What AI models can a GeForce RTX 3080 Ti run?

1603 of 2118 indexed open-weight models fit a GeForce RTX 3080 Ti at 131,072 context with q8_0 KV cache, the largest being ultravox-v0_5-llama-3_2-1b at Q4_K_M. 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.

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. 1603 of 2118 indexed models fit at 128K 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

Context4K 8K 16K 32K 64K 128K

KV cachef16 q8_0 q4_0quantizing the KV cache is a ~2× lever on the dominant term at long context

text 1343vision language 157audio asr 39video 16embedding 26image 1audio tts 21

What fits at 128K context

largest quantization that fits, per model · 1603 of 2118 indexed

Model	Best quant	Params	Weights●	KV●	Total in memory◐	Headroom◐	tok/s◐
ultravox-v0_5-llama-3_2-1b	Q4_K_M	683M	0.75 GiB	17.00 GiB	18.60 GiB	0.00 GiB	31±12.9%
CycleGRPO-4B	F16	4.8B	8.23 GiB	9.56 GiB	18.60 GiB	0.00 GiB	31±12.9%
EXAONE-4.0-32B	Q3_K_S	32.0B	13.00 GiB	4.70 GiB	18.60 GiB	0.00 GiB	31±12.9%
Jan-v3-4B-base-instruct	BF16	4.4B	8.22 GiB	9.56 GiB	18.60 GiB	0.00 GiB	31±12.9%
Jan-code-4b	BF16	4.4B	8.22 GiB	9.56 GiB	18.60 GiB	0.00 GiB	31±12.9%
dolphin-2.9.2-Phi-3-MediumKV unresolved	IQ2_M	14.0B	4.45 GiB	13.28 GiB	18.59 GiB	0.01 GiB	31±12.9%
t5-v1_1-xxl	F32	4.8B	17.74 GiB	0.00 GiB	18.59 GiB	0.01 GiB	31±12.9%
EVA-abliterated-TIES-Qwen2.5-14B	I1-IQ2_M	14.8B	4.99 GiB	12.75 GiB	18.58 GiB	0.02 GiB	31±12.9%
Neuron-V1-14B-Instruct	I1-IQ2_M	14.8B	4.99 GiB	12.75 GiB	18.58 GiB	0.02 GiB	31±12.9%
Ektome-Qwen2.5-Coder-14B-Instruct-PristinelyUncensored	I1-IQ2_M	14.8B	4.99 GiB	12.75 GiB	18.58 GiB	0.02 GiB	31±12.9%
Qwen2.5-14B-Instruct-1M-abliterated	I1-IQ2_M	14.8B	4.99 GiB	12.75 GiB	18.58 GiB	0.02 GiB	31±12.9%
DeepCoder-14B-Preview	IQ2_M	14.8B	4.99 GiB	12.75 GiB	18.58 GiB	0.02 GiB	31±12.9%
Deepseeker-Kunou-Qwen2.5-14b	I1-IQ2_M	14.8B	4.99 GiB	12.75 GiB	18.58 GiB	0.02 GiB	31±12.9%
SuperNova-Medius	IQ2_M	14.8B	4.99 GiB	12.75 GiB	18.58 GiB	0.02 GiB	31±12.9%
14B-Qwen2.5-Kunou-v1	I1-IQ2_M	14.8B	4.99 GiB	12.75 GiB	18.58 GiB	0.02 GiB	31±12.9%
