NVIDIA · workstation

RTX 5000 Ada Generation

RTX 5000 Ada Generation has 32 GB of VRAM at 576 GB/s — about 29.76 GiB usable after driver and compositor overhead. 2011 of 2118 indexed models fit at 8K context with f16 KV.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed
Memory
32 GB
GDDR6
Bandwidth
576 GB/s
256-bit bus
Tensor FP16
261 TF
dense
TDP
250 W
$4000 MSRP
Context4K8K16K32K64K128K
KV cachef16q8_0q4_0quantizing the KV cache is a ~2× lever on the dominant term at long context
text 1726video 16vision language 181image 2embedding 26audio tts 21audio asr 39

What fits at 8K context

largest quantization that fits, per model · 2011 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
Phi-3.5-MoE-instructMoEKV unresolvedQ5_K_L41.9B27.75 GiB1.00 GiB29.76 GiB0.00 GiB31±37%
Bernini-RQ8_014.3B28.71 GiB0.00 GiB29.76 GiB0.00 GiB12±22%
Yi-34B-200K-DARE-megamerge-v8Q6_K34.4B26.78 GiB1.88 GiB29.74 GiB0.02 GiB12±22%
Nous-Hermes-2-Yi-34BI1-Q6_K34.4B26.78 GiB1.88 GiB29.74 GiB0.02 GiB12±22%
Nous-Capybara-limarpv3-34BI1-Q6_K34.4B26.78 GiB1.88 GiB29.74 GiB0.02 GiB12±22%
Apertus-70B-Instruct-2509UD-IQ3_XXS70.6B26.01 GiB2.50 GiB29.69 GiB0.07 GiB12±22%
Rombo-LLM-V3.0-Qwen-72bI1-IQ2_S72.7B26.02 GiB2.50 GiB29.65 GiB0.11 GiB12±22%
Qwen2.5-72B-Instruct-abliteratedI1-IQ2_S72.7B26.02 GiB2.50 GiB29.65 GiB0.11 GiB12±22%
Qwen2.5-72B-Instruct-abliterated-v2I1-IQ2_S72.7B26.02 GiB2.50 GiB29.65 GiB0.11 GiB12±22%
HuatuoGPT-o1-72BIQ2_S72.7B26.02 GiB2.50 GiB29.65 GiB0.11 GiB12±22%
MiroThinker-v1.0-72BI1-IQ2_S72.7B26.02 GiB2.50 GiB29.65 GiB0.11 GiB12±22%
Malaysian-Qwen2.5-72B-InstructI1-IQ2_S72.7B26.02 GiB2.50 GiB29.65 GiB0.11 GiB12±22%
Qwen2.5-72BI1-IQ2_S72.7B26.02 GiB2.50 GiB29.65 GiB0.11 GiB12±22%
magnum-v4-72bI1-IQ2_S72.7B26.02 GiB2.50 GiB29.65 GiB0.11 GiB12±22%
KAT-Dev-72B-ExpIQ2_S72.7B26.02 GiB2.50 GiB29.65 GiB0.11 GiB12±22%
Homer-v1.0-Qwen2.5-72BIQ2_S72.7B26.02 GiB2.50 GiB29.65 GiB0.11 GiB12±22%
Tower-Plus-72B-ultra-uncensored-hereticI1-IQ2_S72.7B26.02 GiB2.50 GiB29.65 GiB0.11 GiB12±22%
Qwen2.5-VL-72B-InstructIQ2_S73.4B26.02 GiB2.50 GiB29.65 GiB0.11 GiB12±22%
UI-TARS-72B-DPOIQ2_S73.4B26.02 GiB2.50 GiB29.65 GiB0.11 GiB12±22%
Salience-1.5-ProMoEQ6_K36.0B28.43 GiB0.16 GiB29.59 GiB0.17 GiB67±37%
Qwable-v1MoEQ6_K36.0B28.43 GiB0.16 GiB29.59 GiB0.17 GiB67±37%
T-SearchMoEQ6_K36.0B28.43 GiB0.16 GiB29.59 GiB0.17 GiB67±37%
archangel_sft-kto_llama30bIQ4_XS32.5B16.28 GiB12.19 GiB29.53 GiB0.23 GiB12±22%
Qwen3.5-88BMoEI1-Q2_K_S87.7B28.30 GiB0.19 GiB29.51 GiB0.25 GiB60±37%
Kimi-Linear-48B-A3B-InstructMoEQ4_K_L49.1B28.26 GiB0.24 GiB29.50 GiB0.26 GiB12±22%
Gemma-4-Novelist-Eclipse-31BQ6_K32.7B25.97 GiB2.42 GiB29.47 GiB0.29 GiB12±22%
Gemma-4-31B-StyleTuneQ6_K32.7B25.97 GiB2.42 GiB29.47 GiB0.29 GiB12±22%
MiniMax-M2.1-REAP-139B-A10BMoEI1-IQ1_S139B26.55 GiB1.94 GiB29.47 GiB0.29 GiB38±37%
m51Lab-MiniMax-M2.7-REAP-139B-A10BMoEI1-IQ1_S139B26.55 GiB1.94 GiB29.47 GiB0.29 GiB38±37%
Apriel-1.6-15b-ThinkerBF1614.9B26.88 GiB1.50 GiB29.43 GiB0.33 GiB12±22%
