NVIDIA · workstation

RTX 4000 Ada Generation

RTX 4000 Ada Generation has 20 GB of VRAM at 360 GB/s — about 18.60 GiB usable after driver and compositor overhead. 1952 of 2118 indexed models fit at 4K context with q8_0 KV.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed
Memory
20 GB
GDDR6
Bandwidth
360 GB/s
160-bit bus
Tensor FP16
107 TF
dense
TDP
130 W
$1250 MSRP
Context4K8K16K32K64K128K
KV cachef16q8_0q4_0quantizing the KV cache is a ~2× lever on the dominant term at long context
text 1675vision language 173video 16image 2audio asr 39audio tts 21embedding 26

What fits at 4K context

largest quantization that fits, per model · 1952 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
Qwen3.6-34B-80L-Fable-5-HereticI1-IQ4_XS33.4B17.37 GiB0.17 GiB18.60 GiB0.00 GiB12±22%
Qwen3.6-27B-uncensored-heretic-v2Q5_K_S27.4B17.40 GiB0.13 GiB18.59 GiB0.01 GiB12±22%
Qwen3.5-27B-Engineer-Deckard-GeminiI1-Q5_K_S27.7B17.40 GiB0.13 GiB18.59 GiB0.01 GiB12±22%
Qwen3.5-27B-HERETIC-Polaris-Advanced-Thinking-Alpha-uncensoredI1-Q5_K_S27.4B17.40 GiB0.13 GiB18.59 GiB0.01 GiB12±22%
Qwen3.5-27B-Deckard-PKD-Heretic-Uncensored-ThinkingI1-Q5_K_S27.4B17.40 GiB0.13 GiB18.59 GiB0.01 GiB12±22%
Qwen3.6-27B-Heretic2-ThinkingI1-Q5_K_S27.4B17.40 GiB0.13 GiB18.59 GiB0.01 GiB12±22%
Qwen3.6-27B-Uncensored-AggressiveI1-Q5_K_S27.4B17.40 GiB0.13 GiB18.59 GiB0.01 GiB12±22%
Qwen-3.5-Opus-GLM-27BI1-Q5_K_S26.9B17.40 GiB0.13 GiB18.59 GiB0.01 GiB12±22%
Qwen3.6-27B-abliteratedI1-Q5_K_S27.4B17.40 GiB0.13 GiB18.59 GiB0.01 GiB12±22%
KoQweopus-3.5-27B-experimentalI1-Q5_K_S27.8B17.40 GiB0.13 GiB18.59 GiB0.01 GiB12±22%
Webcoda-AI-27BI1-Q5_K_S27.4B17.40 GiB0.13 GiB18.59 GiB0.01 GiB12±22%
Qwen3.5-27B-imabari-v2I1-Q5_K_S27.8B17.40 GiB0.13 GiB18.59 GiB0.01 GiB12±22%
Qwen3.6-27B-AEON-Ultimate-Uncensored-BF16Q5_K_S27.4B17.40 GiB0.13 GiB18.59 GiB0.01 GiB12±22%
Huihui-Qwen3.5-27B-abliteratedI1-Q5_K_S27.8B17.40 GiB0.13 GiB18.59 GiB0.01 GiB12±22%
Qwen3.5-27B-uncensored-heretic-v1I1-Q5_K_S27.4B17.40 GiB0.13 GiB18.59 GiB0.01 GiB12±22%
Qwen3.5-27B-Unredacted-MAXI1-Q5_K_S27.4B17.40 GiB0.13 GiB18.59 GiB0.01 GiB12±22%
Qwen3.5-27B-hereticI1-Q5_K_S27.4B17.40 GiB0.13 GiB18.59 GiB0.01 GiB12±22%
Carnice-V2-27bI1-Q5_K_S27.4B17.40 GiB0.13 GiB18.59 GiB0.01 GiB12±22%
Qwen3.5-Queen-27BI1-Q5_K_S27.4B17.40 GiB0.13 GiB18.59 GiB0.01 GiB12±22%
GRaPE-2-ProI1-Q5_K_S27.8B17.40 GiB0.13 GiB18.59 GiB0.01 GiB12±22%
Huihui-Qwen3.6-27B-abliteratedQ5_K_S27.8B17.40 GiB0.13 GiB18.59 GiB0.01 GiB12±22%
Qwen3.5-27B-abliteratedQ5_K_S26.9B17.40 GiB0.13 GiB18.59 GiB0.01 GiB12±22%
Qwen3.5-27B-DerestrictedI1-Q5_K_S27.8B17.40 GiB0.13 GiB18.59 GiB0.01 GiB12±22%
ThinkingCap-Qwen3.6-27B-hereticQ5_K_S27.4B17.40 GiB0.13 GiB18.59 GiB0.01 GiB12±22%
Fara1.5-27BQ5_027.4B17.40 GiB0.13 GiB18.59 GiB0.01 GiB12±22%
Qwen3.6-27B-Omnimerge-v4Q5_K_S27.8B17.40 GiB0.13 GiB18.59 GiB0.01 GiB12±22%
Qwen3.6-27B-Claude-Opus-Reasoning-Distill-v2-hereticQ5_027.4B17.40 GiB0.13 GiB18.59 GiB0.01 GiB12±22%
Darwin-28B-REASONI1-Q5_K_S26.9B17.40 GiB0.13 GiB18.59 GiB0.01 GiB12±22%
Huihui-Qwen3.5-27B-Claude-4.6-Opus-abliteratedI1-Q5_K_S27.8B17.40 GiB0.13 GiB18.59 GiB0.01 GiB12±22%
Qwen3.5-27B-Claude-4.6-Opus-Reasoning-DistilledI1-Q5_K_S27.8B17.40 GiB0.13 GiB18.59 GiB0.01 GiB12±22%
