NVIDIA · consumer

GeForce RTX 3080 Ti

GeForce RTX 3080 Ti has 20 GB of VRAM at 760 GB/s — about 18.60 GiB usable after driver and compositor overhead. 1949 of 2118 indexed models fit at 16K context with q4_0 KV.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed
Memory
20 GB
GDDR6X
Bandwidth
760 GB/s
320-bit bus
Tensor FP16
136 TF
dense
TDP
350 W
$1199 MSRP
Context4K8K16K32K64K128K
KV cachef16q8_0q4_0quantizing the KV cache is a ~2× lever on the dominant term at long context
text 1672vision language 173video 16audio asr 39image 2audio tts 21embedding 26

What fits at 16K context

largest quantization that fits, per model · 1949 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
Gemma-4-Novelist-Eclipse-31BIQ4_XS32.7B16.67 GiB1.03 GiB18.59 GiB0.01 GiB31±12.9%
Gemma-4-31B-StyleTuneIQ4_XS32.7B16.67 GiB1.03 GiB18.59 GiB0.01 GiB31±12.9%
t5-v1_1-xxlF324.8B17.74 GiB0.00 GiB18.59 GiB0.01 GiB31±12.9%
Skyfall-31B-v4.2-hereticI1-Q4_K_S31.4B16.71 GiB0.95 GiB18.58 GiB0.02 GiB31±12.9%
Skyfall-31B-v4.2I1-Q4_K_S31.4B16.71 GiB0.95 GiB18.58 GiB0.02 GiB31±12.9%
GLM-Z1-32B-0414-uncensored-heretic-v2Q4_K_S32.6B17.42 GiB0.27 GiB18.58 GiB0.02 GiB31±12.9%
Qwen3.6-34B-80L-Fable-5-HereticI1-IQ4_XS33.4B17.37 GiB0.35 GiB18.58 GiB0.02 GiB31±12.9%
GRM-2.6-Plus-0628Q4_K_L27.8B17.43 GiB0.28 GiB18.57 GiB0.03 GiB31±12.9%
ThinkingCap-Qwen3.6-27BQ4_K_L27.4B17.43 GiB0.28 GiB18.57 GiB0.03 GiB31±12.9%
Tess-4-27BQ4_K_L27.8B17.43 GiB0.28 GiB18.57 GiB0.03 GiB31±12.9%
Huihui-GLM-4.7-Flash-abliterated-57BMoEI1-IQ2_M57.3B17.14 GiB0.59 GiB18.57 GiB0.03 GiB111±37%
GLM-4-32B-0414-Korean-CultureI1-Q4_K_S32.6B17.41 GiB0.27 GiB18.57 GiB0.03 GiB31±12.9%
GLM-Z1-32B-0414Q4_K_S32.6B17.41 GiB0.27 GiB18.57 GiB0.03 GiB31±12.9%
GLM-4-32B-0414Q4_K_S32.6B17.41 GiB0.27 GiB18.57 GiB0.03 GiB31±12.9%
WizardCoder-Python-34B-V1.0I1-IQ4_XS33.7B16.83 GiB0.84 GiB18.57 GiB0.03 GiB31±12.9%
Phind-CodeLlama-34B-Python-v1I1-IQ4_XS33.7B16.83 GiB0.84 GiB18.57 GiB0.03 GiB31±12.9%
Phind-CodeLlama-34B-v2I1-IQ4_XS33.7B16.83 GiB0.84 GiB18.57 GiB0.03 GiB31±12.9%
Qwen3-Coder-30B-A3B-InstructMoEQ4_K_M30.5B17.35 GiB0.42 GiB18.57 GiB0.03 GiB117±37%
Qwen3-VL-30B-A3B-ThinkingMoEQ4_K_M31.1B17.35 GiB0.42 GiB18.57 GiB0.03 GiB117±37%
MiroThinker-v1.0-30BMoEQ4_K_M30.5B17.35 GiB0.42 GiB18.57 GiB0.03 GiB117±37%
Qwen3-30B-A3BMoEQ4_K_M30.5B17.35 GiB0.42 GiB18.57 GiB0.03 GiB117±37%
Qwen3-30B-A3B-Instruct-2507MoEQ4_K_M30.5B17.35 GiB0.42 GiB18.57 GiB0.03 GiB117±37%
Qwen3-30B-A3B-Thinking-2507MoEQ4_K_M30.5B17.35 GiB0.42 GiB18.57 GiB0.03 GiB117±37%
Pantheon-Proto-RP-1.8-30B-A3BMoEQ4_K_M30.5B17.35 GiB0.42 GiB18.57 GiB0.03 GiB117±37%
Olmo-3.1-32B-InstructQ4_K_S32.2B17.15 GiB0.52 GiB18.57 GiB0.03 GiB31±12.9%
Olmo-3.1-32B-ThinkQ4_K_S32.2B17.15 GiB0.52 GiB18.57 GiB0.03 GiB31±12.9%
Olmo-3-32B-ThinkQ4_K_S32.2B17.15 GiB0.52 GiB18.57 GiB0.03 GiB31±12.9%
Tongyi-DeepResearch-30B-A3BMoEQ4_K_M30.5B17.35 GiB0.42 GiB18.57 GiB0.03 GiB117±37%
North-Mini-Code-1.0MoEQ4_K_L30.5B17.58 GiB0.20 GiB18.56 GiB0.04 GiB126±37%
GLM-4.7-Flash-DerestrictedMoEI1-Q4_131.2B17.52 GiB0.23 GiB18.56 GiB0.04 GiB125±37%
