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. 1869 of 2118 indexed models fit at 32K context with q8_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 1594vision language 171video 16embedding 26audio tts 21audio asr 39image 2

What fits at 32K context

largest quantization that fits, per model · 1869 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
QwQ-32B-Preview-abliterated-linear25I1-IQ3_S32.8B13.45 GiB4.25 GiB18.59 GiB0.01 GiB31±12.9%
openhands-lm-32b-v0.1I1-IQ3_S32.8B13.45 GiB4.25 GiB18.59 GiB0.01 GiB31±12.9%
Qwen2.5-Coder-32B-abliteratedI1-IQ3_S32.8B13.45 GiB4.25 GiB18.59 GiB0.01 GiB31±12.9%
m1-32bI1-IQ3_S32.8B13.45 GiB4.25 GiB18.59 GiB0.01 GiB31±12.9%
XMainframe-v2-Instruct-32bI1-IQ3_S32.8B13.45 GiB4.25 GiB18.59 GiB0.01 GiB31±12.9%
Qwen2.5-Coder-32B-Python-SpecialistI1-IQ3_S32.8B13.45 GiB4.25 GiB18.59 GiB0.01 GiB31±12.9%
Qwen2.5-32b-RP-InkI1-IQ3_S32.8B13.45 GiB4.25 GiB18.59 GiB0.01 GiB31±12.9%
Qwen2.5-Coder-32BIQ3_S32.8B13.45 GiB4.25 GiB18.59 GiB0.01 GiB31±12.9%
DeepSeek-R1-Distill-Qwen-32B-hereticI1-IQ3_S32.8B13.45 GiB4.25 GiB18.59 GiB0.01 GiB31±12.9%
InnoSpark-HPC-RM-32BI1-IQ3_S32.8B13.45 GiB4.25 GiB18.59 GiB0.01 GiB31±12.9%
Qwen2.5-Coder-32B-Instruct-UncensoredI1-IQ3_S32.8B13.45 GiB4.25 GiB18.59 GiB0.01 GiB31±12.9%
Qwen3.5-40B-Claude-4.6-Opus-Deckard-Heretic-Uncensored-ThinkingI1-IQ3_S39.5B16.14 GiB1.59 GiB18.59 GiB0.01 GiB31±12.9%
cogito-v1-preview-qwen-32BI1-IQ3_S32.8B13.44 GiB4.25 GiB18.59 GiB0.01 GiB31±12.9%
QwQ-32B-Snowdrop-v0I1-IQ3_S32.8B13.44 GiB4.25 GiB18.59 GiB0.01 GiB31±12.9%
DeepSeek-R1-Distill-Qwen-32B-UncensoredI1-IQ3_S32.8B13.44 GiB4.25 GiB18.59 GiB0.01 GiB31±12.9%
RoguePlanet-DeepSeek-R1-Qwen-32B-RPI1-IQ3_S32.8B13.44 GiB4.25 GiB18.59 GiB0.01 GiB31±12.9%
ERNIE-4.5-21B-A3B-ThinkingQ6_K21.8B16.84 GiB0.93 GiB18.59 GiB0.01 GiB31±12.9%
ERNIE-4.5-21B-A3B-PTQ6_K21.9B16.84 GiB0.93 GiB18.59 GiB0.01 GiB31±12.9%
t5-v1_1-xxlF324.8B17.74 GiB0.00 GiB18.59 GiB0.01 GiB31±12.9%
Nemotron-Cascade-2-30B-A3BMoEQ3_K_S31.6B16.94 GiB0.86 GiB18.59 GiB0.01 GiB111±37%
Qwen3-VL-32B-Instruct-ultra-uncensored-hereticI1-IQ3_S33.4B13.44 GiB4.25 GiB18.58 GiB0.02 GiB31±12.9%
Huihui-Qwen3-VL-32B-Instruct-abliteratedI1-IQ3_S33.4B13.44 GiB4.25 GiB18.58 GiB0.02 GiB31±12.9%
ColorGUI-32BI1-IQ3_S33.4B13.44 GiB4.25 GiB18.58 GiB0.02 GiB31±12.9%
Qwen3-32B-UncensoredI1-IQ3_S32.8B13.44 GiB4.25 GiB18.58 GiB0.02 GiB31±12.9%
Qwen3-32B-abliteratedI1-IQ3_S32.8B13.44 GiB4.25 GiB18.58 GiB0.02 GiB31±12.9%
AReaL-boba-2-32BI1-IQ3_S32.8B13.44 GiB4.25 GiB18.58 GiB0.02 GiB31±12.9%
Assistant_Pepe_32BI1-IQ3_S32.8B13.44 GiB4.25 GiB18.58 GiB0.02 GiB31±12.9%
Qwen3-VL-30B-A3B-ThinkingMoEIQ4_NL31.1B16.19 GiB1.59 GiB18.58 GiB0.02 GiB85±37%
MiroThinker-v1.0-30BMoEIQ4_NL30.5B16.19 GiB1.59 GiB18.58 GiB0.02 GiB85±37%
Qwen3-30B-A3BMoEIQ4_NL30.5B16.19 GiB1.59 GiB18.58 GiB0.02 GiB85±37%
Qwen3-30B-A3B-Instruct-2507MoEIQ4_NL30.5B16.19 GiB1.59 GiB18.58 GiB0.02 GiB85±37%