Assistant_Pepe_70BQ2_K70.6B25.79 GiB2.50 GiB29.42 GiB0.34 GiB12±22%
Wizard-Vicuna-30B-UncensoredI1-IQ4_XS32.5B16.15 GiB12.19 GiB29.41 GiB0.35 GiB12±22%
Hunyuan-A13B-InstructMoEQ2_K80.4B27.40 GiB1.00 GiB29.39 GiB0.37 GiB12±22%
ALIA-40b-fc-2606I1-Q5_K_M40.4B26.78 GiB1.50 GiB29.39 GiB0.37 GiB12±22%
ALIA-40b-instruct-2606I1-Q5_K_M40.4B26.78 GiB1.50 GiB29.39 GiB0.37 GiB12±22%
Qwen3-Coder-NextMoEUD-IQ3_S79.7B27.65 GiB0.75 GiB29.39 GiB0.37 GiB63±37%
Hermes-4-70BUD-IQ3_XXS70.6B25.76 GiB2.50 GiB29.38 GiB0.38 GiB12±22%
Llama-3.3-70B-InstructUD-IQ3_XXS70.6B25.76 GiB2.50 GiB29.38 GiB0.38 GiB12±22%
DeepSeek-R1-Distill-Llama-70BUD-IQ3_XXS70.6B25.76 GiB2.50 GiB29.38 GiB0.38 GiB12±22%
Darwin-35B-A3B-OpusMoEQ6_K_L36.0B28.22 GiB0.16 GiB29.38 GiB0.38 GiB68±37%
Aurora-Code-1MoEQ6_K_L34.7B28.22 GiB0.16 GiB29.38 GiB0.38 GiB68±37%
grug-35b-v2MoEQ6_K_L35.1B28.22 GiB0.16 GiB29.38 GiB0.38 GiB68±37%
grug-35bMoEQ6_K_L35.1B28.22 GiB0.16 GiB29.38 GiB0.38 GiB68±37%
WorldSim-Opus-3.6-35B-A3BMoEQ6_K_L35.1B28.22 GiB0.16 GiB29.38 GiB0.38 GiB68±37%
Qwen3.6-35B-A3B-AnkoMoEQ6_K_L35.1B28.22 GiB0.16 GiB29.38 GiB0.38 GiB68±37%
KAT-Coder-V2.5-DevMoEQ6_K_L34.7B28.22 GiB0.16 GiB29.38 GiB0.38 GiB68±37%
Ornith-1.0-35BMoEQ6_K_L34.7B28.22 GiB0.16 GiB29.38 GiB0.38 GiB68±37%
Nex-N2-miniMoEQ6_K_L35.1B28.22 GiB0.16 GiB29.38 GiB0.38 GiB68±37%
Qwen3.6-27B-Heretic2-Uncensored-Finetune-ThinkingQ8_027.4B27.82 GiB0.50 GiB29.38 GiB0.38 GiB12±22%
WizardLM-Uncensored-SuperCOT-StoryTelling-30bQ3_K_L32.5B16.09 GiB12.19 GiB29.35 GiB0.41 GiB12±22%
Qwen3-42B-A3B-2507-Thinking-Abliterated-uncensored-TOTAL-RECALL-v2-Medium-MASTER-CODERMoEI1-Q5_K_S42.4B27.23 GiB1.05 GiB29.27 GiB0.49 GiB43±37%
CalmeRys-78B-Orpo-v0.1I1-IQ2_XXS78.0B25.43 GiB2.69 GiB29.25 GiB0.51 GiB12±22%
calme-2.3-rys-78bIQ2_XXS78.0B25.43 GiB2.69 GiB29.25 GiB0.51 GiB12±22%
dolphin-2.9.1-yi-1.5-34b-hereticQ6_K34.4B26.28 GiB1.88 GiB29.24 GiB0.52 GiB12±22%
dolphin-2.9.1-yi-1.5-34bI1-Q6_K34.4B26.28 GiB1.88 GiB29.24 GiB0.52 GiB12±22%
OrionStar-Yi-34B-Chat-LlamaI1-Q6_K34.4B26.28 GiB1.88 GiB29.24 GiB0.52 GiB12±22%
Yi-34B-200K-LlamafiedI1-Q6_K34.4B26.28 GiB1.88 GiB29.24 GiB0.52 GiB12±22%
Yi-1.5-34BQ6_K34.4B26.28 GiB1.88 GiB29.24 GiB0.52 GiB12±22%
Merged-RP-Stew-V2-34BI1-Q6_K34.4B26.28 GiB1.88 GiB29.24 GiB0.52 GiB12±22%
Capybara-Tess-Yi-34B-200KQ6_K34.4B26.28 GiB1.88 GiB29.24 GiB0.52 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.

Questions people ask

What AI models can a RTX 5000 Ada Generation run?
2011 of 2118 indexed open-weight models fit a RTX 5000 Ada Generation at 8,192 context with f16 KV cache, the largest being Phi-3.5-MoE-instruct at Q5_K_L. That covers text, vision-language, image, video and speech models.
How much usable memory does a RTX 5000 Ada Generation actually have?
Its nameplate is 32 GB, but about 29.76 GiB is available to a model once driver and compositor overhead is accounted for.
Is a RTX 5000 Ada Generation fast for local AI?
Its memory bandwidth is 576 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.