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. 1931 of 2118 indexed models fit at 32K 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
vision language 171text 1656audio asr 39video 16image 2audio tts 21embedding 26

What fits at 32K context

largest quantization that fits, per model · 1931 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
Gemma4-Gutenberg-31BIQ4_XS31.3B15.98 GiB1.74 GiB18.59 GiB0.01 GiB31±12.9%
gemma-4-31B-itIQ4_XS31.3B15.98 GiB1.74 GiB18.59 GiB0.01 GiB31±12.9%
Gemma4-Gutenberg-31B-HereticIQ4_XS31.3B15.98 GiB1.74 GiB18.59 GiB0.01 GiB31±12.9%
Equinox-31BIQ4_XS31.3B15.98 GiB1.74 GiB18.59 GiB0.01 GiB31±12.9%
gemma-4-31B-it-SDFT-Heretic-RPIQ4_XS30.7B15.98 GiB1.74 GiB18.59 GiB0.01 GiB31±12.9%
OLMo-2-1124-13B-InstructQ6_K_L13.7B10.72 GiB7.03 GiB18.59 GiB0.01 GiB31±12.9%
Qwen3-53B-A3B-2507-THINKING-TOTAL-RECALL-v2-MASTER-CODERMoEI1-IQ2_M53.0B16.32 GiB1.48 GiB18.59 GiB0.01 GiB82±37%
t5-v1_1-xxlF324.8B17.74 GiB0.00 GiB18.59 GiB0.01 GiB31±12.9%
CodeLlama-70b-Instruct-hfI1-IQ1_M69.0B14.85 GiB2.81 GiB18.58 GiB0.02 GiB31±12.9%
CodeLlama-70b-Python-hfI1-IQ1_M69.0B14.85 GiB2.81 GiB18.58 GiB0.02 GiB31±12.9%
Nous-Hermes-Llama2-70bI1-IQ1_M69.0B14.85 GiB2.81 GiB18.58 GiB0.02 GiB31±12.9%
Midnight-Miqu-70B-v1.5I1-IQ1_M69.0B14.85 GiB2.81 GiB18.58 GiB0.02 GiB31±12.9%
Pantheon-Reasoning-26B-A4B-1.1MoEQ5_K_S26.5B17.36 GiB0.43 GiB18.58 GiB0.02 GiB31±12.9%
Llama3.2-30B-A3B-II-Dark-Champion-INSTRUCT-Heretic-Abliterated-UncensoredMoEI1-Q4_K_S30.0B16.11 GiB1.65 GiB18.57 GiB0.03 GiB67±37%
Trinity-MiniMoEQ5_K_L26.1B17.59 GiB0.17 GiB18.56 GiB0.04 GiB127±37%
Skyfall-31B-v4.2-hereticI1-IQ4_XS31.4B15.74 GiB1.90 GiB18.56 GiB0.04 GiB31±12.9%
Skyfall-31B-v4.2I1-IQ4_XS31.4B15.74 GiB1.90 GiB18.56 GiB0.04 GiB31±12.9%
Gemma-3-27B-MeditronFOI1-Q4_128.8B16.81 GiB0.87 GiB18.56 GiB0.04 GiB31±12.9%
North-Mini-Code-1.0MoEQ4_K_M30.5B17.46 GiB0.32 GiB18.55 GiB0.05 GiB121±37%
magnum-v4-27bQ4_K_L27.2B15.77 GiB1.84 GiB18.55 GiB0.05 GiB31±12.9%
gemma-2-27b-itQ4_K_L27.2B15.77 GiB1.84 GiB18.55 GiB0.05 GiB31±12.9%
Huihui-Qwen3.5-35B-A3B-abliteratedMoEIQ4_XS36.0B17.57 GiB0.18 GiB18.55 GiB0.05 GiB161±37%
Qwen3.5-35B-A3B-BaseMoEIQ4_XS36.0B17.57 GiB0.18 GiB18.55 GiB0.05 GiB161±37%
GRM-2.6-Plus-0628Q4_K_M27.8B17.12 GiB0.56 GiB18.54 GiB0.06 GiB31±12.9%
GLM-4.7-Flash-hereticMoEQ4_K_M29.9B17.27 GiB0.46 GiB18.54 GiB0.06 GiB116±37%
Qwen3.5-40B-Claude-4.6-Opus-Deckard-Heretic-Uncensored-ThinkingI1-IQ3_M39.5B16.83 GiB0.84 GiB18.54 GiB0.06 GiB31±12.9%
Qwen3.5-40B-RoughHouse-Claude-4.6-Opus-Polar-Deckard-Uncensored-Heretic-ThinkingI1-IQ3_M39.5B16.83 GiB0.84 GiB18.54 GiB0.06 GiB31±12.9%
DeepSeek-R1-Distill-Llama-70BUD-IQ1_S70.6B14.79 GiB2.81 GiB18.53 GiB0.07 GiB31±12.9%
