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. 1954 of 2118 indexed models fit at 8K context with q4_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
Context4K8K16K32K64K128K
KV cachef16q8_0q4_0quantizing the KV cache is a ~2× lever on the dominant term at long context
text 1677vision language 173video 16image 2audio asr 39audio tts 21embedding 26

What fits at 8K context

largest quantization that fits, per model · 1954 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
GLM-4.7-FlashMoEQ4_131.2B17.67 GiB0.12 GiB18.59 GiB0.01 GiB130±37%
t5-v1_1-xxlF324.8B17.74 GiB0.00 GiB18.59 GiB0.01 GiB31±12.9%
GLM-4.7-Flash-hereticMoEQ4_129.9B17.65 GiB0.12 GiB18.57 GiB0.03 GiB130±37%
Fallen-Gemma3-27B-v1Q5_K_S27.4B17.48 GiB0.26 GiB18.57 GiB0.03 GiB31±12.9%
Qwen3-VL-30B-A3B-ThinkingMoEQ4_K_L31.1B17.57 GiB0.21 GiB18.57 GiB0.03 GiB126±37%
MiroThinker-v1.0-30BMoEQ4_K_L30.5B17.57 GiB0.21 GiB18.57 GiB0.03 GiB126±37%
Qwen3-30B-A3BMoEQ4_K_L30.5B17.57 GiB0.21 GiB18.57 GiB0.03 GiB126±37%
Qwen3-30B-A3B-Instruct-2507MoEQ4_K_L30.5B17.57 GiB0.21 GiB18.57 GiB0.03 GiB126±37%
Qwen3-30B-A3B-Thinking-2507MoEQ4_K_L30.5B17.57 GiB0.21 GiB18.57 GiB0.03 GiB126±37%
Pantheon-Proto-RP-1.8-30B-A3BMoEQ4_K_L30.5B17.57 GiB0.21 GiB18.57 GiB0.03 GiB126±37%
Tongyi-DeepResearch-30B-A3BMoEQ4_K_L30.5B17.57 GiB0.21 GiB18.57 GiB0.03 GiB126±37%
GLM-4.7-Flash-REAP-23B-A3B-absolute-heresyMoEI1-Q6_K23.0B17.61 GiB0.12 GiB18.54 GiB0.06 GiB114±37%
Qwen3.6-34B-80L-Fable-5-HereticIQ4_XS33.4B17.50 GiB0.18 GiB18.54 GiB0.06 GiB31±12.9%
Qwen3.6-27B-uncensored-heretic-v2NVFP427.4B17.51 GiB0.14 GiB18.52 GiB0.08 GiB31±12.9%
gemma-4-26B-A4B-itMoEUD-Q5_K_S26.5B17.56 GiB0.17 GiB18.51 GiB0.09 GiB31±12.9%
Gemma4-Gutenberg-31BQ4_K_S31.3B16.95 GiB0.68 GiB18.51 GiB0.09 GiB31±12.9%
gemma-4-31B-itQ4_K_S31.3B16.95 GiB0.68 GiB18.51 GiB0.09 GiB31±12.9%
Gemma4-Gutenberg-31B-HereticQ4_K_S31.3B16.95 GiB0.68 GiB18.51 GiB0.09 GiB31±12.9%
Equinox-31BQ4_K_S31.3B16.95 GiB0.68 GiB18.51 GiB0.09 GiB31±12.9%
gemma-4-31B-it-SDFT-Heretic-RPQ4_K_S30.7B16.95 GiB0.68 GiB18.51 GiB0.09 GiB31±12.9%
North-Mini-Code-1.0MoEQ4_K_L30.5B17.58 GiB0.15 GiB18.50 GiB0.10 GiB129±37%
Qwen3-42B-A3B-2507-Thinking-Abliterated-uncensored-TOTAL-RECALL-v2-Medium-MASTER-CODERMoEI1-IQ3_M42.4B17.41 GiB0.29 GiB18.50 GiB0.10 GiB123±37%
command-r-35b-writer-v2I1-IQ3_S35.0B14.77 GiB2.81 GiB18.49 GiB0.11 GiB31±12.9%
Holo3-35B-A3BMoEIQ4_XS35.1B17.64 GiB0.04 GiB18.49 GiB0.11 GiB172±37%
Qwen35B-Agent-R2-AbliteratedMoEIQ4_XS34.7B17.64 GiB0.04 GiB18.49 GiB0.11 GiB172±37%
Darwin-35B-A3B-OpusMoEIQ4_XS36.0B17.64 GiB0.04 GiB18.49 GiB0.11 GiB172±37%
spoomplesmaxx-flash-35B-A3MoEIQ4_XS35.1B17.64 GiB0.04 GiB18.49 GiB0.11 GiB172±37%
Holo-3.1-35B-A3BMoEIQ4_XS35.1B17.64 GiB0.04 GiB18.49 GiB0.11 GiB172±37%
