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. 1397 of 2118 indexed models fit at 128K context with q8_0 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
vision language 136text 1160video 15embedding 26audio tts 21audio asr 38image 1

What fits at 128K context

largest quantization that fits, per model · 1397 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
Qwen3.6-27B-Heretic2-Uncensored-Finetune-ThinkingIQ2_M27.4B9.77 GiB4.25 GiB14.88 GiB0.00 GiB38±12.9%
dolphin-2.9.3-mistral-7B-32kI1-Q6_K7.2B5.54 GiB8.50 GiB14.88 GiB0.00 GiB37±12.9%
Mistral-7B-v0.3Q6_K7.2B5.54 GiB8.50 GiB14.88 GiB0.00 GiB37±12.9%
Mistral-7B-Instruct-v0.3-ParasiteI1-Q6_K7.2B5.54 GiB8.50 GiB14.88 GiB0.00 GiB37±12.9%
Mistral-7B-Instruct-v0.3-JbliteratedI1-Q6_K7.2B5.54 GiB8.50 GiB14.88 GiB0.00 GiB37±12.9%
Mistral-7B-Instruct-v0.3Q6_K7.2B5.54 GiB8.50 GiB14.88 GiB0.00 GiB37±12.9%
Mistral-7B-v0.3-Chinese-ChatQ6_K7.2B5.54 GiB8.50 GiB14.88 GiB0.00 GiB37±12.9%
mistral-7b-v0.3-bnb-4bitQ6_K7.5B5.54 GiB8.50 GiB14.88 GiB0.00 GiB37±12.9%
Mathstral-7B-v0.1Q6_K7.2B5.54 GiB8.50 GiB14.88 GiB0.00 GiB37±12.9%
openchat-3.5-0106KV unresolvedQ6_K7.2B5.53 GiB8.50 GiB14.87 GiB0.01 GiB37±12.9%
dolphin-2.6-mistral-7bQ6_K7.2B5.53 GiB8.50 GiB14.87 GiB0.01 GiB37±12.9%
Silicon-Maid-7BKV unresolvedQ6_K7.2B5.53 GiB8.50 GiB14.87 GiB0.01 GiB37±12.9%
deepseek-coder-1.3b-instructQ8_01.3B1.33 GiB12.75 GiB14.87 GiB0.01 GiB37±12.9%
deepseek-coder-1.3b-baseQ8_01.3B1.33 GiB12.75 GiB14.87 GiB0.01 GiB37±12.9%
SciPhi-Self-RAG-Mistral-7B-32kKV unresolvedI1-Q6_K7.2B5.53 GiB8.50 GiB14.87 GiB0.01 GiB37±12.9%
dolphin-2.2.1-mistral-7bKV unresolvedI1-Q6_K7.2B5.53 GiB8.50 GiB14.87 GiB0.01 GiB37±12.9%
OpenChat-3.5-7B-Qwen-v2.0KV unresolvedI1-Q6_K7.2B5.53 GiB8.50 GiB14.87 GiB0.01 GiB37±12.9%
CapybaraHermes-2.5-Mistral-7BKV unresolvedQ6_K7.2B5.53 GiB8.50 GiB14.87 GiB0.01 GiB37±12.9%
dolphin-2.8-mistral-7b-v02Q6_K7.2B5.53 GiB8.50 GiB14.87 GiB0.01 GiB37±12.9%
openchat-3.5-1210KV unresolvedQ6_K7.2B5.53 GiB8.50 GiB14.87 GiB0.01 GiB37±12.9%
Mistral-7B-v0.2Q6_K7.2B5.53 GiB8.50 GiB14.87 GiB0.01 GiB37±12.9%
OpenHermes-2.5-Mistral-7BKV unresolvedQ6_K7.2B5.53 GiB8.50 GiB14.87 GiB0.01 GiB37±12.9%
Hermes-Trismegistus-Mistral-7BKV unresolvedQ6_K7.2B5.53 GiB8.50 GiB14.87 GiB0.01 GiB37±12.9%
Mistral-7B-OpenOrcaKV unresolvedQ6_K7.2B5.53 GiB8.50 GiB14.87 GiB0.01 GiB37±12.9%
dolphin-2.1-mistral-7bKV unresolvedQ6_K7.2B5.53 GiB8.50 GiB14.87 GiB0.01 GiB37±12.9%
OpenHermes-2-Mistral-7BKV unresolvedQ6_K7.2B5.53 GiB8.50 GiB14.87 GiB0.01 GiB37±12.9%
dolphin-2.6-mistral-7b-dpo-laserQ6_K7.2B5.53 GiB8.50 GiB14.87 GiB0.01 GiB37±12.9%
Mistral-7B-Instruct-v0.1KV unresolvedI1-Q6_K7.2B5.53 GiB8.50 GiB14.87 GiB0.01 GiB37±12.9%
Mistral-7B-Instruct-v0.2I1-Q6_K7.2B5.53 GiB8.50 GiB14.87 GiB0.01 GiB37±12.9%
ContextualKunoichi_KTO-7BI1-Q6_K7.2B5.53 GiB8.50 GiB14.87 GiB0.01 GiB37±12.9%
