NVIDIA · workstation

RTX PRO 6000 Blackwell Workstation Edition

RTX PRO 6000 Blackwell 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 128K 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
504 TF
dense
TDP
600 W
$8565 MSRP
Context4K8K16K32K64K128K
KV cachef16q8_0q4_0quantizing the KV cache is a ~2× lever on the dominant term at long context
text 1788vision language 189audio tts 21image 2audio asr 39video 16embedding 26

What fits at 128K context

largest quantization that fits, per model · 2081 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
MiMo-V2-FlashMoEKV unresolvedI1-IQ2_XS310B84.00 GiB4.22 GiB89.27 GiB0.01 GiB49±37%
ERNIE-4.5-300B-A47B-PTUD-IQ1_S300B80.54 GiB7.59 GiB89.26 GiB0.02 GiB12±22%
HarmonicHarlequin_v5-20BI1-Q3_K_M33.3B15.04 GiB73.13 GiB89.20 GiB0.08 GiB12±22%
Qwen3.5-122B-A10BMoEUD-Q5_K_M125B87.21 GiB0.84 GiB89.08 GiB0.20 GiB66±37%
GLM-4.7-REAP-218B-A32BMoEQ2_K_L218B75.08 GiB12.94 GiB89.06 GiB0.22 GiB25±37%
GLM-4.5MoEIQ1_M358B74.85 GiB12.94 GiB88.82 GiB0.46 GiB28±37%
Hy3MoEIQ2_XXS299B76.47 GiB11.25 GiB88.76 GiB0.52 GiB31±37%
Mixtral-8x22B-Instruct-v0.1MoEQ4_K_M141B79.71 GiB7.88 GiB88.65 GiB0.63 GiB19±37%
Mixtral-8x22B-v0.1MoEQ4_K_M141B79.71 GiB7.88 GiB88.65 GiB0.63 GiB19±37%
Mixtral-8x22B-v0.1MoEQ4_K_M141B79.71 GiB7.88 GiB88.64 GiB0.64 GiB19±37%
Qwen3.5-REAP-212B-A17BMoEIQ3_M212B86.49 GiB1.05 GiB88.60 GiB0.68 GiB61±37%
GLM-4.7MoEIQ1_M358B74.53 GiB12.94 GiB88.51 GiB0.77 GiB28±37%
dots.llm1.instMoEUD-IQ2_M143B52.59 GiB34.88 GiB88.49 GiB0.79 GiB14±37%
MiniMax-M2.1-REAP-139B-A10BMoEI1-Q4_K_M139B78.40 GiB8.72 GiB88.11 GiB1.17 GiB34±37%
m51Lab-MiniMax-M2.7-REAP-139B-A10BMoEI1-Q4_K_M139B78.40 GiB8.72 GiB88.11 GiB1.17 GiB34±37%
WizardLM-Uncensored-SuperCOT-StoryTelling-30bQ8_032.5B32.19 GiB54.84 GiB88.11 GiB1.17 GiB12±22%
archangel_sft-kto_llama30bQ8_032.5B32.19 GiB54.84 GiB88.11 GiB1.17 GiB12±22%
Wizard-Vicuna-30B-UncensoredQ8_032.5B32.19 GiB54.84 GiB88.11 GiB1.17 GiB12±22%
GLM-4.6-Derestricted-v3MoEIQ1_M357B74.10 GiB12.94 GiB88.08 GiB1.20 GiB28±37%
GLM-4.6MoEIQ1_M357B74.10 GiB12.94 GiB88.08 GiB1.20 GiB28±37%
DeepSeek-Coder-V2-Instruct-0724MoEIQ3_XXS236B84.61 GiB2.37 GiB88.02 GiB1.26 GiB55±37%
DeepSeek-V2.5MoEIQ3_XXS236B84.61 GiB2.37 GiB88.02 GiB1.26 GiB55±37%
DeepSeek-Coder-V2-InstructMoEIQ3_XXS236B84.61 GiB2.37 GiB88.02 GiB1.26 GiB55±37%
Trinity-Large-ThinkingMoEIQ1_M399B84.63 GiB2.33 GiB87.99 GiB1.29 GiB65±37%
Mistral-Medium-3.5-128BQ4_1128B74.29 GiB12.38 GiB87.82 GiB1.46 GiB12±22%
NVIDIA-Nemotron-3-Super-120B-A12B-BF16MoEUD-Q5_K_S124B83.56 GiB3.09 GiB87.65 GiB1.63 GiB50±37%
step-3.5-flashIQ3_XXS199B72.82 GiB13.79 GiB87.64 GiB1.64 GiB12±22%
Qwen3-VL-235B-A22B-ThinkingMoEQ2_K_L236B79.94 GiB6.61 GiB87.58 GiB1.70 GiB36±37%
Qwen3-VL-235B-A22B-InstructMoEQ2_K_L236B79.94 GiB6.61 GiB87.58 GiB1.70 GiB36±37%
Qwen3-235B-A22BMoEQ2_K_L235B79.94 GiB6.61 GiB87.58 GiB1.70 GiB36±37%
