 GB | 74k116k with q8 cache | 77tok/sest. | 1856 tok/s | chat, coding, agents, long documents |
| Qwen 3.5 27B Alibaba Qwen | Q4_K_M | fits | 20.8 GB | 101k178k with q8 cache | 20tok/sest. | 308 tok/s | chat, coding |
| Gemma 4 26B-A4B | UD-Q5_K_M | fits | 22.2 GB | 110k178k with q8 cache | 52tok/sest. | 1901 tok/s | chat, coding, agents, long documents |
| Qwen 3.5 35B-A3B Alibaba Qwen | Q4_K_M | fits | 22.9 GB | 84k131k with q8 cache | 77tok/sest. | 1856 tok/s | chat, coding, agents, long documents |
| GLM 4.7 Flash Z.ai | Q4_K_M | fits | 21.4 GB | 98k169k with q8 cache | 38tok/sest. | 921 tok/s | chat, coding, agents, long documents |
| Nemotron 3.5 Lightning 30B-A3B NVIDIA | MXFP4 | fits | 23.1 GB | 88k115k with q8 cache | 94tok/sest. | 2289 tok/s | chat, coding, agents, long documents |
| gpt-oss 20B OpenAI | MXFP4 | fits | 13.9 GB | 128k | 62tok/sest. | 1421 tok/s | chat, coding, agents, long documents |
| Gemma 4 12B | Q8_0 | fits | 14.2 GB | 256k | 30tok/sest. | 731 tok/s | chat, coding, long documents |
| Qwen 3.5 9B Alibaba Qwen | Q8_0 | fits | 12.0 GB | 256k | 36tok/sest. | 894 tok/s | chat, coding, long documents |
| Gemma 3 27B | Q4_K_M | fits | 21.9 GB | 81k128k with q8 cache | 19tok/sest. | 345 tok/s | chat, coding |
| Ministral 3 14B Mistral AI | Q6_K | fits | 21.4 GB | 76k139k with q8 cache | 19tok/sest. | 428 tok/s | chat, coding |
| Qwen3 14B Alibaba Qwen | Q6_K | fits | 22.4 GB | 40k | 18tok/sest. | 374 tok/s | chat, coding |
| DeepSeek R1 0528 Qwen3 8B DeepSeek | Q8_0 | fits | 18.2 GB | 99k128k with q8 cache | 23tok/sest. | 591 tok/s | chat, coding |
| Qwen3 8B Alibaba Qwen | Q8_0 | fits | 18.2 GB | 40k | 23tok/sest. | 591 tok/s | chat, coding |
| Phi-4 14B Microsoft | Q5_K_M | fits | 23.3 GB | 16k | 18tok/sest. | 376 tok/s | chat, coding |
| Qwen 3.5 4B Alibaba Qwen | Q8_0 | fits | 7.3 GB | 256k | 62tok/sest. | 1550 tok/s | chat, coding, long documents |
| Ministral 3 8B Mistral AI | Q8_0 | fits | 18.0 GB | 103k187k with q8 cache | 23tok/sest. | 585 tok/s | chat, coding |
| Granite 4.2 8B IBM | Q8_0 | fits | 19.8 GB | 86k128k with q8 cache | 21tok/sest. | 540 tok/s | chat, coding |
| Gemma 4 E4B | Q8_0 | fits | 10.5 GB | 128k | 41tok/sest. | 1171 tok/s | chat, coding, long documents |
| LFM2.5 8B-A1B Liquid AI | Q8_0 | fits | 12.5 GB | 125k | 58tok/sest. | 2500 tok/s | chat, coding, long documents |
| Llama 3.1 8B Meta | Q6_K | fits | 15.2 GB | 126k128k with q8 cache | 27tok/sest. | 635 tok/s | chat, coding, long documents |
| Qwen2.5 0.5B Alibaba Qwen | Q8_0 | fits | 2.3 GB | 32k | 312tok/sest. | 5544 tok/s | chat, coding, long documents |
| Qwen3.5 0.8B Alibaba Qwen | Q8_0 | fits | 2.6 GB | 256k | 255tok/sest. | 6286 tok/s | chat, coding, agents, long documents |
| Qwen3.5 2B Alibaba Qwen | Q5_K_M | fits | 3.2 GB | 256k | 184tok/sest. | 3427 tok/s | chat, coding, agents, long documents |
| DeepSeek R1 Distill Qwen 1.5B DeepSeek | Q8_0 | fits | 4.6 GB | 128k | 110tok/sest. | 2296 tok/s | chat, coding, long documents |
