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. 1450 of 2118 indexed models fit at 64K context with q4_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 1239vision language 113image 1video 12embedding 26audio tts 21audio asr 38

What fits at 64K context

largest quantization that fits, per model · 1450 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
Bonsai-8B-unpackedQ5_K_L8.2B5.94 GiB2.53 GiB9.30 GiB0.00 GiB63±12.9%
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%
INTELLECT-1-InstructI1-IQ4_NL10.2B5.50 GiB2.95 GiB9.30 GiB0.00 GiB63±12.9%
Cydonia-v1.3-Magnum-v4-22BI1-IQ1_S22.2B4.50 GiB3.94 GiB9.30 GiB0.00 GiB63±12.9%
Mistral-Small-22B-ArliAI-RPMax-v1.1I1-IQ1_S22.2B4.50 GiB3.94 GiB9.30 GiB0.00 GiB63±12.9%
magnum-v4-22bI1-IQ1_S22.2B4.50 GiB3.94 GiB9.30 GiB0.00 GiB63±12.9%
Codestral-22B-v0.1IQ1_S22.2B4.50 GiB3.94 GiB9.30 GiB0.00 GiB63±12.9%
Codestral-22B-v0.1-hfIQ1_S22.2B4.50 GiB3.94 GiB9.30 GiB0.00 GiB63±12.9%
Goetia-26B-A4B-v1.3-Absolute-Heretic-ARAMoEI1-IQ1_S25.8B7.72 GiB0.79 GiB9.29 GiB0.01 GiB63±12.9%
Frank-26B-A4BMoEI1-IQ1_S26.5B7.72 GiB0.79 GiB9.29 GiB0.01 GiB63±12.9%
G4-MeroMero-26B-A4B-it-uncensored-hereticMoEI1-IQ1_S25.8B7.72 GiB0.79 GiB9.29 GiB0.01 GiB63±12.9%
EVE-26b-XENO-HATMoEI1-IQ1_S25.8B7.72 GiB0.79 GiB9.29 GiB0.01 GiB63±12.9%
Gemma-4-26B-A4B-Animus-V14.1-FFT-hereticMoEI1-IQ1_S25.8B7.72 GiB0.79 GiB9.29 GiB0.01 GiB63±12.9%
gemma-4-26B-A4B-it-qat-q4_0-unquantized-hereticMoEI1-IQ1_S25.8B7.72 GiB0.79 GiB9.29 GiB0.01 GiB63±12.9%
Huihui-gemma-4-26B-A4B-it-qat-q4_0-unquantized-abliteratedMoEI1-IQ1_S26.5B7.72 GiB0.79 GiB9.29 GiB0.01 GiB63±12.9%
G4-MeroMero-26B-A4BMoEI1-IQ1_S25.8B7.72 GiB0.79 GiB9.29 GiB0.01 GiB63±12.9%
G4-Dark-Soul-26B-A4BMoEI1-IQ1_S25.8B7.72 GiB0.79 GiB9.29 GiB0.01 GiB63±12.9%
gemma-4-26B-A4B-it-local-abliterated-sota-internal-t34MoEI1-IQ1_S25.8B7.72 GiB0.79 GiB9.29 GiB0.01 GiB63±12.9%
gemma-4-26B-A4B-it-SOMPOA-heresyMoEI1-IQ1_S25.8B7.72 GiB0.79 GiB9.29 GiB0.01 GiB63±12.9%
gemma-4-26B-A4B-it-hereticMoEI1-IQ1_S25.8B7.72 GiB0.79 GiB9.29 GiB0.01 GiB63±12.9%
gemma-4-26B-A4B-it-abliterixMoEI1-IQ1_S25.8B7.72 GiB0.79 GiB9.29 GiB0.01 GiB63±12.9%
gemma-4-26B-A4B-it-heretic-ara-v2MoEI1-IQ1_S25.8B7.72 GiB0.79 GiB9.29 GiB0.01 GiB63±12.9%
Gemma-4-26B-A4B-it-heretic-antislopMoEI1-IQ1_S25.8B7.72 GiB0.79 GiB9.29 GiB0.01 GiB63±12.9%
gemma-4-26B-A4B-it-Claude-Opus-Distill-v2MoEI1-IQ1_S26.5B7.72 GiB0.79 GiB9.29 GiB0.01 GiB63±12.9%
gemma-4-26B-A4B-Heretic-StableMoEI1-IQ1_S25.8B7.72 GiB0.79 GiB9.29 GiB0.01 GiB63±12.9%
gemma-4-26B-A4B-it-Uncensored-MAXMoEI1-IQ1_S25.8B7.72 GiB0.79 GiB9.29 GiB0.01 GiB63±12.9%
gemma-4-26B-A4B-it-ultra-uncensored-hereticMoEI1-IQ1_S25.8B7.72 GiB0.79 GiB9.29 GiB0.01 GiB63±12.9%