xLAM-7b-rI1-Q6_K7.2B5.53 GiB8.50 GiB14.87 GiB0.01 GiB37±12.9%
mistral-7b-uncensoredKV unresolvedQ6_K7.2B5.53 GiB8.50 GiB14.87 GiB0.01 GiB37±12.9%
MegaBeam-Mistral-7B-512kQ6_K7.2B5.53 GiB8.50 GiB14.87 GiB0.01 GiB37±12.9%
Yarn-Mistral-7b-128kKV unresolvedQ6_K7.2B5.53 GiB8.50 GiB14.87 GiB0.01 GiB37±12.9%
Ninja-v1-RP-WIPKV unresolvedI1-Q6_K7.2B5.53 GiB8.50 GiB14.87 GiB0.01 GiB37±12.9%
BioMistral-7BKV unresolvedQ6_K7.2B5.53 GiB8.50 GiB14.87 GiB0.01 GiB37±12.9%
Mistral-7B-Instruct-v0.2-code-ftKV unresolvedQ6_K7.2B5.53 GiB8.50 GiB14.87 GiB0.01 GiB37±12.9%
SpydazWeb_AI_CyberTron_Ultra_7bKV unresolvedQ6_K7.2B5.53 GiB8.50 GiB14.87 GiB0.01 GiB37±12.9%
MetaMath-Cybertron-StarlingKV unresolvedQ6_K7.2B5.53 GiB8.50 GiB14.87 GiB0.01 GiB37±12.9%
zephyr-7b-betaKV unresolvedQ6_K7.2B5.53 GiB8.50 GiB14.87 GiB0.01 GiB37±12.9%
japanese-stablelm-instruct-gamma-7bKV unresolvedQ6_K7.2B5.53 GiB8.50 GiB14.87 GiB0.01 GiB37±12.9%
Kunoichi-DPO-v2-7BKV unresolvedQ6_K7.2B5.53 GiB8.50 GiB14.87 GiB0.01 GiB37±12.9%
dolphin-2.0-mistral-7bKV unresolvedQ6_K7.2B5.53 GiB8.50 GiB14.87 GiB0.01 GiB37±12.9%
SciPhi-Mistral-7B-32kKV unresolvedQ6_K7.2B5.53 GiB8.50 GiB14.87 GiB0.01 GiB37±12.9%
Kimiko-Mistral-7BKV unresolvedQ6_K7.2B5.53 GiB8.50 GiB14.87 GiB0.01 GiB37±12.9%
next-8bI1-Q4_K_S8.2B4.47 GiB9.56 GiB14.87 GiB0.01 GiB37±12.9%
Supertron2-Reranker-8BI1-Q4_K_S8.8B4.47 GiB9.56 GiB14.87 GiB0.01 GiB37±12.9%
next-ocrI1-Q4_K_S8.8B4.47 GiB9.56 GiB14.87 GiB0.01 GiB37±12.9%
Qwen3-VL-8B-GLM-4.7-Flash-Heretic-Uncensored-ThinkingI1-Q4_K_S8.8B4.47 GiB9.56 GiB14.87 GiB0.01 GiB37±12.9%
Midas-FableAgent-8BI1-Q4_K_S8.2B4.47 GiB9.56 GiB14.87 GiB0.01 GiB37±12.9%
Qwen3-VL-8B-Heretic-1.3.0I1-Q4_K_S8.8B4.47 GiB9.56 GiB14.87 GiB0.01 GiB37±12.9%
Qwen3-VL-8B-Thinking-Unredacted-MAXI1-Q4_K_S8.8B4.47 GiB9.56 GiB14.87 GiB0.01 GiB37±12.9%
Qwen3-VL-8B-Instruct-Minecraft-MT-en-zhI1-Q4_K_S8.8B4.47 GiB9.56 GiB14.87 GiB0.01 GiB37±12.9%
Qwen-3-VL-8B-Instruct-hereticI1-Q4_K_S8.8B4.47 GiB9.56 GiB14.87 GiB0.01 GiB37±12.9%
Poe-8B-GLM5-Opus4.6-Sonnet4.5-Kimi-Grok-Gemini-3-pro-preview-HERETICI1-Q4_K_S8.8B4.47 GiB9.56 GiB14.87 GiB0.01 GiB37±12.9%
ToolCUA-8BI1-Q4_K_S8.8B4.47 GiB9.56 GiB14.87 GiB0.01 GiB37±12.9%
Qwen3-VL-8B-ThinkingQ4_K_S8.8B4.47 GiB9.56 GiB14.87 GiB0.01 GiB37±12.9%
Huihui-Qwen3-VL-8B-Instruct-abliteratedI1-Q4_K_S8.8B4.47 GiB9.56 GiB14.87 GiB0.01 GiB37±12.9%
Qwen3-VL-8B-Instruct-Unredacted-MAXQ4_K_S8.8B4.47 GiB9.56 GiB14.87 GiB0.01 GiB37±12.9%
Qwen3-VL-Reranker-8BI1-Q4_K_S8.8B4.47 GiB9.56 GiB14.87 GiB0.01 GiB37±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.

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?
1397 of 2118 indexed open-weight models fit a GeForce RTX 4080 Super at 131,072 context with q8_0 KV cache, the largest being Qwen3.6-27B-Heretic2-Uncensored-Finetune-Thinking at IQ2_M. 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.