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. 1817 of 2118 indexed models fit at 32K context with f16 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 1543vision language 170audio asr 39audio tts 21image 2video 16embedding 26

What fits at 32K context

largest quantization that fits, per model · 1817 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
Huihui-Qwen3-Coder-30B-A3B-Instruct-abliteratedMoEI1-Q3_K_L30.5B14.81 GiB3.00 GiB18.60 GiB0.00 GiB64±37%
Qwen3-Coder-30B-A3B-Instruct-RTPurboMoEI1-Q3_K_L30.5B14.81 GiB3.00 GiB18.60 GiB0.00 GiB64±37%
GRM-2.6-Plus-0628IQ4_NL27.8B15.73 GiB2.00 GiB18.59 GiB0.01 GiB31±12.9%
Skyfall-31B-v4.2-hereticI1-Q2_K31.4B10.92 GiB6.75 GiB18.59 GiB0.01 GiB31±12.9%
Skyfall-31B-v4.2I1-Q2_K31.4B10.92 GiB6.75 GiB18.59 GiB0.01 GiB31±12.9%
Llama-3.1-8BQ4_K_M8.0B13.75 GiB4.00 GiB18.59 GiB0.01 GiB31±12.9%
medgemma-27b-itI1-Q4_K_S28.8B14.60 GiB3.11 GiB18.59 GiB0.01 GiB31±12.9%
gemma-3-27b-it-abliterated-refined-visionI1-Q4_K_S27.4B14.60 GiB3.11 GiB18.59 GiB0.01 GiB31±12.9%
gemma-3-27b-it-abliteratedQ4_K_S27.4B14.60 GiB3.11 GiB18.59 GiB0.01 GiB31±12.9%
Nidum-Gemma-3-27B-it-UncensoredI1-Q4_K_S27.4B14.60 GiB3.11 GiB18.59 GiB0.01 GiB31±12.9%
gemma-3-27b-itQ4_K_S27.4B14.60 GiB3.11 GiB18.59 GiB0.01 GiB31±12.9%
AtomicGPT-gemma3-27bI1-Q4_K_S27.4B14.60 GiB3.11 GiB18.59 GiB0.01 GiB31±12.9%
Magistry-24B-v1.1IQ4_NL23.6B12.67 GiB5.00 GiB18.59 GiB0.01 GiB31±12.9%
Unbound-v1.12.0-27BI1-Q4_K_S27.4B14.60 GiB3.11 GiB18.59 GiB0.01 GiB31±12.9%
Mira-v1.12-Ties-27BI1-Q4_K_S27.4B14.60 GiB3.11 GiB18.59 GiB0.01 GiB31±12.9%
Medgamma27BI1-Q4_K_S27.0B14.60 GiB3.11 GiB18.59 GiB0.01 GiB31±12.9%
medgemma-27b-text-itQ4_K_S27.0B14.60 GiB3.11 GiB18.59 GiB0.01 GiB31±12.9%
t5-v1_1-xxlF324.8B17.74 GiB0.00 GiB18.59 GiB0.01 GiB31±12.9%
gemma-7bI1-IQ3_S8.5B3.71 GiB14.00 GiB18.58 GiB0.02 GiB31±12.9%
Pantheon-Reasoning-26B-A4B-1.1MoEQ4_K_M26.5B16.25 GiB1.54 GiB18.58 GiB0.02 GiB31±12.9%
Gemma-4-31B-Isometry-RPI1-Q2_K32.7B11.53 GiB6.17 GiB18.58 GiB0.02 GiB31±12.9%
Gemma-4-Dark-Gemistry-31BI1-Q2_K32.7B11.53 GiB6.17 GiB18.58 GiB0.02 GiB31±12.9%
Prosopon-31BI1-Q2_K32.7B11.53 GiB6.17 GiB18.58 GiB0.02 GiB31±12.9%
Gemma-4-Novelist-Eclipse-31BI1-Q2_K32.7B11.53 GiB6.17 GiB18.58 GiB0.02 GiB31±12.9%
Giftige-Blume-31B-v1-StyleSwapI1-Q2_K32.7B11.53 GiB6.17 GiB18.58 GiB0.02 GiB31±12.9%
G4-MeroMero-31B-StyleSwapI1-Q2_K32.7B11.53 GiB6.17 GiB18.58 GiB0.02 GiB31±12.9%
Gemma-4-31B-StyleTune-heretic-araI1-Q2_K32.7B11.53 GiB6.17 GiB18.58 GiB0.02 GiB31±12.9%
Pantheon-Reasoning-31B-1.1I1-Q2_K32.7B11.53 GiB6.17 GiB18.58 GiB0.02 GiB31±12.9%
Gemma-4-31B-StyleTuneI1-Q2_K32.7B11.53 GiB6.17 GiB18.58 GiB0.02 GiB31±12.9%