gemma-4-26B-A4B-it-ara-abliteratedMoEI1-IQ1_S25.8B7.72 GiB0.79 GiB9.29 GiB0.01 GiB63±12.9%
Huihui-gemma-4-26B-A4B-it-abliteratedMoEI1-IQ1_S26.5B7.72 GiB0.79 GiB9.29 GiB0.01 GiB63±12.9%
Gemma-4-26B-A4B-AbliteratedMoEI1-IQ1_S25.8B7.72 GiB0.79 GiB9.29 GiB0.01 GiB63±12.9%
gemma4-26b-fiction-bf16MoEI1-IQ1_S25.8B7.72 GiB0.79 GiB9.29 GiB0.01 GiB63±12.9%
gemma-4-26B-A4B-it-heretic-araMoEI1-IQ1_S25.8B7.72 GiB0.79 GiB9.29 GiB0.01 GiB63±12.9%
dolphin-2.9.1-mixtral-1x22bMoEI1-IQ1_S22.2B4.49 GiB3.94 GiB9.29 GiB0.01 GiB37±37%
ERNIE-4.5-21B-A3B-ThinkingUD-IQ2_M21.8B7.48 GiB0.98 GiB9.29 GiB0.01 GiB63±12.9%
NuExtract-1.5IQ3_M3.8B1.73 GiB6.75 GiB9.28 GiB0.02 GiB63±12.9%
Phi-3.5-mini-instructIQ3_M3.8B1.73 GiB6.75 GiB9.28 GiB0.02 GiB63±12.9%
Phi-3.5-mini-instruct_UncensoredIQ3_M3.8B1.73 GiB6.75 GiB9.28 GiB0.02 GiB63±12.9%
Phi-3-mini-128k-instructIQ3_M3.8B1.73 GiB6.75 GiB9.28 GiB0.02 GiB63±12.9%
Phi-3-mini-4k-instructIQ3_M3.8B1.73 GiB6.75 GiB9.28 GiB0.02 GiB63±12.9%
AceReason-Nemotron-14BUD-IQ2_M14.8B5.06 GiB3.38 GiB9.28 GiB0.02 GiB63±12.9%
ERNIE-4.5-21B-A3B-PTUD-IQ2_M21.9B7.47 GiB0.98 GiB9.28 GiB0.02 GiB63±12.9%
legitus-instruct-v1I1-Q6_K8.1B6.16 GiB2.25 GiB9.28 GiB0.02 GiB63±12.9%
Apertus-8B-Instruct-2509I1-Q6_K8.1B6.16 GiB2.25 GiB9.28 GiB0.02 GiB63±12.9%
Ministral-3-8B-Instruct-2512-BF16Q5_K_M8.9B6.04 GiB2.39 GiB9.27 GiB0.03 GiB63±12.9%
Skywork-R1V3-38BIQ2_XXS38.4B8.41 GiB0.00 GiB9.27 GiB0.03 GiB63±12.9%
Qwen3-14BUD-IQ3_XXS14.8B5.60 GiB2.81 GiB9.27 GiB0.03 GiB63±12.9%
Phi-4-reasoningUD-IQ2_M14.7B4.90 GiB3.52 GiB9.27 GiB0.03 GiB63±12.9%
Phi-4-reasoning-plusUD-IQ2_M14.7B4.90 GiB3.52 GiB9.27 GiB0.03 GiB63±12.9%
Nanbeige4.1-3BF163.9B7.33 GiB1.13 GiB9.27 GiB0.03 GiB63±12.9%
HunyuanImage-2.1Q3_K_S17.5B8.42 GiB0.00 GiB9.27 GiB0.03 GiB63±12.9%
UncensoredLM-DeepSeek-R1-Distill-Qwen-14BQ2_K14.2B5.19 GiB3.23 GiB9.27 GiB0.03 GiB63±12.9%
LFM2-8B-A1BMoEQ8_08.3B8.26 GiB0.21 GiB9.27 GiB0.03 GiB166±37%
EVA-Yi-1.5-9B-32K-V1I1-Q6_K8.8B6.75 GiB1.69 GiB9.26 GiB0.04 GiB63±12.9%
Yi-Coder-9B-ChatQ6_K8.8B6.75 GiB1.69 GiB9.26 GiB0.04 GiB63±12.9%
Yi-1.5-9B-ChatQ6_K8.8B6.75 GiB1.69 GiB9.26 GiB0.04 GiB63±12.9%
Qwen3.5-14B-A3B-Claude-4.6-Opus-Reasoning-Distilled-reapMoEQ4_K_M14.1B8.10 GiB0.35 GiB9.25 GiB0.05 GiB184±37%
North-Mini-Code-1.0MoEIQ2_XXS30.5B7.93 GiB0.55 GiB9.25 GiB0.05 GiB185±37%
Ling-liteMoEQ3_K_S16.8B7.47 GiB0.98 GiB9.25 GiB0.05 GiB137±37%
Le-Chaton-Slim-23BMoEI1-IQ2_S23.3B6.61 GiB1.83 GiB9.25 GiB0.05 GiB86±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 run?
1450 of 2118 indexed open-weight models fit a GeForce RTX 3080 at 65,536 context with q4_0 KV cache, the largest being Bonsai-8B-unpacked at Q5_K_L. 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.