Sugoi-14B-Ultra-HF	I1-IQ2_M	14.8B	4.99 GiB	12.75 GiB	18.58 GiB	0.02 GiB	31±12.9%
Qwen2.5-Coder-14B-Instruct-abliterated	IQ2_M	14.8B	4.99 GiB	12.75 GiB	18.58 GiB	0.02 GiB	31±12.9%
OpenCodeReasoning-Nemotron-14B	IQ2_M	14.8B	4.99 GiB	12.75 GiB	18.58 GiB	0.02 GiB	31±12.9%
Qwen2.5-14B-Instruct	IQ2_M	14.8B	4.99 GiB	12.75 GiB	18.58 GiB	0.02 GiB	31±12.9%
DeepSeek-R1-Distill-Qwen-14B-abliterated-v2	I1-IQ2_M	14.8B	4.99 GiB	12.75 GiB	18.58 GiB	0.02 GiB	31±12.9%
C1-Tachu	I1-IQ2_M	14.8B	4.99 GiB	12.75 GiB	18.58 GiB	0.02 GiB	31±12.9%
DeepSeek-R1-Distill-Qwen-14B-abliterated	I1-IQ2_M	14.8B	4.99 GiB	12.75 GiB	18.58 GiB	0.02 GiB	31±12.9%
0x-lite	IQ2_M	14.8B	4.99 GiB	12.75 GiB	18.58 GiB	0.02 GiB	31±12.9%
Qwen2.5-Coder-14B-Instruct	IQ2_M	14.8B	4.99 GiB	12.75 GiB	18.58 GiB	0.02 GiB	31±12.9%
Tessera-4	I1-IQ2_M	14.8B	4.99 GiB	12.75 GiB	18.58 GiB	0.02 GiB	31±12.9%
Qwen2.5-14B-Instruct	IQ2_M	14.8B	4.99 GiB	12.75 GiB	18.58 GiB	0.02 GiB	31±12.9%
Tessera-4.1	I1-IQ2_M	14.8B	4.99 GiB	12.75 GiB	18.58 GiB	0.02 GiB	31±12.9%
Qwen2.5-14B-Instruct-1M	IQ2_M	14.8B	4.99 GiB	12.75 GiB	18.58 GiB	0.02 GiB	31±12.9%
Qwen2.5-Coder-14B	IQ2_M	14.8B	4.99 GiB	12.75 GiB	18.58 GiB	0.02 GiB	31±12.9%
AceReason-Nemotron-14B	I1-IQ2_M	14.8B	4.99 GiB	12.75 GiB	18.58 GiB	0.02 GiB	31±12.9%
DeepSeek-R1-Distill-Qwen-14B	IQ2_M	14.8B	4.99 GiB	12.75 GiB	18.58 GiB	0.02 GiB	31±12.9%
UwU-14B-Math-v0.2	I1-IQ2_M	14.8B	4.99 GiB	12.75 GiB	18.58 GiB	0.02 GiB	31±12.9%
EVA-Qwen2.5-14B-v0.2	I1-IQ2_M	14.8B	4.99 GiB	12.75 GiB	18.58 GiB	0.02 GiB	31±12.9%
EVA-Qwen2.5-14B-v0.0	I1-IQ2_M	14.8B	4.99 GiB	12.75 GiB	18.58 GiB	0.02 GiB	31±12.9%
EVA-Qwen2.5-14B-v0.1	I1-IQ2_M	14.8B	4.99 GiB	12.75 GiB	18.58 GiB	0.02 GiB	31±12.9%
oxy-1-small	IQ2_M	14.8B	4.99 GiB	12.75 GiB	18.58 GiB	0.02 GiB	31±12.9%
Impish_QWEN_14B-1M	I1-IQ2_M	14.8B	4.99 GiB	12.75 GiB	18.58 GiB	0.02 GiB	31±12.9%
Phi-3-medium-128k-instruct	Q2_K_S	14.0B	4.44 GiB	13.28 GiB	18.58 GiB	0.02 GiB	31±12.9%
Lamarck-14B-v0.7	I1-IQ2_M	14.8B	4.99 GiB	12.75 GiB	18.58 GiB	0.02 GiB	31±12.9%
QwenStock-14B	I1-IQ2_M	14.8B	4.99 GiB	12.75 GiB	18.58 GiB	0.02 GiB	31±12.9%
