NVIDIA · consumer

GeForce RTX 3080

GeForce RTX 3080 has 10 GB of VRAM at 760 GB/s — about 9.30 GiB usable after driver and compositor overhead. 1545 of 2118 indexed models fit at 16K context with q8_0 KV.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed
Memory
10 GB
GDDR6X
Bandwidth
760 GB/s
320-bit bus
Tensor FP16
119 TF
dense
TDP
320 W
$699 MSRP
Context4K8K16K32K64K128K
KV cachef16q8_0q4_0quantizing the KV cache is a ~2× lever on the dominant term at long context
text 1330vision language 115image 2video 12audio asr 39audio tts 21embedding 26

What fits at 16K context

largest quantization that fits, per model · 1545 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
OmniAtlas-Qwen3-30B-A3BI1-IQ2_XS31.7B8.45 GiB0.00 GiB9.30 GiB0.00 GiB63±12.9%
Qwen3-Omni-30B-A3B-CaptionerI1-IQ2_XS31.7B8.45 GiB0.00 GiB9.30 GiB0.00 GiB63±12.9%
gemma-7bI1-Q4_K_S8.5B4.70 GiB3.72 GiB9.29 GiB0.01 GiB63±12.9%
Rocinante-XL-16B-v1IQ3_XS16.1B6.65 GiB1.79 GiB9.29 GiB0.01 GiB63±12.9%
gemma-3-12b-it-vl-Gemini-3-Pro-Preview-Heretic-Uncensored-ThinkingI1-Q5_K_S12.2B7.67 GiB0.78 GiB9.29 GiB0.01 GiB63±12.9%
gemma-3-12b-it-vl-Deepseek-v3.1-Heretic-Uncensored-ThinkingI1-Q5_K_S12.2B7.67 GiB0.78 GiB9.29 GiB0.01 GiB63±12.9%
gemma-3-12b-it-ultra-uncensored-hereticQ5_K_S12.2B7.67 GiB0.78 GiB9.29 GiB0.01 GiB63±12.9%
gemma-3-12b-it-vl-GLM-4.7-Flash-Heretic-Uncensored-ThinkingI1-Q5_K_S12.2B7.67 GiB0.78 GiB9.29 GiB0.01 GiB63±12.9%
Floppa-12B-Gemma3-UncensoredI1-Q5_K_S12.2B7.67 GiB0.78 GiB9.29 GiB0.01 GiB63±12.9%
gemma-3-12b-it-hereticI1-Q5_K_S12.2B7.67 GiB0.78 GiB9.29 GiB0.01 GiB63±12.9%
gemma-3-12b-it-abliteratedQ5_K_S12.2B7.67 GiB0.78 GiB9.29 GiB0.01 GiB63±12.9%
gemma-3-12b-it-abliterated-v2Q5_K_S11.8B7.67 GiB0.78 GiB9.29 GiB0.01 GiB63±12.9%
gemma-3-12b-itQ5_K_S12.2B7.67 GiB0.78 GiB9.29 GiB0.01 GiB63±12.9%
deepseek-math-7b-instructQ5_K_S6.9B4.48 GiB3.98 GiB9.29 GiB0.01 GiB63±12.9%
deepseek-llm-7b-chatQ5_06.9B4.48 GiB3.98 GiB9.29 GiB0.01 GiB63±12.9%
Janus-Pro-7BI1-Q5_K_S7.4B4.48 GiB3.98 GiB9.29 GiB0.01 GiB63±12.9%
deepseek-coder-7b-instruct-v1.5I1-Q5_K_S6.9B4.48 GiB3.98 GiB9.29 GiB0.01 GiB63±12.9%
NVIDIA-Nemotron-Nano-9B-v2Q5_K_M8.9B6.58 GiB1.86 GiB9.29 GiB0.01 GiB63±12.9%
openNemo-9B-abliteratedQ5_K_M8.9B6.58 GiB1.86 GiB9.29 GiB0.01 GiB63±12.9%
Tini-Cybersec-8B-A1BMoEQ8_08.5B8.39 GiB0.10 GiB9.29 GiB0.01 GiB175±37%
LFM2.5-8B-A1B-KO-SFTMoEQ8_08.5B8.39 GiB0.10 GiB9.29 GiB0.01 GiB175±37%
LFM2.5-8B-A1B-SOMPOA-heresyMoEQ8_08.5B8.39 GiB0.10 GiB9.29 GiB0.01 GiB175±37%
Huihui-LFM2.5-8B-A1B-abliteratedMoEQ8_08.5B8.39 GiB0.10 GiB9.29 GiB0.01 GiB175±37%
LFM2.5-8B-A1BMoEQ8_08.5B8.39 GiB0.10 GiB9.29 GiB0.01 GiB175±37%
Supertron2.1-8B-A1BMoEQ8_08.5B8.39 GiB0.10 GiB9.29 GiB0.01 GiB175±37%
LFM2.5-8B-A1B-hereticMoEQ8_08.5B8.39 GiB0.10 GiB9.29 GiB0.01 GiB175±37%
Qwen3-VL-8B-Instruct-HereticI1-IQ3_M8.8B7.26 GiB1.20 GiB9.29 GiB0.01 GiB63±12.9%
gemma-4-12B-it-uncensored-hereticQ4_K_S12.0B7.66 GiB0.78 GiB9.28 GiB0.02 GiB63±12.9%
Kimi-VL-A3B-Thinking-2506MoEIQ4_XS16.4B8.21 GiB0.25 GiB9.28 GiB0.02 GiB187±37%
