NVIDIA · consumer

GeForce RTX 4080 Super

GeForce RTX 4080 Super has 16 GB of VRAM at 736 GB/s — about 14.88 GiB usable after driver and compositor overhead. 1688 of 2118 indexed models fit at 32K context with f16 KV.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed
Memory
16 GB
GDDR6X
Bandwidth
736 GB/s
256-bit bus
Tensor FP16
209 TF
dense
TDP
320 W
$999 MSRP
Context4K8K16K32K64K128K
KV cachef16q8_0q4_0quantizing the KV cache is a ~2× lever on the dominant term at long context
text 1435vision language 151audio asr 39video 15embedding 26audio tts 21image 1

What fits at 32K context

largest quantization that fits, per model · 1688 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
GLM-4.7-Flash-REAP-23B-A3BMoEQ4_K_S23.0B12.41 GiB1.65 GiB14.87 GiB0.01 GiB86±37%
Qwen3.6-27B-Heretic2-Uncensored-Finetune-ThinkingIQ3_M27.4B12.01 GiB2.00 GiB14.87 GiB0.01 GiB38±12.9%
DeepSeek-Coder-V2-Lite-BaseMoEI1-Q6_K15.7B13.10 GiB0.95 GiB14.86 GiB0.02 GiB99±37%
DeepSeek-Coder-V2-Lite-InstructMoEQ6_K15.7B13.10 GiB0.95 GiB14.86 GiB0.02 GiB99±37%
DeepSeek-V2-Lite-ChatMoEQ6_K15.7B13.10 GiB0.95 GiB14.86 GiB0.02 GiB99±37%
DeepSeek-V2-Lite-Chat-Uncensored-Unbiased-ReasonerMoEQ6_K15.7B13.10 GiB0.95 GiB14.86 GiB0.02 GiB99±37%
GLM-Z1-32B-0414UD-IQ3_XXS32.6B12.06 GiB1.91 GiB14.86 GiB0.02 GiB38±12.9%
GLM-4-32B-0414UD-IQ3_XXS32.6B12.06 GiB1.91 GiB14.86 GiB0.02 GiB38±12.9%
Huihui-Qwen3.5-35B-A3B-abliteratedMoEI1-IQ3_XS36.0B13.43 GiB0.63 GiB14.86 GiB0.02 GiB150±37%
Qwen3.5-35B-A3B-BaseMoEI1-IQ3_XS36.0B13.43 GiB0.63 GiB14.86 GiB0.02 GiB150±37%
Qwen3.5-35B-A3B-Claude-4.6-Opus-Reasoning-DistilledMoEI1-IQ3_XS36.0B13.43 GiB0.63 GiB14.86 GiB0.02 GiB150±37%
DeepSeek-V2-Lite-Chat-UncensoredMoEQ6_K15.7B13.09 GiB0.95 GiB14.85 GiB0.03 GiB99±37%
Phi-3.5-mini-instructQ4_K_S3.8B2.04 GiB12.00 GiB14.85 GiB0.03 GiB37±12.9%
EXAONE-4.0-32BQ2_K32.0B11.11 GiB2.84 GiB14.85 GiB0.03 GiB38±12.9%
Llama-3.2-8X3B-MOE-Dark-Champion-Instruct-uncensored-abliterated-18.4BMoEQ4_K_M18.4B10.54 GiB3.50 GiB14.84 GiB0.04 GiB45±37%
NuExtract-1.5Q4_K_S3.8B2.04 GiB12.00 GiB14.84 GiB0.04 GiB37±12.9%
Phi-3.5-mini-instructQ4_K_S3.8B2.04 GiB12.00 GiB14.84 GiB0.04 GiB37±12.9%
Phi-3.5-mini-instruct_UncensoredQ4_K_S3.8B2.04 GiB12.00 GiB14.84 GiB0.04 GiB37±12.9%
Phi-3-mini-128k-instructQ4_K_S3.8B2.04 GiB12.00 GiB14.84 GiB0.04 GiB37±12.9%
