NVIDIA · consumer

GeForce RTX 3080

GeForce RTX 3080 has 12 GB of VRAM at 912 GB/s — about 11.16 GiB usable after driver and compositor overhead. 1795 of 2118 indexed models fit at 4K context with q4_0 KV.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed
Memory
12 GB
GDDR6X
Bandwidth
912 GB/s
384-bit bus
Tensor FP16
123 TF
dense
TDP
350 W
$799 MSRP
Context4K8K16K32K64K128K
KV cachef16q8_0q4_0quantizing the KV cache is a ~2× lever on the dominant term at long context
video 14vision language 150text 1543audio asr 39audio tts 21image 2embedding 26

What fits at 4K context

largest quantization that fits, per model · 1795 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
Wan2.1-FLF2V-14B-720PQ4_116.4B10.32 GiB0.00 GiB11.16 GiB0.00 GiB62±12.9%
Magistral-Small-2509-VisionQ2_K_L24.0B10.06 GiB0.18 GiB11.16 GiB0.00 GiB62±12.9%
Wan2.1-I2V-14B-480PQ4_116.4B10.32 GiB0.00 GiB11.15 GiB0.01 GiB62±12.9%
Wan2.1-I2V-14B-720PQ4_116.4B10.32 GiB0.00 GiB11.15 GiB0.01 GiB62±12.9%
GLM-4.7-Flash-DerestrictedMoEI1-Q2_K31.2B10.28 GiB0.06 GiB11.15 GiB0.01 GiB245±37%
Huihui-GLM-4.7-Flash-abliteratedMoEI1-Q2_K31.2B10.28 GiB0.06 GiB11.15 GiB0.01 GiB245±37%
Moonlight-16B-A3B-InstructMoEQ5_016.0B10.30 GiB0.03 GiB11.15 GiB0.01 GiB207±37%
Qwen3.6-35B-A3B-REAM-192-hereticMoEIQ2_S27.0B10.32 GiB0.02 GiB11.15 GiB0.01 GiB288±37%
c4ai-command-r-08-2024IQ2_S32.3B10.05 GiB0.18 GiB11.14 GiB0.02 GiB63±12.9%
INTELLECT-1-InstructQ8_010.2B10.11 GiB0.18 GiB11.14 GiB0.02 GiB62±12.9%
reka-flash-3.1I1-Q3_K_M20.9B10.12 GiB0.15 GiB11.14 GiB0.02 GiB62±12.9%
reka-flash-3Q3_K_M20.9B10.12 GiB0.15 GiB11.14 GiB0.02 GiB62±12.9%
Qwen3.6-27B-Omnimerge-v4Q2_K27.8B10.20 GiB0.07 GiB11.13 GiB0.03 GiB62±12.9%
Fallen-Gemma3-27B-v1Q2_K_L27.4B10.10 GiB0.19 GiB11.13 GiB0.03 GiB62±12.9%
Qwen3-Coder-REAP-25B-A3BMoEQ3_K_S24.9B10.23 GiB0.11 GiB11.13 GiB0.03 GiB218±37%
Trinity-2-Codestral-22B-v0.2Q3_K_M22.2B10.02 GiB0.25 GiB11.12 GiB0.04 GiB62±12.9%
Cydonia-v1.3-Magnum-v4-22BI1-Q3_K_M22.2B10.02 GiB0.25 GiB11.12 GiB0.04 GiB62±12.9%
Mistral-Small-22B-ArliAI-RPMax-v1.1I1-Q3_K_M22.2B10.02 GiB0.25 GiB11.12 GiB0.04 GiB62±12.9%
Mistral-Small-Drummer-22BQ3_K_M22.2B10.02 GiB0.25 GiB11.12 GiB0.04 GiB62±12.9%
magnum-v4-22bI1-Q3_K_M22.2B10.02 GiB0.25 GiB11.12 GiB0.04 GiB62±12.9%
Codestral-22B-v0.1Q3_K22.2B10.02 GiB0.25 GiB11.12 GiB0.04 GiB62±12.9%
Codestral-22B-v0.1-hfQ3_K_M22.2B10.02 GiB0.25 GiB11.12 GiB0.04 GiB62±12.9%
medgemma-27b-itI1-IQ3_XXS28.8B9.98 GiB0.26 GiB11.12 GiB0.04 GiB62±12.9%
gemma-3-27b-it-abliterated-refined-visionI1-IQ3_XXS27.4B9.98 GiB0.26 GiB11.12 GiB0.04 GiB62±12.9%
Nidum-Gemma-3-27B-it-UncensoredI1-IQ3_XXS27.4B9.98 GiB0.26 GiB11.12 GiB0.04 GiB62±12.9%
gemma-3-27b-it-abliteratedIQ3_XXS27.4B9.98 GiB0.26 GiB11.12 GiB0.04 GiB62±12.9%
AtomicGPT-gemma3-27bI1-IQ3_XXS27.4B9.98 GiB0.26 GiB11.12 GiB0.04 GiB62±12.9%
Unbound-v1.12.0-27BI1-IQ3_XXS27.4B9.98 GiB0.26 GiB11.12 GiB0.04 GiB62±12.9%