DeepSeek-R1-Distill-Qwen-14B-Uncensored	I1-IQ2_M	14.8B	4.99 GiB	12.75 GiB	18.58 GiB	0.02 GiB	31±12.9%
Smilodon-9B-v1	I1-Q5_K_M	10.2B	6.19 GiB	11.55 GiB	18.58 GiB	0.02 GiB	31±12.9%
bella-bartender-v2	I1-Q5_K_M	9.2B	6.19 GiB	11.55 GiB	18.58 GiB	0.02 GiB	31±12.9%
Gemma-The-Writer-9B-HERETIC-Uncensored-Abliterated	I1-Q5_K_M	9.2B	6.19 GiB	11.55 GiB	18.58 GiB	0.02 GiB	31±12.9%
Dirty-Muse-Writer-v01-Uncensored-Erotica-NSFW	I1-Q5_K_M	9.2B	6.19 GiB	11.55 GiB	18.58 GiB	0.02 GiB	31±12.9%
Gemma-2-9B-It-SPPO-Iter3	I1-Q5_K_M	9.2B	6.19 GiB	11.55 GiB	18.58 GiB	0.02 GiB	31±12.9%
Gemma-SEA-LION-v3-9B-IT	I1-Q5_K_M	9.2B	6.19 GiB	11.55 GiB	18.58 GiB	0.02 GiB	31±12.9%
G2-Darkest-Writer-Dirty-Shirley-9B-v2	I1-Q5_K_M	9.2B	6.19 GiB	11.55 GiB	18.58 GiB	0.02 GiB	31±12.9%
G2-Darkest-Writer-9B-v1	I1-Q5_K_M	9.2B	6.19 GiB	11.55 GiB	18.58 GiB	0.02 GiB	31±12.9%
Tiger-Gemma-9B-v3	I1-Q5_K_M	9.2B	6.19 GiB	11.55 GiB	18.58 GiB	0.02 GiB	31±12.9%
gemma-2-9b-it-abliterated	Q5_K_M	9.2B	6.19 GiB	11.55 GiB	18.58 GiB	0.02 GiB	31±12.9%
gemma-2-9b-it	Q5_K_M	9.2B	6.19 GiB	11.55 GiB	18.58 GiB	0.02 GiB	31±12.9%
Tiger-Gemma-9B-v1	Q5_K_M	9.2B	6.19 GiB	11.55 GiB	18.58 GiB	0.02 GiB	31±12.9%
magnum-v4-9b	Q5_K_M	9.2B	6.19 GiB	11.55 GiB	18.58 GiB	0.02 GiB	31±12.9%
gemma-2-9b	Q5_K_M	9.2B	6.19 GiB	11.55 GiB	18.58 GiB	0.02 GiB	31±12.9%
Llama3.2-24B-A3B-II-Dark-Champion-INSTRUCT-Heretic-Abliterated-UncensoredMoE	I1-Q4_K_M	18.0B	10.33 GiB	7.44 GiB	18.58 GiB	0.02 GiB	33±37%
Homunculus	Q4_K_M	12.5B	7.10 GiB	10.63 GiB	18.57 GiB	0.03 GiB	31±12.9%
Fimbulvetr-11B-v2	I1-Q3_K_M	10.7B	4.98 GiB	12.75 GiB	18.57 GiB	0.03 GiB	31±12.9%
ERNIE-21B-A3B-Thinking-Gemini-3-Pro-High-Reasoning-V2	I1-Q5_K_S	21.8B	14.02 GiB	3.72 GiB	18.56 GiB	0.04 GiB	31±12.9%
ERNIE-21B-A3B-Claude-4.5-High-OPUS-Thinking	I1-Q5_K_S	21.8B	14.02 GiB	3.72 GiB	18.56 GiB	0.04 GiB	31±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?: 1603 of 2118 indexed open-weight models fit a GeForce RTX 3080 Ti at 131,072 context with q8_0 KV cache, the largest being ultravox-v0_5-llama-3_2-1b at Q4_K_M. 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.