| MiniCPM5 2B openbmb | Q8_0 | fits | 6.2 GB | 128k | 76tok/sest. | 1294 tok/s | chat, coding, agents, long documents |
| Spark X2.5 4B XHToken | mlx-4bit | fits | 6.2 GB | 483k808k with q8 cache | 76tok/sest. | 1589 tok/s | chat, coding, agents, long documents |
| Mellum2.1 12B A2.5B Thinking JetBrains | Q8_0 | fits | 14.0 GB | 128k | 69tok/sest. | 2317 tok/s | chat, coding, agents, long documents |
| FrogNano 4B 2609 Microsoft | Q8_0 | fits | 7.4 GB | 256k | 61tok/sest. | 1550 tok/s | chat, coding, agents, long documents |
| LFM2.5 2.6B Liquid AI | Q8_0 | fits | 7.5 GB | 128k | 60tok/sest. | 1569 tok/s | chat, coding, agents, long documents |
| LFM2.5 VL 3B Liquid AI | Q8_0 | fits | 7.5 GB | 32k | 60tok/sest. | 1476 tok/s | chat, coding, agents, long documents |
| Kimi VL A3B Thinking 2506 Moonshot AI | Q8_0 | fits | 18.8 GB | 128k | 49tok/sest. | 1593 tok/s | chat, coding, long documents |
| Kimi VL A3B Moonshot AI | mlx-8bit | fits | 19.6 GB | 128k | 47tok/sest. | 1593 tok/s | chat, coding, agents, long documents |
| granite vision 4.1 4b IBM | Q8_0 | fits | 9.4 GB | 128k | 46tok/sest. | 986 tok/s | chat, coding, long documents |
| granite 4.2 3b IBM | Q8_0 | fits | 9.7 GB | 128k | 45tok/sest. | 1018 tok/s | chat, coding, agents, long documents |
| Ministral 3 3B Instruct 2512 Mistral AI | Q8_0 | fits | 11.0 GB | 179k256k with q8 cache | 39tok/sest. | 975 tok/s | chat, coding, agents, long documents |
| GLM 4 9B 0414 Z.ai | mlx-5bit | fits | 11.3 GB | 32k | 38tok/sest. | 524 tok/s | chat, coding, agents, long documents |
| LFM2.5 24B A2B Liquid AI | Q5_K_M | fits | 23.3 GB | 65k115k with q8 cache | 36tok/sest. | 1432 tok/s | chat, coding, agents |
| DeepSeek R1 Distill Qwen 7B DeepSeek | Q8_0 | fits | 12.1 GB | 128k | 35tok/sest. | 744 tok/s | chat, coding, long documents |
| GLM 4.6V Flash Z.ai | Q8_0 | fits | 12.9 GB | 128k | 33tok/sest. | 502 tok/s | chat, coding, agents, long documents |
| GLM 4.1V 9B Thinking Z.ai | mlx-8bit | fits | 14.5 GB | 64k | 29tok/sest. | 502 tok/s | chat, coding, long documents |
| DeepSeek R1 Distill Llama 8B DeepSeek | Q8_0 | fits | 17.0 GB | 112k128k with q8 cache | 24tok/sest. | 635 tok/s | chat, coding |
| Ministral 3 8B Reasoning 2512 Mistral AI | Q8_0 | fits | 18.0 GB | 103k187k with q8 cache | 23tok/sest. | 585 tok/s | chat, coding, agents |
| Muse Glimmer 30B meta-models | Q5_K_M | fits | 20.7 GB | 128k | 20tok/sest. | 311 tok/s | chat, coding |
| Ministral 3 14B Reasoning 2512 Mistral AI | Q6_K | fits | 21.4 GB | 76k139k with q8 cache | 19tok/sest. | 428 tok/s | chat, coding |
| DeepSeek R1 Distill Qwen 14B DeepSeek | Q5_K_M | fits | 22.9 GB | 66k122k with q8 cache | 18tok/sest. | 342 tok/s | chat, coding |
| Qwen 3.5 122B-A10B Alibaba Qwen | UD-Q2_K_XL | offload | 41.7 GB | – | 12tok/sest. | 468 tok/s | chat |
| Gemma 4 31B | Q4_K_M | offload | 23.9 GB | 57k110k with q8 cache | 15tok/sest. | 278 tok/s | chat |