Qwen3-30B-A3B-Thinking-2507MoEIQ4_NL30.5B16.19 GiB1.59 GiB18.58 GiB0.02 GiB85±37%
Pantheon-Proto-RP-1.8-30B-A3BMoEIQ4_NL30.5B16.19 GiB1.59 GiB18.58 GiB0.02 GiB85±37%
Carnice-Qwen3.6-MoE-35B-A3BMoEI1-IQ4_XS36.0B17.44 GiB0.33 GiB18.58 GiB0.02 GiB150±37%
Qwen35B-Agent-R2-AbliteratedMoEI1-IQ4_XS34.7B17.44 GiB0.33 GiB18.58 GiB0.02 GiB150±37%
Qwen3.6-35B-A3B-Claude-4.6-Opus-Reasoning-DistilledMoEI1-IQ4_XS36.0B17.44 GiB0.33 GiB18.58 GiB0.02 GiB150±37%
Darwin-35B-A3B-OpusMoEI1-IQ4_XS36.0B17.44 GiB0.33 GiB18.58 GiB0.02 GiB150±37%
Qwen35B-Agent-R2MoEI1-IQ4_XS34.7B17.44 GiB0.33 GiB18.58 GiB0.02 GiB150±37%
Carnice-MoE-35B-A3BMoEI1-IQ4_XS36.0B17.44 GiB0.33 GiB18.58 GiB0.02 GiB150±37%
spoomplesmaxx-flash-35B-A3MoEI1-IQ4_XS35.1B17.44 GiB0.33 GiB18.58 GiB0.02 GiB150±37%
Huihui-Qwen3.6-35B-A3B-Claude-4.7-Opus-abliteratedMoEI1-IQ4_XS36.0B17.44 GiB0.33 GiB18.58 GiB0.02 GiB150±37%
Qwen3.6-35B-A3B-Uncensored-AggressiveMoEI1-IQ4_XS35.1B17.44 GiB0.33 GiB18.58 GiB0.02 GiB150±37%
WorldSim-Opus-3.6-35B-A3BMoEI1-IQ4_XS35.1B17.44 GiB0.33 GiB18.58 GiB0.02 GiB150±37%
Qwen3.6-35B-A3B-abliterated-MAXMoEI1-IQ4_XS35.1B17.44 GiB0.33 GiB18.58 GiB0.02 GiB150±37%
Huihui-Qwen3.6-35B-A3B-abliteratedMoEI1-IQ4_XS36.0B17.44 GiB0.33 GiB18.58 GiB0.02 GiB150±37%
Qwopus3.6-35B-A3B-v1MoEI1-IQ4_XS36.0B17.44 GiB0.33 GiB18.58 GiB0.02 GiB150±37%
Qwen3.6-35B-A3B-StyleTuneMoEI1-IQ4_XS35.1B17.44 GiB0.33 GiB18.58 GiB0.02 GiB150±37%
Qwen3.6-35B-A3B-abliteratedMoEI1-IQ4_XS35.1B17.44 GiB0.33 GiB18.58 GiB0.02 GiB150±37%
Qwen3.6-35B-A3B-abliterated-v4MoEIQ4_XS34.7B17.44 GiB0.33 GiB18.58 GiB0.02 GiB150±37%
0GM-1.0-35B-A3B-0427MoEI1-IQ4_XS36.0B17.44 GiB0.33 GiB18.58 GiB0.02 GiB150±37%
Qwen3.5-40B-RoughHouse-Claude-4.6-Opus-Polar-Deckard-Uncensored-Heretic-ThinkingI1-Q3_K_S39.5B16.12 GiB1.59 GiB18.58 GiB0.02 GiB31±12.9%
Qwen3.6-27B-A3B-CoderMoEI1-Q5_K_M26.7B17.44 GiB0.33 GiB18.58 GiB0.02 GiB127±37%
Tongyi-DeepResearch-30B-A3BMoEIQ4_NL30.5B16.19 GiB1.59 GiB18.58 GiB0.02 GiB85±37%
Yi-34B-200K-DARE-megamerge-v8IQ3_XS34.4B13.71 GiB3.98 GiB18.58 GiB0.02 GiB31±12.9%
Nous-Hermes-2-Yi-34BI1-IQ3_XS34.4B13.71 GiB3.98 GiB18.58 GiB0.02 GiB31±12.9%
NVIDIA-Nemotron-3-Nano-30B-A3B-BF16MoEIQ4_NL31.6B16.93 GiB0.86 GiB18.58 GiB0.02 GiB111±37%
GLM-4.7-Flash-DerestrictedMoEI1-Q4_K_M31.2B16.89 GiB0.88 GiB18.57 GiB0.03 GiB102±37%
Huihui-GLM-4.7-Flash-abliteratedMoEI1-Q4_K_M31.2B16.89 GiB0.88 GiB18.57 GiB0.03 GiB102±37%
GLM-4.7-Flash-Claude-Opus-4.5-High-Reasoning-DistillMoEQ4_K_M31.2B16.89 GiB0.88 GiB18.57 GiB0.03 GiB102±37%
Qwen3-Coder-30B-A3B-InstructMoEQ4_030.5B16.19 GiB1.59 GiB18.57 GiB0.03 GiB85±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.

Questions people ask

What AI models can a GeForce RTX 3080 Ti run?
1869 of 2118 indexed open-weight models fit a GeForce RTX 3080 Ti at 32,768 context with q8_0 KV cache, the largest being QwQ-32B-Preview-abliterated-linear25 at I1-IQ3_S. 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.