NVIDIA · workstation

RTX PRO 6000 Blackwell Max-Q Workstation Edition

RTX PRO 6000 Blackwell Max-Q Workstation Edition has 96 GB of VRAM at 1792 GB/s — about 89.28 GiB usable after driver and compositor overhead. 2086 of 2118 indexed models fit at 64K context with q4_0 KV.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed
Memory
96 GB
GDDR7
Bandwidth
1792 GB/s
512-bit bus
Tensor FP16
441 TF
dense
TDP
300 W
$8565 MSRP
Context4K8K16K32K64K128K
KV cachef16q8_0q4_0quantizing the KV cache is a ~2× lever on the dominant term at long context
text 1792vision language 190audio tts 21image 2audio asr 39video 16embedding 26

What fits at 64K context

largest quantization that fits, per model · 2086 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
MiniMax-M2.1MoEIQ3_XXS229B83.91 GiB4.36 GiB89.26 GiB0.02 GiB48±37%
MiniMax-M2MoEIQ3_XXS229B83.91 GiB4.36 GiB89.26 GiB0.02 GiB48±37%
command-a-plus-05-2026-bf16MoEIQ3_XXS219B87.48 GiB0.68 GiB89.16 GiB0.12 GiB52±37%
gpt-oss-120b-Uncensored-xCloudMoEI1-Q5_K_M117B87.49 GiB0.64 GiB89.12 GiB0.16 GiB67±37%
gpt-oss-120b-abliteratedMoEI1-Q5_K_M117B87.49 GiB0.64 GiB89.12 GiB0.16 GiB67±37%
Gemma-4-Dark-Gemistry-31BQ6_K32.7B84.44 GiB3.14 GiB88.66 GiB0.62 GiB12±22%
Qwen3.5-122B-A10BMoEUD-Q5_K_M125B87.21 GiB0.42 GiB88.66 GiB0.62 GiB69±37%
Qwen3-235B-A22B-abliteratedMoEI1-IQ3_XXS235B84.17 GiB3.30 GiB88.51 GiB0.77 GiB43±37%
grok-2MoEUD-IQ1_S270B82.82 GiB4.50 GiB88.47 GiB0.81 GiB20±37%
MiMo-V2-FlashMoEKV unresolvedI1-IQ2_S310B85.29 GiB2.11 GiB88.45 GiB0.83 GiB58±37%
Mistral-Medium-3.5-128BQ5_K_S128B81.09 GiB6.19 GiB88.43 GiB0.85 GiB12±22%
GLM-4.5-Air-REAP-82B-A12BMoEQ8_081.9B84.17 GiB3.23 GiB88.43 GiB0.85 GiB39±37%
step-3.5-flashQ3_K_S199B80.06 GiB7.04 GiB88.13 GiB1.15 GiB12±22%
Qwen3.5-REAP-212B-A17BMoEIQ3_M212B86.49 GiB0.53 GiB88.07 GiB1.21 GiB65±37%
Behemoth-X-123B-v2Q5_K_M123B80.55 GiB6.19 GiB87.89 GiB1.39 GiB12±22%
Mistral-Large-Instruct-2411Q5_K_M123B80.55 GiB6.19 GiB87.89 GiB1.39 GiB12±22%
Codestral-22B-v0.1F3222.2B82.88 GiB3.94 GiB87.88 GiB1.40 GiB12±22%
Step-3.5-Flash-REAP-121B-A11BI1-Q5_K_M121B79.79 GiB7.04 GiB87.86 GiB1.42 GiB12±22%
Mixtral-8x22B-v0.1MoEF32141B82.84 GiB3.94 GiB87.84 GiB1.44 GiB21±37%
Devstral-2-123B-Instruct-2512Q5_K_S125B80.27 GiB6.19 GiB87.61 GiB1.67 GiB12±22%
XORTRON-NXTXPRTXXLI1-Q5_K_S128B80.27 GiB6.19 GiB87.61 GiB1.67 GiB12±22%
Llama-4-Scout-17B-16E-InstructMoEKV unresolvedQ6_K_L109B83.13 GiB3.38 GiB87.53 GiB1.75 GiB43±37%
MiniMax-M3MoEIQ1_S427B84.31 GiB2.11 GiB87.43 GiB1.85 GiB59±37%
MiniMax-M2.7-BF16-ultra-uncensored-hereticMoEI1-IQ3_XXS229B81.91 GiB4.36 GiB87.26 GiB2.02 GiB49±37%
MiniMax-M2.5MoEI1-IQ3_XXS229B81.91 GiB4.36 GiB87.26 GiB2.02 GiB49±37%
