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. 2087 of 2118 indexed models fit at 8K 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
vision language 190text 1793audio tts 21image 2audio asr 39video 16embedding 26

What fits at 8K context

largest quantization that fits, per model · 2087 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-MTPQ6_K27.8B88.00 GiB0.14 GiB89.21 GiB0.07 GiB12±22%
GLM-4.6-REAP-268B-A32BMoEUD-IQ2_M269B87.26 GiB0.81 GiB89.11 GiB0.17 GiB50±37%
MiniMax-M2.1MoEIQ3_XS229B87.32 GiB0.54 GiB88.85 GiB0.43 GiB68±37%
MiniMax-M2MoEIQ3_XS229B87.32 GiB0.54 GiB88.85 GiB0.43 GiB68±37%
MiniMax-M2.7MoEUD-Q3_K_S229B87.21 GiB0.54 GiB88.74 GiB0.54 GiB68±37%
command-a-plus-05-2026-bf16MoEIQ3_XXS219B87.48 GiB0.19 GiB88.67 GiB0.61 GiB54±37%
MiMo-V2-FlashMoEKV unresolvedUD-IQ1_M310B87.36 GiB0.26 GiB88.67 GiB0.61 GiB70±37%
gpt-oss-120b-Uncensored-xCloudMoEI1-Q5_K_M117B87.49 GiB0.09 GiB88.57 GiB0.71 GiB71±37%
gpt-oss-120b-abliteratedMoEI1-Q5_K_M117B87.49 GiB0.09 GiB88.57 GiB0.71 GiB71±37%
MiniMax-M2.7-BF16-ultra-uncensored-hereticMoEI1-IQ3_XS229B86.92 GiB0.54 GiB88.45 GiB0.83 GiB68±37%
MiniMax-M2.5MoEI1-IQ3_XS229B86.92 GiB0.54 GiB88.45 GiB0.83 GiB68±37%
ERNIE-4.5-300B-A47B-PTUD-IQ1_M300B86.78 GiB0.47 GiB88.38 GiB0.90 GiB12±22%
xLAM-8x7b-rMoEBF1646.7B86.99 GiB0.28 GiB88.31 GiB0.97 GiB22±37%
Qwen3.5-122B-A10BMoEUD-Q5_K_M125B87.21 GiB0.05 GiB88.29 GiB0.99 GiB72±37%
Hy3MoEIQ2_S299B86.43 GiB0.70 GiB88.17 GiB1.11 GiB61±37%
Qwen3-235B-A22B-Thinking-2507MoEIQ3_XXS235B86.68 GiB0.41 GiB88.13 GiB1.15 GiB53±37%
Qwen3-235B-A22B-Instruct-2507MoEIQ3_XXS235B86.68 GiB0.41 GiB88.13 GiB1.15 GiB53±37%
MiMo-V2.5MoEKV unresolvedIQ2_XS311B86.68 GiB0.26 GiB87.99 GiB1.29 GiB71±37%
Mistral-Medium-3.5-128BQ5_K_L128B85.78 GiB0.77 GiB87.71 GiB1.57 GiB12±22%
Qwen3.5-REAP-212B-A17BMoEIQ3_M212B86.49 GiB0.07 GiB87.61 GiB1.67 GiB68±37%
Step-3.7-FlashIQ3_XS201B85.44 GiB1.13 GiB87.60 GiB1.68 GiB12±22%
dots.llm1.instMoEQ4_1143B84.24 GiB2.18 GiB87.44 GiB1.84 GiB52±37%
GLM-4.7-REAP-218B-A32BMoEIQ3_XS218B85.49 GiB0.81 GiB87.34 GiB1.94 GiB45±37%
Ornith-1.0-397BMoEIQ1_M397B85.09 GiB0.07 GiB86.20 GiB3.08 GiB83±37%
Gemma-4-Dark-Gemistry-31BQ6_K32.7B84.44 GiB0.68 GiB86.20 GiB3.08 GiB12±22%
step-3.5-flashQ3_K_M199B84.04 GiB1.13 GiB86.20 GiB3.08 GiB12±22%
Trinity-Large-ThinkingMoEIQ1_M399B84.63 GiB0.35 GiB86.01 GiB3.27 GiB84±37%
DeepSeek-Coder-V2-Instruct-0724MoEIQ3_XXS236B84.61 GiB0.15 GiB85.79 GiB3.49 GiB69±37%
DeepSeek-V2.5MoEIQ3_XXS236B84.61 GiB0.15 GiB85.79 GiB3.49 GiB69±37%