Huihui-GLM-4.7-Flash-abliteratedMoEI1-Q4_131.2B17.52 GiB0.23 GiB18.56 GiB0.04 GiB125±37%
Qwen3.6-27B-uncensored-heretic-v2Q5_K_S27.4B17.40 GiB0.28 GiB18.54 GiB0.06 GiB31±12.9%
Qwen3.5-27B-Engineer-Deckard-GeminiI1-Q5_K_S27.7B17.40 GiB0.28 GiB18.54 GiB0.06 GiB31±12.9%
Qwen3.5-27B-HERETIC-Polaris-Advanced-Thinking-Alpha-uncensoredI1-Q5_K_S27.4B17.40 GiB0.28 GiB18.54 GiB0.06 GiB31±12.9%
Qwen3.5-27B-Deckard-PKD-Heretic-Uncensored-ThinkingI1-Q5_K_S27.4B17.40 GiB0.28 GiB18.54 GiB0.06 GiB31±12.9%
Qwen3.6-27B-Heretic2-ThinkingI1-Q5_K_S27.4B17.40 GiB0.28 GiB18.54 GiB0.06 GiB31±12.9%
Qwen3.6-27B-Uncensored-AggressiveI1-Q5_K_S27.4B17.40 GiB0.28 GiB18.54 GiB0.06 GiB31±12.9%
Qwen-3.5-Opus-GLM-27BI1-Q5_K_S26.9B17.40 GiB0.28 GiB18.54 GiB0.06 GiB31±12.9%
Qwen3.6-27B-abliteratedI1-Q5_K_S27.4B17.40 GiB0.28 GiB18.54 GiB0.06 GiB31±12.9%
KoQweopus-3.5-27B-experimentalI1-Q5_K_S27.8B17.40 GiB0.28 GiB18.54 GiB0.06 GiB31±12.9%
Webcoda-AI-27BI1-Q5_K_S27.4B17.40 GiB0.28 GiB18.54 GiB0.06 GiB31±12.9%
Qwen3.5-27B-imabari-v2I1-Q5_K_S27.8B17.40 GiB0.28 GiB18.54 GiB0.06 GiB31±12.9%
Qwen3.6-27B-AEON-Ultimate-Uncensored-BF16Q5_K_S27.4B17.40 GiB0.28 GiB18.54 GiB0.06 GiB31±12.9%
Huihui-Qwen3.5-27B-abliteratedI1-Q5_K_S27.8B17.40 GiB0.28 GiB18.54 GiB0.06 GiB31±12.9%
Qwen3.5-27B-uncensored-heretic-v1I1-Q5_K_S27.4B17.40 GiB0.28 GiB18.54 GiB0.06 GiB31±12.9%
Qwen3.5-27B-Unredacted-MAXI1-Q5_K_S27.4B17.40 GiB0.28 GiB18.54 GiB0.06 GiB31±12.9%
Qwen3.5-27B-hereticI1-Q5_K_S27.4B17.40 GiB0.28 GiB18.54 GiB0.06 GiB31±12.9%
Carnice-V2-27bI1-Q5_K_S27.4B17.40 GiB0.28 GiB18.54 GiB0.06 GiB31±12.9%
Qwen3.5-Queen-27BI1-Q5_K_S27.4B17.40 GiB0.28 GiB18.54 GiB0.06 GiB31±12.9%
GRaPE-2-ProI1-Q5_K_S27.8B17.40 GiB0.28 GiB18.54 GiB0.06 GiB31±12.9%
Huihui-Qwen3.6-27B-abliteratedQ5_K_S27.8B17.40 GiB0.28 GiB18.54 GiB0.06 GiB31±12.9%
Qwen3.5-27B-abliteratedQ5_K_S26.9B17.40 GiB0.28 GiB18.54 GiB0.06 GiB31±12.9%
Qwen3.5-27B-DerestrictedI1-Q5_K_S27.8B17.40 GiB0.28 GiB18.54 GiB0.06 GiB31±12.9%
ThinkingCap-Qwen3.6-27B-hereticQ5_K_S27.4B17.40 GiB0.28 GiB18.54 GiB0.06 GiB31±12.9%
Fara1.5-27BQ5_027.4B17.40 GiB0.28 GiB18.54 GiB0.06 GiB31±12.9%
Qwen3.6-27B-Omnimerge-v4Q5_K_S27.8B17.40 GiB0.28 GiB18.54 GiB0.06 GiB31±12.9%
Qwen3.6-27B-Claude-Opus-Reasoning-Distill-v2-hereticQ5_027.4B17.40 GiB0.28 GiB18.54 GiB0.06 GiB31±12.9%
Darwin-28B-REASONI1-Q5_K_S26.9B17.40 GiB0.28 GiB18.54 GiB0.06 GiB31±12.9%
Huihui-Qwen3.5-27B-Claude-4.6-Opus-abliteratedI1-Q5_K_S27.8B17.40 GiB0.28 GiB18.54 GiB0.06 GiB31±12.9%
Qwen3.5-27B-Claude-4.6-Opus-Reasoning-DistilledI1-Q5_K_S27.8B17.40 GiB0.28 GiB18.54 GiB0.06 GiB31±12.9%
●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 GeForce RTX 3080 Ti run?
1949 of 2118 indexed open-weight models fit a GeForce RTX 3080 Ti at 16,384 context with q4_0 KV cache, the largest being Gemma-4-Novelist-Eclipse-31B at IQ4_XS. That covers text, vision-language, image, video and speech models.
How much usable memory does a GeForce RTX 3080 Ti 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 GeForce RTX 3080 Ti fast for local AI?
Its memory bandwidth is 760 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.