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. 2081 of 2118 indexed models fit at 64K context with q8_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 189text 1788audio tts 21image 2audio asr 39video 16embedding 26

What fits at 64K context

largest quantization that fits, per model · 2081 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
Qwen3.5-122B-A10BMoEUD-Q5_K_M125B87.21 GiB0.80 GiB89.03 GiB0.25 GiB66±37%
MiMo-V2-FlashMoEKV unresolvedI1-IQ2_XS310B84.00 GiB3.98 GiB89.03 GiB0.25 GiB50±37%
c4ai-command-r-plus-08-2024Q6_K104B79.32 GiB8.50 GiB89.00 GiB0.28 GiB12±22%
Step-3.7-FlashUD-IQ3_S201B74.54 GiB13.30 GiB88.86 GiB0.42 GiB12±22%
ERNIE-4.5-300B-A47B-PTUD-IQ1_S300B80.54 GiB7.17 GiB88.84 GiB0.44 GiB12±22%
HarmonicHarlequin_v5-20BI1-Q4_K_M33.3B18.71 GiB69.06 GiB88.81 GiB0.47 GiB12±22%
Qwen3.5-REAP-212B-A17BMoEIQ3_M212B86.49 GiB1.00 GiB88.54 GiB0.74 GiB62±37%
GLM-4.7-REAP-218B-A32BMoEQ2_K_L218B75.08 GiB12.22 GiB88.34 GiB0.94 GiB26±37%
Mixtral-8x22B-Instruct-v0.1MoEQ4_K_M141B79.71 GiB7.44 GiB88.21 GiB1.07 GiB19±37%
Mixtral-8x22B-v0.1MoEQ4_K_M141B79.71 GiB7.44 GiB88.21 GiB1.07 GiB19±37%
Mixtral-8x22B-v0.1MoEQ4_K_M141B79.71 GiB7.44 GiB88.21 GiB1.07 GiB19±37%
Hy3MoEIQ2_XXS299B76.47 GiB10.63 GiB88.14 GiB1.14 GiB32±37%
GLM-4.5MoEIQ1_M358B74.85 GiB12.22 GiB88.11 GiB1.17 GiB29±37%
Trinity-Large-ThinkingMoEIQ1_M399B84.63 GiB2.41 GiB88.07 GiB1.21 GiB65±37%
DeepSeek-Coder-V2-Instruct-0724MoEIQ3_XXS236B84.61 GiB2.24 GiB87.89 GiB1.39 GiB55±37%
DeepSeek-V2.5MoEIQ3_XXS236B84.61 GiB2.24 GiB87.89 GiB1.39 GiB55±37%
DeepSeek-Coder-V2-InstructMoEIQ3_XXS236B84.61 GiB2.24 GiB87.89 GiB1.39 GiB55±37%
GLM-4.7MoEIQ1_M358B74.53 GiB12.22 GiB87.79 GiB1.49 GiB29±37%
dots.llm1.instMoEIQ2_M143B53.68 GiB32.94 GiB87.65 GiB1.63 GiB15±37%
MiniMax-M2.1-REAP-139B-A10BMoEI1-Q4_K_M139B78.40 GiB8.23 GiB87.62 GiB1.66 GiB34±37%
m51Lab-MiniMax-M2.7-REAP-139B-A10BMoEI1-Q4_K_M139B78.40 GiB8.23 GiB87.62 GiB1.66 GiB34±37%
NVIDIA-Nemotron-3-Super-120B-A12B-BF16MoEUD-Q5_K_S124B83.56 GiB2.92 GiB87.47 GiB1.81 GiB51±37%
GLM-4.6-Derestricted-v3MoEIQ1_M357B74.10 GiB12.22 GiB87.36 GiB1.92 GiB29±37%
GLM-4.6MoEIQ1_M357B74.10 GiB12.22 GiB87.36 GiB1.92 GiB29±37%
Qwen3-VL-235B-A22B-ThinkingMoEQ2_K_L236B79.94 GiB6.24 GiB87.22 GiB2.06 GiB37±37%
Qwen3-VL-235B-A22B-InstructMoEQ2_K_L236B79.94 GiB6.24 GiB87.22 GiB2.06 GiB37±37%
Qwen3-235B-A22BMoEQ2_K_L235B79.94 GiB6.24 GiB87.22 GiB2.06 GiB37±37%
Qwen3-235B-A22B-Instruct-2507MoEQ2_K_L235B79.94 GiB6.24 GiB87.22 GiB2.06 GiB37±37%
Qwen3-235B-A22B-Thinking-2507MoEQ2_K_L235B79.94 GiB6.24 GiB87.22 GiB2.06 GiB37±37%
step-3.5-flashIQ3_XXS199B72.82 GiB13.30 GiB87.15 GiB2.13 GiB12±22%