grug-35b-v2MoEIQ4_XS35.1B17.64 GiB0.04 GiB18.49 GiB0.11 GiB172±37%
Qwen3.6-35B-A3B-abliterated-MAXMoEIQ4_XS35.1B17.64 GiB0.04 GiB18.49 GiB0.11 GiB172±37%
Qwen3.6-35B-A3B-StyleTuneMoEIQ4_XS35.1B17.64 GiB0.04 GiB18.49 GiB0.11 GiB172±37%
Qwen3.6-35B-A3B-Kimi-K2.6-Reasoning-DistilledMoEIQ4_XS36.0B17.64 GiB0.04 GiB18.49 GiB0.11 GiB172±37%
Qwen3.6-35B-A3B-Claude-4.7-Opus-Reasoning-DistilledMoEIQ4_XS36.0B17.64 GiB0.04 GiB18.49 GiB0.11 GiB172±37%
Qwen3.6-35B-A3B-abliteratedMoEIQ4_XS35.1B17.64 GiB0.04 GiB18.49 GiB0.11 GiB172±37%
Huihui-Qwen3.5-35B-A3B-Claude-4.6-Opus-abliteratedMoEIQ4_XS36.0B17.64 GiB0.04 GiB18.49 GiB0.11 GiB172±37%
solar-pro-preview-instructKV unresolvedQ6_K22.1B16.92 GiB0.70 GiB18.48 GiB0.12 GiB31±12.9%
Trinity-MiniMoEQ5_K_L26.1B17.59 GiB0.07 GiB18.46 GiB0.14 GiB133±37%
GLM-Z1-32B-0414-uncensored-heretic-v2Q4_K_S32.6B17.42 GiB0.13 GiB18.45 GiB0.15 GiB31±12.9%
GLM-4.6V-FlashBF1610.3B17.52 GiB0.09 GiB18.45 GiB0.15 GiB31±12.9%
GLM-Z1-9B-0414BF169.4B17.52 GiB0.09 GiB18.45 GiB0.15 GiB31±12.9%
GLM-4.1V-9B-ThinkingF1610.3B17.52 GiB0.09 GiB18.45 GiB0.15 GiB31±12.9%
GLM-4-9B-0414BF169.4B17.52 GiB0.09 GiB18.45 GiB0.15 GiB31±12.9%
Trinity-2-Codestral-22B-v0.2Q6_K_L22.2B17.09 GiB0.49 GiB18.44 GiB0.16 GiB31±12.9%
Mistral-Small-Drummer-22BQ6_K_L22.2B17.09 GiB0.49 GiB18.44 GiB0.16 GiB31±12.9%
Mistral-Small-Instruct-2409Q6_K_L22.2B17.09 GiB0.49 GiB18.44 GiB0.16 GiB31±12.9%
Mistral-Small-22B-ArliAI-RPMax-v1.1Q6_K_L22.2B17.09 GiB0.49 GiB18.44 GiB0.16 GiB31±12.9%
magnum-v4-22bQ6_K_L22.2B17.09 GiB0.49 GiB18.44 GiB0.16 GiB31±12.9%
GLM-4.7-Flash-DerestrictedMoEI1-Q4_131.2B17.52 GiB0.12 GiB18.44 GiB0.16 GiB131±37%
Huihui-GLM-4.7-Flash-abliteratedMoEI1-Q4_131.2B17.52 GiB0.12 GiB18.44 GiB0.16 GiB131±37%
Huihui-gemma-4-26B-A4B-it-abliteratedMoEUD-Q5_K_S26.5B17.48 GiB0.17 GiB18.44 GiB0.16 GiB31±12.9%
GLM-4-32B-0414-Korean-CultureI1-Q4_K_S32.6B17.41 GiB0.13 GiB18.44 GiB0.16 GiB31±12.9%
GLM-Z1-32B-0414Q4_K_S32.6B17.41 GiB0.13 GiB18.44 GiB0.16 GiB31±12.9%
GLM-4-32B-0414Q4_K_S32.6B17.41 GiB0.13 GiB18.44 GiB0.16 GiB31±12.9%
GRM-2.6-Plus-0628Q4_K_L27.8B17.43 GiB0.14 GiB18.43 GiB0.17 GiB31±12.9%
ThinkingCap-Qwen3.6-27BQ4_K_L27.4B17.43 GiB0.14 GiB18.43 GiB0.17 GiB31±12.9%
Tess-4-27BQ4_K_L27.8B17.43 GiB0.14 GiB18.43 GiB0.17 GiB31±12.9%
Qwen3-Coder-REAP-25B-A3BMoEQ5_124.9B17.43 GiB0.21 GiB18.43 GiB0.17 GiB115±37%
Olmo-3.1-32B-InstructQ4_K_S32.2B17.15 GiB0.38 GiB18.43 GiB0.17 GiB31±12.9%
Olmo-3.1-32B-ThinkQ4_K_S32.2B17.15 GiB0.38 GiB18.43 GiB0.17 GiB31±12.9%
Olmo-3-32B-ThinkQ4_K_S32.2B17.15 GiB0.38 GiB18.43 GiB0.17 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?
1954 of 2118 indexed open-weight models fit a GeForce RTX 3080 Ti at 8,192 context with q4_0 KV cache, the largest being GLM-4.7-Flash at Q4_1. 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.