dots.llm1.instMoEQ3_K_M143B68.74 GiB17.44 GiB87.20 GiB2.08 GiB24±37%
Trinity-Large-ThinkingMoEIQ1_M399B84.63 GiB1.28 GiB86.93 GiB2.35 GiB74±37%
DeepSeek-Coder-V2-Instruct-0724MoEIQ3_XXS236B84.61 GiB1.19 GiB86.83 GiB2.45 GiB61±37%
DeepSeek-V2.5MoEIQ3_XXS236B84.61 GiB1.19 GiB86.83 GiB2.45 GiB61±37%
DeepSeek-Coder-V2-InstructMoEIQ3_XXS236B84.61 GiB1.19 GiB86.83 GiB2.45 GiB61±37%
Ornith-1.0-397BMoEIQ1_M397B85.09 GiB0.53 GiB86.67 GiB2.61 GiB78±37%
MiniMax-M2.1-REAP-139B-A10BMoEI1-Q4_1139B81.21 GiB4.36 GiB86.55 GiB2.73 GiB44±37%
m51Lab-MiniMax-M2.7-REAP-139B-A10BMoEI1-Q4_1139B81.21 GiB4.36 GiB86.55 GiB2.73 GiB44±37%
Step-3.7-FlashIQ3_XXS201B78.38 GiB7.04 GiB86.45 GiB2.83 GiB12±22%
GLM-4.7MoEUD-TQ1_0358B78.69 GiB6.47 GiB86.20 GiB3.08 GiB39±37%
NVIDIA-Nemotron-3-Super-120B-A12B-BF16MoEUD-Q5_K_S124B83.56 GiB1.55 GiB86.10 GiB3.18 GiB57±37%
GLM-4.5MoEUD-TQ1_0358B78.54 GiB6.47 GiB86.05 GiB3.23 GiB39±37%
Llama-3_1-Nemotron-51B-InstructQ6_K_L51.5B39.83 GiB45.00 GiB85.97 GiB3.31 GiB12±22%
GLM-4.6MoEUD-TQ1_0357B78.36 GiB6.47 GiB85.87 GiB3.41 GiB39±37%
DeepSeek-V4-Flash-0731MoEUD-IQ2_M304B84.68 GiB0.02 GiB85.75 GiB3.53 GiB81±37%
DeepSeek-V4-FlashMoEUD-IQ2_M291B84.68 GiB0.02 GiB85.75 GiB3.53 GiB81±37%
ERNIE-4.5-300B-A47B-PTUD-IQ1_S300B80.54 GiB3.80 GiB85.47 GiB3.81 GiB12±22%
c4ai-command-r-plus-08-2024Q6_K104B79.32 GiB4.50 GiB85.00 GiB4.28 GiB12±22%
Llama-3_3-Nemotron-Super-49B-v1_5Q6_K_L49.9B38.58 GiB45.00 GiB84.72 GiB4.56 GiB12±22%
Valkyrie-49B-v2.1Q6_K_L49.9B38.58 GiB45.00 GiB84.72 GiB4.56 GiB12±22%
Mixtral-8x22B-Instruct-v0.1MoEQ4_K_M141B79.71 GiB3.94 GiB84.71 GiB4.57 GiB21±37%
Mixtral-8x22B-v0.1MoEQ4_K_M141B79.71 GiB3.94 GiB84.71 GiB4.57 GiB21±37%
Mistral-Small-4-119B-2603MoEUD-Q5_K_M119B83.04 GiB0.40 GiB84.46 GiB4.82 GiB72±37%
Trinity-Large-TrueBaseMoEI1-IQ1_M399B82.07 GiB1.28 GiB84.38 GiB4.90 GiB76±37%
CalmeRys-78B-Orpo-v0.1Q8_078.0B77.16 GiB6.05 GiB84.34 GiB4.94 GiB12±22%
calme-2.3-rys-78bQ8_078.0B77.16 GiB6.05 GiB84.34 GiB4.94 GiB12±22%
GLM-4.6-REAP-268B-A32BMoEUD-IQ1_M269B76.78 GiB6.47 GiB84.29 GiB4.99 GiB36±37%
Qwen3-VL-235B-A22B-ThinkingMoEQ2_K_L236B79.94 GiB3.30 GiB84.28 GiB5.00 GiB45±37%
Qwen3-VL-235B-A22B-InstructMoEQ2_K_L236B79.94 GiB3.30 GiB84.28 GiB5.00 GiB45±37%
Qwen3-235B-A22BMoEQ2_K_L235B79.94 GiB3.30 GiB84.28 GiB5.00 GiB45±37%
Qwen3-235B-A22B-Instruct-2507MoEQ2_K_L235B79.94 GiB3.30 GiB84.28 GiB5.00 GiB45±37%
Qwen3-235B-A22B-Thinking-2507MoEQ2_K_L235B79.94 GiB3.30 GiB84.28 GiB5.00 GiB45±37%
Llama-3_3-Nemotron-Super-49B-v1Q6_K49.9B38.11 GiB45.00 GiB84.25 GiB5.03 GiB12±22%
Qwen3.5-397B-A17BMoEIQ1_S403B82.64 GiB0.53 GiB84.22 GiB5.06 GiB80±37%
OYM-Qimi-122B-A10B-K2.6MoEI1-Q5_K_M125B82.62 GiB0.42 GiB84.07 GiB5.21 GiB72±37%
●From the file◐Predictedwhat these mean

