NVIDIA · consumer

GeForce RTX 2070 SUPER

GeForce RTX 2070 SUPER has 8 GB of VRAM at 448 GB/s — about 7.44 GiB usable after driver and compositor overhead. 1357 of 2118 indexed models fit at 16K context with q8_0 KV.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed
Memory
8 GB
GDDR6
Bandwidth
448 GB/s
256-bit bus
Tensor FP16
73 TF
dense
TDP
215 W
$499 MSRP
Context4K8K16K32K64K128K
KV cachef16q8_0q4_0quantizing the KV cache is a ~2× lever on the dominant term at long context
text 1162vision language 100video 8image 2embedding 26audio tts 21audio asr 38

What fits at 16K context

largest quantization that fits, per model · 1357 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
dolphin-2.9.3-mistral-7B-32kI1-Q6_K7.2B5.54 GiB1.06 GiB7.44 GiB0.00 GiB48±12.9%
Mistral-7B-v0.3Q6_K7.2B5.54 GiB1.06 GiB7.44 GiB0.00 GiB48±12.9%
Mistral-7B-Instruct-v0.3-ParasiteI1-Q6_K7.2B5.54 GiB1.06 GiB7.44 GiB0.00 GiB48±12.9%
Mistral-7B-Instruct-v0.3-JbliteratedI1-Q6_K7.2B5.54 GiB1.06 GiB7.44 GiB0.00 GiB48±12.9%
Mistral-7B-Instruct-v0.3Q6_K7.2B5.54 GiB1.06 GiB7.44 GiB0.00 GiB48±12.9%
Mistral-7B-v0.3-Chinese-ChatQ6_K7.2B5.54 GiB1.06 GiB7.44 GiB0.00 GiB48±12.9%
mistral-7b-v0.3-bnb-4bitQ6_K7.5B5.54 GiB1.06 GiB7.44 GiB0.00 GiB48±12.9%
Mathstral-7B-v0.1Q6_K7.2B5.54 GiB1.06 GiB7.44 GiB0.00 GiB48±12.9%
OmniAtlas-Qwen3-30B-A3BI1-IQ1_M31.7B6.59 GiB0.00 GiB7.44 GiB0.00 GiB48±12.9%
Qwen3-Omni-30B-A3B-CaptionerI1-IQ1_M31.7B6.59 GiB0.00 GiB7.44 GiB0.00 GiB48±12.9%
Qwythos-9B-v2Q4_K_L9.7B6.34 GiB0.27 GiB7.44 GiB0.00 GiB48±12.9%
Tess-4-9BQ4_K_L9.7B6.34 GiB0.27 GiB7.44 GiB0.00 GiB48±12.9%
Qwen3.5-9B-BaseQ5_19.7B6.33 GiB0.27 GiB7.44 GiB0.00 GiB48±12.9%
Teuken-7B-instruct-research-v0.4Q6_K_L7.5B6.33 GiB0.27 GiB7.43 GiB0.01 GiB48±12.9%
deepseek-coder-6.7B-kexerI1-Q2_K6.7B2.36 GiB4.25 GiB7.43 GiB0.01 GiB48±12.9%
Magicoder-S-DS-6.7BI1-Q2_K6.7B2.36 GiB4.25 GiB7.43 GiB0.01 GiB48±12.9%
deepseek-coder-6.7b-baseI1-Q2_K6.7B2.36 GiB4.25 GiB7.43 GiB0.01 GiB48±12.9%
openchat-3.5-0106KV unresolvedQ6_K7.2B5.53 GiB1.06 GiB7.43 GiB0.01 GiB48±12.9%
dolphin-2.6-mistral-7bQ6_K7.2B5.53 GiB1.06 GiB7.43 GiB0.01 GiB48±12.9%