Barcenas-StyleTune-31B-FableI1-Q2_K32.1B11.53 GiB6.17 GiB18.58 GiB0.02 GiB31±12.9%
CallerIQ2_S32.8B9.67 GiB8.00 GiB18.57 GiB0.03 GiB31±12.9%
Dumpling-Qwen2.5-32BIQ2_S32.8B9.67 GiB8.00 GiB18.57 GiB0.03 GiB31±12.9%
OREAL-32BIQ2_S32.8B9.67 GiB8.00 GiB18.57 GiB0.03 GiB31±12.9%
QwQ-32B-Preview-abliterated-linear25I1-IQ2_S32.8B9.67 GiB8.00 GiB18.57 GiB0.03 GiB31±12.9%
openhands-lm-32b-v0.1I1-IQ2_S32.8B9.67 GiB8.00 GiB18.57 GiB0.03 GiB31±12.9%
Qwen2.5-Coder-32B-abliteratedI1-IQ2_S32.8B9.67 GiB8.00 GiB18.57 GiB0.03 GiB31±12.9%
m1-32bI1-IQ2_S32.8B9.67 GiB8.00 GiB18.57 GiB0.03 GiB31±12.9%
XMainframe-v2-Instruct-32bI1-IQ2_S32.8B9.67 GiB8.00 GiB18.57 GiB0.03 GiB31±12.9%
Qwen2.5-Coder-32B-Python-SpecialistI1-IQ2_S32.8B9.67 GiB8.00 GiB18.57 GiB0.03 GiB31±12.9%
Qwen2.5-32b-RP-InkI1-IQ2_S32.8B9.67 GiB8.00 GiB18.57 GiB0.03 GiB31±12.9%
LongWriter-Zero-32BIQ2_S32.8B9.67 GiB8.00 GiB18.57 GiB0.03 GiB31±12.9%
OpenCodeReasoning-Nemotron-32B-IOIIQ2_S32.8B9.67 GiB8.00 GiB18.57 GiB0.03 GiB31±12.9%
Qwen2.5-Coder-32B-Instruct-abliteratedIQ2_S32.8B9.67 GiB8.00 GiB18.57 GiB0.03 GiB31±12.9%
OlympicCoder-32BIQ2_S32.8B9.67 GiB8.00 GiB18.57 GiB0.03 GiB31±12.9%
OpenCodeReasoning-Nemotron-32BIQ2_S32.8B9.67 GiB8.00 GiB18.57 GiB0.03 GiB31±12.9%
OpenThinker-32BIQ2_S32.8B9.67 GiB8.00 GiB18.57 GiB0.03 GiB31±12.9%
QwQ-32B-ArliAI-RpR-v4IQ2_S32.8B9.67 GiB8.00 GiB18.57 GiB0.03 GiB31±12.9%
Qwen2.5-Coder-32B-InstructIQ2_S32.8B9.67 GiB8.00 GiB18.57 GiB0.03 GiB31±12.9%
Qwen2.5-Coder-32BIQ2_S32.8B9.67 GiB8.00 GiB18.57 GiB0.03 GiB31±12.9%
QwQ-32B-abliteratedIQ2_S32.8B9.67 GiB8.00 GiB18.57 GiB0.03 GiB31±12.9%
DeepSeek-R1-Distill-Qwen-32B-hereticI1-IQ2_S32.8B9.67 GiB8.00 GiB18.57 GiB0.03 GiB31±12.9%
InnoSpark-HPC-RM-32BI1-IQ2_S32.8B9.67 GiB8.00 GiB18.57 GiB0.03 GiB31±12.9%
OpenThinker2-32BIQ2_S32.8B9.67 GiB8.00 GiB18.57 GiB0.03 GiB31±12.9%
INTELLECT-2IQ2_S32.8B9.67 GiB8.00 GiB18.57 GiB0.03 GiB31±12.9%
Qwen2.5-32B-InstructIQ2_S32.8B9.67 GiB8.00 GiB18.57 GiB0.03 GiB31±12.9%
Qwen2.5-Coder-32B-Instruct-UncensoredI1-IQ2_S32.8B9.67 GiB8.00 GiB18.57 GiB0.03 GiB31±12.9%
QwQ-32B-PreviewIQ2_S32.8B9.67 GiB8.00 GiB18.57 GiB0.03 GiB31±12.9%
TinyR1-32B-PreviewIQ2_S32.8B9.67 GiB8.00 GiB18.57 GiB0.03 GiB31±12.9%
deepseek-r1-qwen-2.5-32B-ablatedIQ2_S32.8B9.67 GiB8.00 GiB18.57 GiB0.03 GiB31±12.9%
Rombos-LLM-V2.5-Qwen-32bIQ2_S32.8B9.67 GiB8.00 GiB18.57 GiB0.03 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?
1817 of 2118 indexed open-weight models fit a GeForce RTX 3080 Ti at 32,768 context with f16 KV cache, the largest being Huihui-Qwen3-Coder-30B-A3B-Instruct-abliterated at I1-Q3_K_L. 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.
GeForce RTX 3080 Ti — what AI models can it run locally? — ossmodeldb