GPT-NeoX-20B-ErebusI1-IQ3_S20.6B8.35 GiB9.28 GiB18.52 GiB0.08 GiB31±12.9%
Seed-OSS-36B-Instruct-biprojected-norm-preserving-abliteratedI1-IQ3_M36.2B15.36 GiB2.25 GiB18.51 GiB0.09 GiB31±12.9%
Seed-OSS-36B-InstructIQ3_M36.2B15.36 GiB2.25 GiB18.51 GiB0.09 GiB31±12.9%
Hermes-4.3-36B-hereticI1-IQ3_M36.2B15.36 GiB2.25 GiB18.51 GiB0.09 GiB31±12.9%
Hermes-4.3-36BIQ3_M36.2B15.36 GiB2.25 GiB18.51 GiB0.09 GiB31±12.9%
spoomplesmaxx-v2.1-30BI1-Q4_K_S28.9B15.35 GiB2.25 GiB18.51 GiB0.09 GiB31±12.9%
Huihui-granite-4.1-30b-abliteratedI1-Q4_K_S28.9B15.35 GiB2.25 GiB18.51 GiB0.09 GiB31±12.9%
granite-4.1-30b-hereticI1-Q4_K_S28.9B15.35 GiB2.25 GiB18.51 GiB0.09 GiB31±12.9%
granite-4.1-30bQ4_K_S28.9B15.35 GiB2.25 GiB18.51 GiB0.09 GiB31±12.9%
Yi-34B-200K-DARE-megamerge-v8I1-Q3_K_M34.4B15.51 GiB2.11 GiB18.51 GiB0.09 GiB31±12.9%
dolphin-2.9.1-yi-1.5-34b-hereticQ3_K_M34.4B15.51 GiB2.11 GiB18.51 GiB0.09 GiB31±12.9%
dolphin-2.9.1-yi-1.5-34bI1-Q3_K_M34.4B15.51 GiB2.11 GiB18.51 GiB0.09 GiB31±12.9%
OrionStar-Yi-34B-Chat-LlamaI1-Q3_K_M34.4B15.51 GiB2.11 GiB18.51 GiB0.09 GiB31±12.9%
Yi-34B-200K-LlamafiedI1-Q3_K_M34.4B15.51 GiB2.11 GiB18.51 GiB0.09 GiB31±12.9%
Yi-1.5-34BQ3_K_M34.4B15.51 GiB2.11 GiB18.51 GiB0.09 GiB31±12.9%
Merged-RP-Stew-V2-34BI1-Q3_K_M34.4B15.51 GiB2.11 GiB18.51 GiB0.09 GiB31±12.9%
Capybara-Tess-Yi-34B-200KI1-Q3_K_M34.4B15.51 GiB2.11 GiB18.51 GiB0.09 GiB31±12.9%
Magistry-24B-v1.1Q5_K_L23.6B16.18 GiB1.41 GiB18.50 GiB0.10 GiB31±12.9%
Hermes-4-70BUD-IQ1_S70.6B14.77 GiB2.81 GiB18.50 GiB0.10 GiB31±12.9%
Llama-3.3-70B-InstructUD-IQ1_S70.6B14.77 GiB2.81 GiB18.50 GiB0.10 GiB31±12.9%
Aurora-Code-1MoEIQ4_XS34.7B17.51 GiB0.18 GiB18.49 GiB0.11 GiB161±37%
grug-35bMoEIQ4_XS35.1B17.51 GiB0.18 GiB18.49 GiB0.11 GiB161±37%
WorldSim-Opus-3.6-35B-A3BMoEIQ4_XS35.1B17.51 GiB0.18 GiB18.49 GiB0.11 GiB161±37%
Qwen3.6-35B-A3B-AnkoMoEIQ4_XS35.1B17.51 GiB0.18 GiB18.49 GiB0.11 GiB161±37%
KAT-Coder-V2.5-DevMoEIQ4_XS34.7B17.51 GiB0.18 GiB18.49 GiB0.11 GiB161±37%
Ornith-1.0-35BMoEIQ4_XS34.7B17.51 GiB0.18 GiB18.49 GiB0.11 GiB161±37%
Nex-N2-miniMoEIQ4_XS35.1B17.51 GiB0.18 GiB18.49 GiB0.11 GiB161±37%
EXAONE-4.5-33BIQ4_XS34.4B16.79 GiB0.80 GiB18.49 GiB0.11 GiB31±12.9%
Nous-Hermes-2-Yi-34BQ3_K_M34.4B15.49 GiB2.11 GiB18.49 GiB0.11 GiB31±12.9%
Nous-Capybara-limarpv3-34BQ3_K_M34.4B15.49 GiB2.11 GiB18.49 GiB0.11 GiB31±12.9%
Noromaid-v0.4-Mixtral-Instruct-8x7b-ZlossMoEQ2_K46.7B16.51 GiB1.13 GiB18.47 GiB0.13 GiB51±37%
TildeOpen-30B-Instruct-LVI1-IQ4_XS30.7B15.46 GiB2.11 GiB18.45 GiB0.15 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?
1931 of 2118 indexed open-weight models fit a GeForce RTX 3080 Ti at 32,768 context with q4_0 KV cache, the largest being Gemma4-Gutenberg-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.