Phi-3-mini-4k-instructQ4_K_S3.8B2.04 GiB12.00 GiB14.84 GiB0.04 GiB37±12.9%
octo-netQ4_K_S3.8B2.04 GiB12.00 GiB14.84 GiB0.04 GiB37±12.9%
Qwen3.6-35B-A3B-REAM-160-ru-agentMoEQ4_K_M23.6B13.41 GiB0.63 GiB14.84 GiB0.04 GiB135±37%
Fallen-Gemma3-27B-v1IQ3_M27.4B11.69 GiB2.31 GiB14.84 GiB0.04 GiB38±12.9%
GLM-4.7-FlashMoEQ3_K_S31.2B12.38 GiB1.65 GiB14.84 GiB0.04 GiB92±37%
EVA-abliterated-TIES-Qwen2.5-14BI1-Q4_K_S14.8B7.98 GiB6.00 GiB14.83 GiB0.05 GiB38±12.9%
Neuron-V1-14B-InstructI1-Q4_K_S14.8B7.98 GiB6.00 GiB14.83 GiB0.05 GiB38±12.9%
Ektome-Qwen2.5-Coder-14B-Instruct-PristinelyUncensoredI1-Q4_K_S14.8B7.98 GiB6.00 GiB14.83 GiB0.05 GiB38±12.9%
Qwen2.5-14B-Instruct-1M-abliteratedI1-Q4_K_S14.8B7.98 GiB6.00 GiB14.83 GiB0.05 GiB38±12.9%
DeepCoder-14B-PreviewQ4_K_S14.8B7.98 GiB6.00 GiB14.83 GiB0.05 GiB38±12.9%
Deepseeker-Kunou-Qwen2.5-14bI1-Q4_K_S14.8B7.98 GiB6.00 GiB14.83 GiB0.05 GiB38±12.9%
SuperNova-MediusQ4_K_S14.8B7.98 GiB6.00 GiB14.83 GiB0.05 GiB38±12.9%
14B-Qwen2.5-Kunou-v1I1-Q4_K_S14.8B7.98 GiB6.00 GiB14.83 GiB0.05 GiB38±12.9%
Sugoi-14B-Ultra-HFI1-Q4_K_S14.8B7.98 GiB6.00 GiB14.83 GiB0.05 GiB38±12.9%
Qwen2.5-14B-Instruct-abliterated-v2Q4_K_S14.8B7.98 GiB6.00 GiB14.83 GiB0.05 GiB38±12.9%
Qwen2.5-14B-Instruct-UncensoredQ4_K_S14.8B7.98 GiB6.00 GiB14.83 GiB0.05 GiB38±12.9%
Qwen2.5-Coder-14B-Instruct-abliteratedQ4_K_S14.8B7.98 GiB6.00 GiB14.83 GiB0.05 GiB38±12.9%
OpenCodeReasoning-Nemotron-14BQ4_K_S14.8B7.98 GiB6.00 GiB14.83 GiB0.05 GiB38±12.9%
DeepSeek-R1-Distill-Qwen-14B-abliterated-v2I1-Q4_K_S14.8B7.98 GiB6.00 GiB14.83 GiB0.05 GiB38±12.9%
C1-TachuI1-Q4_K_S14.8B7.98 GiB6.00 GiB14.83 GiB0.05 GiB38±12.9%
DeepSeek-R1-Distill-Qwen-14B-abliteratedI1-Q4_K_S14.8B7.98 GiB6.00 GiB14.83 GiB0.05 GiB38±12.9%
0x-liteQ4_K_S14.8B7.98 GiB6.00 GiB14.83 GiB0.05 GiB38±12.9%
Qwen2.5-Coder-14B-InstructQ4_K_S14.8B7.98 GiB6.00 GiB14.83 GiB0.05 GiB38±12.9%
Tessera-4I1-Q4_K_S14.8B7.98 GiB6.00 GiB14.83 GiB0.05 GiB38±12.9%
AceReason-Nemotron-14BQ4_K_S14.8B7.98 GiB6.00 GiB14.83 GiB0.05 GiB38±12.9%
Qwen2.5-14B-InstructQ4_K_S14.8B7.98 GiB6.00 GiB14.83 GiB0.05 GiB38±12.9%
FinetunedQwen14BQ4_K_S14.8B7.98 GiB6.00 GiB14.83 GiB0.05 GiB38±12.9%
Tessera-4.1I1-Q4_K_S14.8B7.98 GiB6.00 GiB14.83 GiB0.05 GiB38±12.9%
Qwen2.5-14B-Instruct-1MQ4_K_S14.8B7.98 GiB6.00 GiB14.83 GiB0.05 GiB38±12.9%