| Qwen3 Coder Next Alibaba Qwen | Q4_K_M | offload | 47.8 GB | – | 14tok/sest. | 1021 tok/s | chat |
| Mistral Small 4 119B Mistral AI | UD-Q2_K_XL | offload | 39.9 GB | – | 17tok/sest. | 470 tok/s | chat, coding |
| Qwen3 Next 80B-A3B Alibaba Qwen | Q4_K_M | offload | 47.8 GB | – | 16tok/sest. | 1165 tok/s | chat, coding |
| Devstral Small 2 24B Mistral AI | Q4_K_M | offload | 24.4 GB | 57k105k with q8 cache | 13tok/sest. | 296 tok/s | chat, coding |
| Kimi Linear 48B-A3B Moonshot AI | Q4_K_M | offload | 29.6 GB | – | 42tok/sest. | 1297 tok/s | chat, coding, long documents |
| Qwen3 Coder 30B-A3B Alibaba Qwen | Q4_K_M | offload | 24.4 GB | 54k97k with q8 cache | 24tok/sest. | 648 tok/s | chat, coding, agents |
| Qwen3 32B Alibaba Qwen | Q4_K_M | offload | 35.5 GB | 16k30k with q8 cache | 3tok/sest. | 124 tok/s | chat |
| Mistral Small 3.2 24B Mistral AI | Q4_K_M | offload | 24.4 GB | 57k105k with q8 cache | 13tok/sest. | 296 tok/s | chat, coding |
| DeepSeek R1 Distill Qwen 32B DeepSeek | Q4_K_M | offload | 35.6 GB | 16k30k with q8 cache | 3tok/sest. | 153 tok/s | chat |
| Devstral Small 2507 Mistral AI | Q4_K_M | offload | 24.4 GB | 57k105k with q8 cache | 13tok/sest. | 300 tok/s | chat, coding |
| Magistral Small 2506 Mistral AI | Q4_K_M | offload | 24.4 GB | 40k | 13tok/sest. | 300 tok/s | chat, coding |
| Magistral Small 2507 Mistral AI | Q4_K_M | offload | 24.4 GB | 40k | 13tok/sest. | 300 tok/s | chat, coding |
| Magistral Small 2509 Mistral AI | Q4_K_M | offload | 24.4 GB | 57k105k with q8 cache | 13tok/sest. | 296 tok/s | chat, coding |
| Qwen3.8 27B OBLITERATED OBLITERATUS | mlx-4bit | offload | 24.5 GB | 48k84k with q8 cache | 13tok/sest. | 300 tok/s | chat |
| granite 4.2 30b IBM | Q4_K_M | offload | 33.9 GB | 23k42k with q8 cache | 4tok/sest. | 184 tok/s | chat |
| DeepSeek V4 Flash DeepSeek | UD-Q2_K_XL | won't run | 96.6 GB | – | – | – | – |
| GLM 5.3 Flash Z.ai | UD-Q2_K_XL | won't run | 103.1 GB | – | – | – | – |
| Qwen 3.8 Flash Next Alibaba Qwen | UD-Q2_K_XL | won't run | 76.1 GB | – | – | – | – |
| MiniMax M2.7 MiniMax | UD-Q2_K_XL | won't run | 86.7 GB | – | – | – | – |
| Nemotron 3 Super 120B-A12B NVIDIA | UD-Q2_K_XL | won't run | 52.6 GB | – | – | – | – |
| gpt-oss 120B OpenAI | Q4_K_M | won't run | 61.8 GB | – | – | – | – |
| GLM 4.5 Air Z.ai | Q2_K | won't run | 54.8 GB | – | – | – | – |
| Llama 3.3 70B Meta | Q4_K_M | won't run | 60.7 GB | – | – | – | – |
| DeepSeek R1 DeepSeek | Q4_K_M | won't run | 382.0 GB | – | – | – | – |
| DeepSeek R1 0528 DeepSeek | Q4_K_M | won't run | 382.5 GB | – | – | – | – |
| DeepSeek R1 Distill Llama 70B DeepSeek | Q4_K_M | won't run | 60.7 GB | – | – | – | – |
| DeepSeek R1 Zero DeepSeek | Q4_K_M | won't run | 382.0 GB | – | – | – | – |
| DeepSeek V3 DeepSeek | Q4_K_M | won't run | 382.0 GB | – | – | – | – |
| DeepSeek V3 0324 DeepSeek | Q4_K_M | won't run | 382.0 GB | – | – | – | – |
| DeepSeek V3.1 DeepSeek | Q4_K_M | won't run | 382.9 GB | – | – | – | – |
| DeepSeek V3.1 Terminus DeepSeek | Q4_K_M | won't run | 382.9 GB | – | – | – | – |