Kimi-VL-A3B-InstructMoEIQ4_XS16.4B8.21 GiB0.25 GiB9.28 GiB0.02 GiB187±37%
EVA-abliterated-TIES-Qwen2.5-14BI1-Q3_K_M14.8B6.84 GiB1.59 GiB9.28 GiB0.02 GiB63±12.9%
Neuron-V1-14B-InstructI1-Q3_K_M14.8B6.84 GiB1.59 GiB9.28 GiB0.02 GiB63±12.9%
Ektome-Qwen2.5-Coder-14B-Instruct-PristinelyUncensoredI1-Q3_K_M14.8B6.84 GiB1.59 GiB9.28 GiB0.02 GiB63±12.9%
Qwen2.5-14B-Instruct-1M-abliteratedI1-Q3_K_M14.8B6.84 GiB1.59 GiB9.28 GiB0.02 GiB63±12.9%
DeepCoder-14B-PreviewQ3_K_M14.8B6.84 GiB1.59 GiB9.28 GiB0.02 GiB63±12.9%
Deepseeker-Kunou-Qwen2.5-14bI1-Q3_K_M14.8B6.84 GiB1.59 GiB9.28 GiB0.02 GiB63±12.9%
SuperNova-MediusQ3_K_M14.8B6.84 GiB1.59 GiB9.28 GiB0.02 GiB63±12.9%
14B-Qwen2.5-Kunou-v1I1-Q3_K_M14.8B6.84 GiB1.59 GiB9.28 GiB0.02 GiB63±12.9%
Sugoi-14B-Ultra-HFI1-Q3_K_M14.8B6.84 GiB1.59 GiB9.28 GiB0.02 GiB63±12.9%
Qwen2.5-14B-Instruct-abliterated-v2Q3_K_M14.8B6.84 GiB1.59 GiB9.28 GiB0.02 GiB63±12.9%
Qwen2.5-14B-Instruct-UncensoredQ3_K_M14.8B6.84 GiB1.59 GiB9.28 GiB0.02 GiB63±12.9%
Qwen2.5-Coder-14B-Instruct-abliteratedQ3_K_M14.8B6.84 GiB1.59 GiB9.28 GiB0.02 GiB63±12.9%
OpenCodeReasoning-Nemotron-14BQ3_K_M14.8B6.84 GiB1.59 GiB9.28 GiB0.02 GiB63±12.9%
DeepSeek-R1-Distill-Qwen-14B-abliterated-v2I1-Q3_K_M14.8B6.84 GiB1.59 GiB9.28 GiB0.02 GiB63±12.9%
C1-TachuI1-Q3_K_M14.8B6.84 GiB1.59 GiB9.28 GiB0.02 GiB63±12.9%
DeepSeek-R1-Distill-Qwen-14B-abliteratedI1-Q3_K_M14.8B6.84 GiB1.59 GiB9.28 GiB0.02 GiB63±12.9%
0x-liteQ3_K_M14.8B6.84 GiB1.59 GiB9.28 GiB0.02 GiB63±12.9%
Qwen2.5-Coder-14B-InstructQ3_K_M14.8B6.84 GiB1.59 GiB9.28 GiB0.02 GiB63±12.9%
Tessera-4I1-Q3_K_M14.8B6.84 GiB1.59 GiB9.28 GiB0.02 GiB63±12.9%
AceReason-Nemotron-14BQ3_K_M14.8B6.84 GiB1.59 GiB9.28 GiB0.02 GiB63±12.9%
Qwen2.5-14B-InstructQ3_K_M14.8B6.84 GiB1.59 GiB9.28 GiB0.02 GiB63±12.9%
FinetunedQwen14BQ3_K_M14.8B6.84 GiB1.59 GiB9.28 GiB0.02 GiB63±12.9%
Tessera-4.1I1-Q3_K_M14.8B6.84 GiB1.59 GiB9.28 GiB0.02 GiB63±12.9%
Qwen2.5-14B-Instruct-1MQ3_K_M14.8B6.84 GiB1.59 GiB9.28 GiB0.02 GiB63±12.9%
Qwen2.5-Coder-14BQ3_K_M14.8B6.84 GiB1.59 GiB9.28 GiB0.02 GiB63±12.9%
DeepSeek-R1-Distill-Qwen-14BQ3_K_M14.8B6.84 GiB1.59 GiB9.28 GiB0.02 GiB63±12.9%
UwU-14B-Math-v0.2I1-Q3_K_M14.8B6.84 GiB1.59 GiB9.28 GiB0.02 GiB63±12.9%
EVA-Qwen2.5-14B-v0.2I1-Q3_K_M14.8B6.84 GiB1.59 GiB9.28 GiB0.02 GiB63±12.9%
EVA-Qwen2.5-14B-v0.0I1-Q3_K_M14.8B6.84 GiB1.59 GiB9.28 GiB0.02 GiB63±12.9%
EVA-Qwen2.5-14B-v0.1I1-Q3_K_M14.8B6.84 GiB1.59 GiB9.28 GiB0.02 GiB63±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 run?
1545 of 2118 indexed open-weight models fit a GeForce RTX 3080 at 16,384 context with q8_0 KV cache, the largest being OmniAtlas-Qwen3-30B-A3B at I1-IQ2_XS. That covers text, vision-language, image, video and speech models.
How much usable memory does a GeForce RTX 3080 actually have?
Its nameplate is 10 GB, but about 9.30 GiB is available to a model once driver and compositor overhead is accounted for.
Is a GeForce RTX 3080 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.