Mira-v1.12-Ties-27BI1-IQ3_XXS27.4B9.98 GiB0.26 GiB11.12 GiB0.04 GiB62±12.9%
gemma-3-27b-itIQ3_XXS27.4B9.98 GiB0.26 GiB11.12 GiB0.04 GiB62±12.9%
Medgamma27BI1-IQ3_XXS27.0B9.98 GiB0.26 GiB11.12 GiB0.04 GiB62±12.9%
dolphin-2.9.1-mixtral-1x22bMoEI1-Q3_K_M22.2B10.01 GiB0.25 GiB11.12 GiB0.04 GiB36±37%
gemma-4-26B-A4B-itMoEQ2_K26.5B10.20 GiB0.13 GiB11.11 GiB0.05 GiB62±12.9%
Ornith-1.0-35BMoEUD-IQ1_M34.7B10.29 GiB0.02 GiB11.11 GiB0.05 GiB312±37%
OLMo-2-1124-13B-InstructQ5_K_L13.7B9.39 GiB0.88 GiB11.11 GiB0.05 GiB62±12.9%
gemma-4-A4B-98e-v6-coder-itMoEQ3_K_L20.5B10.19 GiB0.13 GiB11.11 GiB0.05 GiB62±12.9%
gemma-4-A4B-98e-v7-coder-itMoEQ3_K_L20.5B10.19 GiB0.13 GiB11.11 GiB0.05 GiB62±12.9%
gemma-4-A4B-98e-v7-coderx-itMoEQ3_K_L20.5B10.19 GiB0.13 GiB11.11 GiB0.05 GiB62±12.9%
GLM-4.7-FlashMoEUD-IQ2_M31.2B10.24 GiB0.06 GiB11.10 GiB0.06 GiB246±37%
Wan2.2-Distill-ModelsQ5_114.3B10.27 GiB0.00 GiB11.10 GiB0.06 GiB62±12.9%
Bernini-RQ5_114.3B10.26 GiB0.00 GiB11.10 GiB0.06 GiB62±12.9%
SkyReels-V2-DF-14B-540PQ5_114.3B10.27 GiB0.00 GiB11.10 GiB0.06 GiB62±12.9%
internlm2-math-plus-20bI1-IQ4_XS19.9B10.03 GiB0.21 GiB11.10 GiB0.06 GiB62±12.9%
Skyfall-31B-v4.2-hereticI1-IQ2_M31.4B9.94 GiB0.24 GiB11.10 GiB0.06 GiB63±12.9%
Skyfall-31B-v4.2I1-IQ2_M31.4B9.94 GiB0.24 GiB11.10 GiB0.06 GiB63±12.9%
GLM-Z1-Rumination-32B-0414IQ2_S33.1B9.93 GiB0.27 GiB11.09 GiB0.07 GiB63±12.9%
InternVL3_5-14BQ5_K_L15.1B10.24 GiB0.00 GiB11.08 GiB0.08 GiB62±12.9%
Delphi-25B-SimpleRL-MathI1-IQ3_XXS25.0B9.03 GiB1.18 GiB11.08 GiB0.08 GiB63±12.9%
Darwin-35B-A3B-OpusMoEIQ2_S36.0B10.25 GiB0.02 GiB11.08 GiB0.08 GiB313±37%
Aurora-Code-1MoEIQ2_S34.7B10.25 GiB0.02 GiB11.08 GiB0.08 GiB313±37%
grug-35b-v2MoEIQ2_S35.1B10.25 GiB0.02 GiB11.08 GiB0.08 GiB313±37%
grug-35bMoEIQ2_S35.1B10.25 GiB0.02 GiB11.08 GiB0.08 GiB313±37%
WorldSim-Opus-3.6-35B-A3BMoEIQ2_S35.1B10.25 GiB0.02 GiB11.08 GiB0.08 GiB313±37%
Qwen3.6-35B-A3B-AnkoMoEIQ2_S35.1B10.25 GiB0.02 GiB11.08 GiB0.08 GiB313±37%
KAT-Coder-V2.5-DevMoEIQ2_S34.7B10.25 GiB0.02 GiB11.08 GiB0.08 GiB313±37%
Nex-N2-miniMoEIQ2_S35.1B10.25 GiB0.02 GiB11.08 GiB0.08 GiB313±37%
dolphin-2.6-mixtral-8x7bMoEI1-IQ1_M46.7B10.10 GiB0.14 GiB11.08 GiB0.08 GiB112±37%
xLAM-8x7b-rMoEIQ1_M46.7B10.10 GiB0.14 GiB11.08 GiB0.08 GiB112±37%
Trinity-MiniMoEIQ3_XS26.1B10.23 GiB0.05 GiB11.07 GiB0.09 GiB247±37%
Qwen3.5-40B-Claude-4.6-Opus-Deckard-Heretic-Uncensored-ThinkingI1-IQ2_XXS39.5B10.10 GiB0.11 GiB11.07 GiB0.09 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?
1795 of 2118 indexed open-weight models fit a GeForce RTX 3080 at 4,096 context with q4_0 KV cache, the largest being Wan2.1-FLF2V-14B-720P at Q4_1. That covers text, vision-language, image, video and speech models.
How much usable memory does a GeForce RTX 3080 actually have?
Its nameplate is 12 GB, but about 11.16 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 912 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.