Qwen3.5-27B-WebNovel-Writer-zhI1-Q5_K_S26.9B17.40 GiB0.13 GiB18.59 GiB0.01 GiB12±22%
Qwen3.5-27B_Homebrew-v2I1-Q5_K_S27.4B17.40 GiB0.13 GiB18.59 GiB0.01 GiB12±22%
EXAONE-4.0-32BQ4_032.0B16.96 GiB0.53 GiB18.59 GiB0.01 GiB12±22%
Pantheon-Reasoning-26B-A4B-1.1MoEQ5_K_S26.5B17.36 GiB0.24 GiB18.58 GiB0.02 GiB12±22%
Gemma-4-Gembrain-X-Core-31BI1-Q4_K_S31.3B16.54 GiB0.95 GiB18.58 GiB0.02 GiB12±22%
Gemma-4-Gembrain-X-31BI1-Q4_K_S31.3B16.54 GiB0.95 GiB18.58 GiB0.02 GiB12±22%
Gemma-4-31B-Isometry-Fabled-PersonaI1-Q4_K_S31.3B16.54 GiB0.95 GiB18.58 GiB0.02 GiB12±22%
Versipellis-31BI1-Q4_K_S31.3B16.54 GiB0.95 GiB18.58 GiB0.02 GiB12±22%
Gemma4-Gutenberg-31BI1-Q4_K_S31.3B16.54 GiB0.95 GiB18.58 GiB0.02 GiB12±22%
G4-MeroMero-31B-uncensored-hereticI1-Q4_K_S31.3B16.54 GiB0.95 GiB18.58 GiB0.02 GiB12±22%
Gemma-4-Novelist-31BI1-Q4_K_S31.3B16.54 GiB0.95 GiB18.58 GiB0.02 GiB12±22%
Wanabi-Gemma4-31BI1-Q4_K_S31.3B16.54 GiB0.95 GiB18.58 GiB0.02 GiB12±22%
G4-Alice-v1.2-31BI1-Q4_K_S31.3B16.54 GiB0.95 GiB18.58 GiB0.02 GiB12±22%
Agares-31B-v1I1-Q4_K_S30.7B16.54 GiB0.95 GiB18.58 GiB0.02 GiB12±22%
Gemma4-Gutenberg-31B-HereticI1-Q4_K_S31.3B16.54 GiB0.95 GiB18.58 GiB0.02 GiB12±22%
gemma-4-Ortenzya-The-Creative-Wordsmith-31B-it-uncensored-hereticI1-Q4_K_S31.3B16.54 GiB0.95 GiB18.58 GiB0.02 GiB12±22%
Gemma-4-Gemsicle-31BI1-Q4_K_S31.3B16.54 GiB0.95 GiB18.58 GiB0.02 GiB12±22%
Gemma-4-Gembrain-31B-it-uncensored-hereticI1-Q4_K_S31.3B16.54 GiB0.95 GiB18.58 GiB0.02 GiB12±22%
Melinoe-Gemma4-31B-VL-hereticI1-Q4_K_S31.3B16.54 GiB0.95 GiB18.58 GiB0.02 GiB12±22%
G4-MeroMero-31BI1-Q4_K_S31.3B16.54 GiB0.95 GiB18.58 GiB0.02 GiB12±22%
Glistening-Gem-31B-v1.0I1-Q4_K_S31.3B16.54 GiB0.95 GiB18.58 GiB0.02 GiB12±22%
Melinoe-Gemma4-31B-VLI1-Q4_K_S31.3B16.54 GiB0.95 GiB18.58 GiB0.02 GiB12±22%
Gemma-4-31B-Storymaxxed3I1-Q4_K_S31.3B16.54 GiB0.95 GiB18.58 GiB0.02 GiB12±22%
Huihui-gemma-4-31B-it-qat-q4_0-unquantized-abliteratedI1-Q4_K_S32.7B16.54 GiB0.95 GiB18.58 GiB0.02 GiB12±22%
gemma-4-31B-Queen-it-qat-q4_0-unquantizedI1-Q4_K_S31.3B16.54 GiB0.95 GiB18.58 GiB0.02 GiB12±22%
gemma-4-31B-it-qat-q4_0-unquantized-hereticI1-Q4_K_S31.3B16.54 GiB0.95 GiB18.58 GiB0.02 GiB12±22%
Gemma-4-AssGuard-31BI1-Q4_K_S31.3B16.54 GiB0.95 GiB18.58 GiB0.02 GiB12±22%
copywriter-gemma4-31bI1-Q4_K_S32.7B16.54 GiB0.95 GiB18.58 GiB0.02 GiB12±22%
gemma-4-31B-heretic-finetuneI1-Q4_K_S30.7B16.54 GiB0.95 GiB18.58 GiB0.02 GiB12±22%
Gemma-4-Garnet-V2-31B-it-ultra-uncensored-hereticI1-Q4_K_S31.3B16.54 GiB0.95 GiB18.58 GiB0.02 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 4000 Ada Generation run?
1952 of 2118 indexed open-weight models fit a RTX 4000 Ada Generation at 4,096 context with q8_0 KV cache, the largest being Qwen3.6-34B-80L-Fable-5-Heretic at I1-IQ4_XS. That covers text, vision-language, image, video and speech models.
How much usable memory does a RTX 4000 Ada Generation actually have?
Its nameplate is 20 GB, but about 18.60 GiB is available to a model once driver and compositor overhead is accounted for.
Is a RTX 4000 Ada Generation fast for local AI?
Its memory bandwidth is 360 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.