DeepSeek-Coder-V2-InstructMoEIQ3_XXS236B84.61 GiB0.15 GiB85.79 GiB3.49 GiB69±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%
Qwen3-235B-A22B-abliteratedMoEI1-IQ3_XXS235B84.17 GiB0.41 GiB85.62 GiB3.66 GiB55±37%
GLM-4.5-Air-REAP-82B-A12BMoEQ8_081.9B84.17 GiB0.40 GiB85.60 GiB3.68 GiB48±37%
MiniMax-M3MoEIQ1_S427B84.31 GiB0.26 GiB85.59 GiB3.69 GiB72±37%
GLM-4.5MoEIQ2_XXS358B83.01 GiB0.81 GiB84.86 GiB4.42 GiB59±37%
NVIDIA-Nemotron-3-Super-120B-A12B-BF16MoEUD-Q5_K_S124B83.56 GiB0.19 GiB84.74 GiB4.54 GiB66±37%
Llama-4-Scout-17B-16E-InstructMoEKV unresolvedQ6_K_L109B83.13 GiB0.42 GiB84.58 GiB4.70 GiB55±37%
GLM-4.7MoEIQ2_XXS358B82.69 GiB0.81 GiB84.54 GiB4.74 GiB59±37%
grok-2MoEUD-IQ1_S270B82.82 GiB0.56 GiB84.53 GiB4.75 GiB23±37%
Codestral-22B-v0.1F3222.2B82.88 GiB0.49 GiB84.43 GiB4.85 GiB12±22%
Mixtral-8x22B-v0.1MoEF32141B82.84 GiB0.49 GiB84.40 GiB4.88 GiB23±37%
Devstral-2-123B-Instruct-2512Q5_K_M125B82.25 GiB0.77 GiB84.18 GiB5.10 GiB12±22%
XORTRON-NXTXPRTXXLI1-Q5_K_M128B82.25 GiB0.77 GiB84.18 GiB5.10 GiB12±22%
Mistral-Small-4-119B-2603MoEUD-Q5_K_M119B83.04 GiB0.05 GiB84.12 GiB5.16 GiB75±37%
GLM-4.6-Derestricted-v3MoEIQ2_XXS357B82.26 GiB0.81 GiB84.11 GiB5.17 GiB59±37%
GLM-4.6MoEIQ2_XXS357B82.26 GiB0.81 GiB84.11 GiB5.17 GiB59±37%
Qwen3.5-397B-A17BMoEIQ1_S403B82.64 GiB0.07 GiB83.75 GiB5.53 GiB86±37%
OYM-Qimi-122B-A10B-K2.6MoEI1-Q5_K_M125B82.62 GiB0.05 GiB83.70 GiB5.58 GiB76±37%
Qwopus3.5-122B-A10B-Kimi-K2.6-destill-healed-abliteratedMoEQ5_K_M123B82.62 GiB0.05 GiB83.70 GiB5.58 GiB76±37%
Hermes-4-405BUD-IQ1_S406B81.23 GiB1.11 GiB83.62 GiB5.66 GiB12±22%
Trinity-Large-TrueBaseMoEI1-IQ1_M399B82.07 GiB0.35 GiB83.45 GiB5.83 GiB86±37%
Ace-Step1.5BF16160M82.03 GiB0.12 GiB83.14 GiB6.14 GiB13±22%
Laguna-S-2.1MoEUD-Q5_K_M118B81.83 GiB0.15 GiB82.99 GiB6.29 GiB70±37%
MiniMax-M2.1-REAP-139B-A10BMoEI1-Q4_1139B81.21 GiB0.54 GiB82.74 GiB6.54 GiB61±37%
m51Lab-MiniMax-M2.7-REAP-139B-A10BMoEI1-Q4_1139B81.21 GiB0.54 GiB82.74 GiB6.54 GiB61±37%
Behemoth-X-123B-v2Q5_K_M123B80.55 GiB0.77 GiB82.48 GiB6.80 GiB13±22%
Mistral-Large-Instruct-2411Q5_K_M123B80.55 GiB0.77 GiB82.48 GiB6.80 GiB13±22%
HopCoder-Mini-35B-A3B-VL36MoEF1635.1B81.19 GiB0.04 GiB82.24 GiB7.04 GiB77±37%
Qwen3.5-122B-A10B-hereticMoEI1-Q5_K_M123B80.98 GiB0.05 GiB82.06 GiB7.22 GiB77±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?
2087 of 2118 indexed open-weight models fit a RTX PRO 6000 Blackwell Max-Q Workstation Edition at 8,192 context with q4_0 KV cache, the largest being Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-MTP at Q6_K. 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.