Qwen3-235B-A22B-Instruct-2507MoEQ2_K_L235B79.94 GiB6.61 GiB87.58 GiB1.70 GiB36±37%
Qwen3-235B-A22B-Thinking-2507MoEQ2_K_L235B79.94 GiB6.61 GiB87.58 GiB1.70 GiB36±37%
MiniMax-M2.7MoEUD-IQ3_S229B77.87 GiB8.72 GiB87.58 GiB1.70 GiB37±37%
Qwen3-235B-A22B-abliteratedMoEI1-Q2_K235B79.81 GiB6.61 GiB87.45 GiB1.83 GiB36±37%
MiniMax-M2.1MoEQ2_K_L229B77.72 GiB8.72 GiB87.42 GiB1.86 GiB37±37%
MiniMax-M2.5MoEQ2_K_L229B77.72 GiB8.72 GiB87.42 GiB1.86 GiB37±37%
MiniMax-M2MoEQ2_K_L229B77.72 GiB8.72 GiB87.42 GiB1.86 GiB37±37%
MiniMax-M2.7-BF16-ultra-uncensored-hereticMoEI1-Q2_K229B77.58 GiB8.72 GiB87.29 GiB1.99 GiB37±37%
Ornith-1.0-397BMoEIQ1_M397B85.09 GiB1.05 GiB87.19 GiB2.09 GiB73±37%
Devstral-2-123B-Instruct-2512Q4_1125B73.08 GiB12.38 GiB86.61 GiB2.67 GiB12±22%
XORTRON-NXTXPRTXXLI1-Q4_1128B73.08 GiB12.38 GiB86.61 GiB2.67 GiB12±22%
grok-2MoEUD-TQ1_0270B76.17 GiB9.00 GiB86.31 GiB2.97 GiB19±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%
GLM-4.5-Air-DerestrictedMoEQ5_K_M110B77.97 GiB6.47 GiB85.47 GiB3.81 GiB37±37%
GLM-4.5-AirMoEQ5_K_M110B77.97 GiB6.47 GiB85.47 GiB3.81 GiB37±37%
Trinity-Large-TrueBaseMoEI1-IQ1_M399B82.07 GiB2.33 GiB85.43 GiB3.85 GiB67±37%
Step-3.5-Flash-REAP-121B-A11BI1-Q4_1121B70.61 GiB13.79 GiB85.43 GiB3.85 GiB12±22%
MiMo-V2.5MoEKV unresolvedIQ2_XXS311B79.83 GiB4.22 GiB85.10 GiB4.18 GiB50±37%
Hunyuan-A13B-InstructMoEQ8_080.4B79.58 GiB4.50 GiB85.08 GiB4.20 GiB12±22%
Behemoth-X-123B-v2Q4_1123B71.45 GiB12.38 GiB84.98 GiB4.30 GiB12±22%
Mistral-Small-4-119B-2603MoEUD-Q5_K_M119B83.04 GiB0.79 GiB84.86 GiB4.42 GiB69±37%
Qwen3.5-397B-A17BMoEIQ1_S403B82.64 GiB1.05 GiB84.74 GiB4.54 GiB75±37%
Ace-Step1.5BF16160M82.03 GiB1.70 GiB84.72 GiB4.56 GiB12±22%
Qwen2.5-72B-InstructQ8_072.7B72.21 GiB11.25 GiB84.59 GiB4.69 GiB12±22%
Laguna-S-2.1MoEUD-Q5_K_M118B81.83 GiB1.73 GiB84.58 GiB4.70 GiB59±37%
OYM-Qimi-122B-A10B-K2.6MoEI1-Q5_K_M125B82.62 GiB0.84 GiB84.49 GiB4.79 GiB69±37%
Qwopus3.5-122B-A10B-Kimi-K2.6-destill-healed-abliteratedMoEQ5_K_M123B82.62 GiB0.84 GiB84.49 GiB4.79 GiB69±37%
HuatuoGPT-o1-72BQ8_072.7B71.96 GiB11.25 GiB84.34 GiB4.94 GiB12±22%
Qwen2.5-72B-Instruct-abliteratedQ8_072.7B71.96 GiB11.25 GiB84.34 GiB4.94 GiB12±22%
●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 processing14316.57 tok/s9546.25–16645.0038
Text generation267.03 tok/s256.42–275.7724
Benchmarked· n=38

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 Workstation Edition run?
2081 of 2118 indexed open-weight models fit a RTX PRO 6000 Blackwell Workstation Edition at 131,072 context with q4_0 KV cache, the largest being MiMo-V2-Flash at I1-IQ2_XS. That covers text, vision-language, image, video and speech models.
How much usable memory does a RTX PRO 6000 Blackwell 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 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.