Mistral-Medium-3.5-128BQ4_1128B74.29 GiB11.69 GiB87.13 GiB2.15 GiB12±22%
Ornith-1.0-397BMoEIQ1_M397B85.09 GiB1.00 GiB87.13 GiB2.15 GiB74±37%
MiniMax-M2.7MoEUD-IQ3_S229B77.87 GiB8.23 GiB87.09 GiB2.19 GiB38±37%
Qwen3-235B-A22B-abliteratedMoEI1-Q2_K235B79.81 GiB6.24 GiB87.08 GiB2.20 GiB37±37%
MiniMax-M2.1MoEQ2_K_L229B77.72 GiB8.23 GiB86.94 GiB2.34 GiB38±37%
MiniMax-M2.5MoEQ2_K_L229B77.72 GiB8.23 GiB86.94 GiB2.34 GiB38±37%
MiniMax-M2MoEQ2_K_L229B77.72 GiB8.23 GiB86.94 GiB2.34 GiB38±37%
MiniMax-M2.7-BF16-ultra-uncensored-hereticMoEI1-Q2_K229B77.58 GiB8.23 GiB86.80 GiB2.48 GiB38±37%
Devstral-2-123B-Instruct-2512Q4_1125B73.08 GiB11.69 GiB85.93 GiB3.35 GiB12±22%
XORTRON-NXTXPRTXXLI1-Q4_1128B73.08 GiB11.69 GiB85.93 GiB3.35 GiB12±22%
grok-2MoEUD-TQ1_0270B76.17 GiB8.50 GiB85.81 GiB3.47 GiB19±37%
DeepSeek-V4-Flash-0731MoEUD-IQ2_M304B84.68 GiB0.03 GiB85.76 GiB3.52 GiB80±37%
DeepSeek-V4-FlashMoEUD-IQ2_M291B84.68 GiB0.03 GiB85.76 GiB3.52 GiB80±37%
Trinity-Large-TrueBaseMoEI1-IQ1_M399B82.07 GiB2.41 GiB85.51 GiB3.77 GiB66±37%
GLM-4.5-Air-DerestrictedMoEQ5_K_M110B77.97 GiB6.11 GiB85.11 GiB4.17 GiB38±37%
GLM-4.5-AirMoEQ5_K_M110B77.97 GiB6.11 GiB85.11 GiB4.17 GiB38±37%
WizardLM-Uncensored-SuperCOT-StoryTelling-30bQ8_032.5B32.19 GiB51.80 GiB85.06 GiB4.22 GiB12±22%
archangel_sft-kto_llama30bQ8_032.5B32.19 GiB51.80 GiB85.06 GiB4.22 GiB12±22%
Wizard-Vicuna-30B-UncensoredQ8_032.5B32.19 GiB51.80 GiB85.06 GiB4.22 GiB12±22%
Step-3.5-Flash-REAP-121B-A11BI1-Q4_1121B70.61 GiB13.30 GiB84.93 GiB4.35 GiB12±22%
MiMo-V2.5MoEKV unresolvedIQ2_XXS311B79.83 GiB3.98 GiB84.86 GiB4.42 GiB51±37%
Hunyuan-A13B-InstructMoEQ8_080.4B79.58 GiB4.25 GiB84.83 GiB4.45 GiB12±22%
Mistral-Small-4-119B-2603MoEUD-Q5_K_M119B83.04 GiB0.75 GiB84.82 GiB4.46 GiB69±37%
Qwen3.5-397B-A17BMoEIQ1_S403B82.64 GiB1.00 GiB84.68 GiB4.60 GiB75±37%
Ace-Step1.5BF16160M82.03 GiB1.61 GiB84.64 GiB4.64 GiB12±22%
Laguna-S-2.1MoEUD-Q5_K_M118B81.83 GiB1.67 GiB84.52 GiB4.76 GiB59±37%
OYM-Qimi-122B-A10B-K2.6MoEI1-Q5_K_M125B82.62 GiB0.80 GiB84.44 GiB4.84 GiB69±37%
Qwopus3.5-122B-A10B-Kimi-K2.6-destill-healed-abliteratedMoEQ5_K_M123B82.62 GiB0.80 GiB84.44 GiB4.84 GiB69±37%
Behemoth-X-123B-v2Q4_1123B71.45 GiB11.69 GiB84.29 GiB4.99 GiB12±22%
gpt-oss-120b-Uncensored-xCloudMoEI1-Q5_K_S117B81.92 GiB1.21 GiB84.12 GiB5.16 GiB66±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?
2081 of 2118 indexed open-weight models fit a RTX PRO 6000 Blackwell Max-Q Workstation Edition at 65,536 context with q8_0 KV cache, the largest being Qwen3.5-122B-A10B at UD-Q5_K_M. 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.