| DeepSeek V3.2 DeepSeek | Q4_K_M | won't run | 382.9 GB | – | – | – | – |
| DeepSeek V4 Flash 0731 DeepSeek | MXFP4 | won't run | 152.1 GB | – | – | – | – |
| DeepSeek V4 Flash Vision Exp DeepSeek | MXFP4 | won't run | 160.9 GB | – | – | – | – |
| Devstral 2 123B Instruct 2512 Mistral AI | Q4_K_M | won't run | 92.8 GB | – | – | – | – |
| GLM 4.5 Z.ai | Q4_K_M | won't run | 225.7 GB | – | – | – | – |
| GLM 4.5V Z.ai | Q4_K_M | won't run | 71.8 GB | – | – | – | – |
| GLM 4.6 Z.ai | Q4_K_M | won't run | 224.9 GB | – | – | – | – |
| GLM 4.6V Z.ai | Q4_K_M | won't run | 78.2 GB | – | – | – | – |
| GLM 4.7 Z.ai | Q4_K_M | won't run | 225.7 GB | – | – | – | – |
| GLM 5 Z.ai | MXFP4 | won't run | 388.9 GB | – | – | – | – |
| GLM 5.1 Z.ai | MXFP4 | won't run | 426.5 GB | – | – | – | – |
| GLM 5.2 Z.ai | UD-Q4_K_M | won't run | 440.4 GB | – | – | – | – |
| GLM 5.3 Z.ai | Q8_0 | won't run | 752.9 GB | – | – | – | – |
| Kimi Dev 72B Moonshot AI | Q6_K | won't run | 81.0 GB | – | – | – | – |
| Kimi K2 Moonshot AI | Q4_K_M | won't run | 583.5 GB | – | – | – | – |
| Kimi K2 Instruct 0905 Moonshot AI | Q4_K_M | won't run | 583.9 GB | – | – | – | – |
| Kimi K2 Thinking Moonshot AI | Q4_K_M | won't run | 583.9 GB | – | – | – | – |
| Kimi K2.5 Moonshot AI | Q4_K_M | won't run | 583.9 GB | – | – | – | – |
| MiniMax M2 MiniMax | Q4_K_M | won't run | 145.4 GB | – | – | – | – |
| MiniMax M2.1 MiniMax | Q4_K_M | won't run | 145.4 GB | – | – | – | – |
| MiniMax M2.5 MiniMax | MXFP4 | won't run | 131.8 GB | – | – | – | – |
| MiniMax M3 MiniMax | MXFP4 | won't run | 248.3 GB | – | – | – | – |
| Mistral Medium 3.5 128B Mistral AI | Q4_K_M | won't run | 92.8 GB | – | – | – | – |
Specifications
From the manufacturer's published specifications; sources are linked where we have them.
- GPU
- RTX PRO 4000 Blackwell
- Memory
- 24 GB GDDR7
- Memory bus
- 192-bit
- Memory bandwidth
- 672 GB/s
- FP16 compute
- 46.9 TFLOPS
- Software
- CUDA
- Board power
- 145 W
- PCIe generation
- Gen 5
- PCIe lanes
- x16
Compare with
Questions
- How much memory does the NVIDIA RTX PRO 4000 Blackwell have for AI models?
- The NVIDIA RTX PRO 4000 Blackwell has 24 GB of GDDR7 memory with 672 GB/s of bandwidth, which sets how fast it generates text.
- What is the best model the NVIDIA RTX PRO 4000 Blackwell can run locally?
- Qwen 3.8 27B (Q4_K_M) fits fully in its memory at 64k context and generates at about 19 tok/s, an estimate from its memory bandwidth. 48 of the 125 models we track fit at that context.
- Can the NVIDIA RTX PRO 4000 Blackwell run DeepSeek V4 Flash?
- No: even at UD-Q2_K_XL, DeepSeek V4 Flash needs about 97 GB at 64k context, more than the NVIDIA RTX PRO 4000 Blackwell and system RAM can hold.
- How much does the NVIDIA RTX PRO 4000 Blackwell cost in the US?
- We have no current price for the NVIDIA RTX PRO 4000 Blackwell in the US.
Speeds on this page are for 64k context. See the best hardware for local AI at every budget.