Qwen2.5-Coder-14BQ4_K_S14.8B7.98 GiB6.00 GiB14.83 GiB0.05 GiB38±12.9%
DeepSeek-R1-Distill-Qwen-14BQ4_K_S14.8B7.98 GiB6.00 GiB14.83 GiB0.05 GiB38±12.9%
UwU-14B-Math-v0.2I1-Q4_K_S14.8B7.98 GiB6.00 GiB14.83 GiB0.05 GiB38±12.9%
EVA-Qwen2.5-14B-v0.2I1-Q4_K_S14.8B7.98 GiB6.00 GiB14.83 GiB0.05 GiB38±12.9%
EVA-Qwen2.5-14B-v0.0I1-Q4_K_S14.8B7.98 GiB6.00 GiB14.83 GiB0.05 GiB38±12.9%
EVA-Qwen2.5-14B-v0.1I1-Q4_K_S14.8B7.98 GiB6.00 GiB14.83 GiB0.05 GiB38±12.9%
oxy-1-smallQ4_K_S14.8B7.98 GiB6.00 GiB14.83 GiB0.05 GiB38±12.9%
Impish_QWEN_14B-1MI1-Q4_K_S14.8B7.98 GiB6.00 GiB14.83 GiB0.05 GiB38±12.9%
Salience-1.5-FlashMoEI1-IQ3_XXS31.1B11.04 GiB3.00 GiB14.83 GiB0.05 GiB68±37%
Huihui-Qwen3-VL-30B-A3B-Instruct-abliteratedMoEI1-IQ3_XXS31.1B11.04 GiB3.00 GiB14.83 GiB0.05 GiB68±37%
Qwen3-30B-A3B-Gemini-Pro-High-Reasoning-2507-ABLITERATED-UNCENSOREDMoEI1-IQ3_XXS30.5B11.04 GiB3.00 GiB14.83 GiB0.05 GiB68±37%
MiroThinker-v1.0-30BMoEI1-IQ3_XXS30.5B11.04 GiB3.00 GiB14.83 GiB0.05 GiB68±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.

Measured on this card

third-party benchmarks, aggregated
Workload◍MedianMiddle 50%Runs
Image generation18.72 it/s14.27–24.78349
Prompt processing7751.62 tok/s7294.24–8117.7818
Text generation185.93 tok/s157.67–186.326
Benchmarked· n=349

Aggregated from community-submitted runs, so the spread is wide by nature — it covers different models, resolutions, step counts and settings, not one controlled configuration. Read the middle 50% rather than the median alone. These figures are reproduced with attribution from vladmandic-sd-data-benchmark, which publishes no licence — so we display and link rather than redistribute them.

Questions people ask

What AI models can a GeForce RTX 4080 Super run?
1688 of 2118 indexed open-weight models fit a GeForce RTX 4080 Super at 32,768 context with f16 KV cache, the largest being GLM-4.7-Flash-REAP-23B-A3B at Q4_K_S. That covers text, vision-language, image, video and speech models.
How much usable memory does a GeForce RTX 4080 Super actually have?
Its nameplate is 16 GB, but about 14.88 GiB is available to a model once driver and compositor overhead is accounted for.
Is a GeForce RTX 4080 Super fast for local AI?
Its memory bandwidth is 736 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.