Speed is modeled, not measured: decode is memory-bandwidth bound, so tokens per second is bytes read per token against achievable bandwidth. Mixture-of-experts models carry a wider band because only the routed experts are read each step, and few have been measured publicly.

Measured on this card

third-party benchmarks, aggregated
Workload◍MedianMiddle 50%Runs
Prompt processing7623.24 tok/s5788.82–12034.7618
Text generation269.96 tok/s249.96–271.269
Benchmarked· n=18

Aggregated from community-submitted runs, so the spread is wide by nature — it covers different models, resolutions, step counts and settings, not one controlled configuration. Read the middle 50% rather than the median alone. These figures are reproduced with attribution from llama.cpp-discussion-15013.

Questions people ask

What AI models can a RTX PRO 6000 Blackwell Max-Q Workstation Edition run?
2086 of 2118 indexed open-weight models fit a RTX PRO 6000 Blackwell Max-Q Workstation Edition at 65,536 context with q4_0 KV cache, the largest being MiniMax-M2.1 at IQ3_XXS. That covers text, vision-language, image, video and speech models.
How much usable memory does a RTX PRO 6000 Blackwell Max-Q Workstation Edition actually have?
Its nameplate is 96 GB, but about 89.28 GiB is available to a model once driver and compositor overhead is accounted for.
Is a RTX PRO 6000 Blackwell Max-Q Workstation Edition fast for local AI?
Its memory bandwidth is 1792 GB/s, and that figure — not teraflops — is what governs token generation speed. Capacity decides what you can run; bandwidth decides how fast it runs.