Silicon-Maid-7BKV unresolvedQ6_K7.2B5.53 GiB1.06 GiB7.43 GiB0.01 GiB48±12.9%
SciPhi-Self-RAG-Mistral-7B-32kKV unresolvedI1-Q6_K7.2B5.53 GiB1.06 GiB7.43 GiB0.01 GiB48±12.9%
dolphin-2.2.1-mistral-7bKV unresolvedI1-Q6_K7.2B5.53 GiB1.06 GiB7.43 GiB0.01 GiB48±12.9%
OpenChat-3.5-7B-Qwen-v2.0KV unresolvedI1-Q6_K7.2B5.53 GiB1.06 GiB7.43 GiB0.01 GiB48±12.9%
CapybaraHermes-2.5-Mistral-7BKV unresolvedQ6_K7.2B5.53 GiB1.06 GiB7.43 GiB0.01 GiB48±12.9%
dolphin-2.8-mistral-7b-v02Q6_K7.2B5.53 GiB1.06 GiB7.43 GiB0.01 GiB48±12.9%
openchat-3.5-1210KV unresolvedQ6_K7.2B5.53 GiB1.06 GiB7.43 GiB0.01 GiB48±12.9%
Mistral-7B-v0.2Q6_K7.2B5.53 GiB1.06 GiB7.43 GiB0.01 GiB48±12.9%
OpenHermes-2.5-Mistral-7BKV unresolvedQ6_K7.2B5.53 GiB1.06 GiB7.43 GiB0.01 GiB48±12.9%
Hermes-Trismegistus-Mistral-7BKV unresolvedQ6_K7.2B5.53 GiB1.06 GiB7.43 GiB0.01 GiB48±12.9%
Mistral-7B-OpenOrcaKV unresolvedQ6_K7.2B5.53 GiB1.06 GiB7.43 GiB0.01 GiB48±12.9%
dolphin-2.1-mistral-7bKV unresolvedQ6_K7.2B5.53 GiB1.06 GiB7.43 GiB0.01 GiB48±12.9%
OpenHermes-2-Mistral-7BKV unresolvedQ6_K7.2B5.53 GiB1.06 GiB7.43 GiB0.01 GiB48±12.9%
dolphin-2.6-mistral-7b-dpo-laserQ6_K7.2B5.53 GiB1.06 GiB7.43 GiB0.01 GiB48±12.9%
Mistral-7B-Instruct-v0.1KV unresolvedI1-Q6_K7.2B5.53 GiB1.06 GiB7.43 GiB0.01 GiB48±12.9%
Mistral-7B-Instruct-v0.2I1-Q6_K7.2B5.53 GiB1.06 GiB7.43 GiB0.01 GiB48±12.9%
ContextualKunoichi_KTO-7BI1-Q6_K7.2B5.53 GiB1.06 GiB7.43 GiB0.01 GiB48±12.9%
xLAM-7b-rI1-Q6_K7.2B5.53 GiB1.06 GiB7.43 GiB0.01 GiB48±12.9%
mistral-7b-uncensoredKV unresolvedQ6_K7.2B5.53 GiB1.06 GiB7.43 GiB0.01 GiB48±12.9%
MegaBeam-Mistral-7B-512kQ6_K7.2B5.53 GiB1.06 GiB7.43 GiB0.01 GiB48±12.9%
Yarn-Mistral-7b-128kKV unresolvedQ6_K7.2B5.53 GiB1.06 GiB7.43 GiB0.01 GiB48±12.9%
Ninja-v1-RP-WIPKV unresolvedI1-Q6_K7.2B5.53 GiB1.06 GiB7.43 GiB0.01 GiB48±12.9%
BioMistral-7BKV unresolvedQ6_K7.2B5.53 GiB1.06 GiB7.43 GiB0.01 GiB48±12.9%
Mistral-7B-Instruct-v0.2-code-ftKV unresolvedQ6_K7.2B5.53 GiB1.06 GiB7.43 GiB0.01 GiB48±12.9%
SpydazWeb_AI_CyberTron_Ultra_7bKV unresolvedQ6_K7.2B5.53 GiB1.06 GiB7.43 GiB0.01 GiB48±12.9%
MetaMath-Cybertron-StarlingKV unresolvedQ6_K7.2B5.53 GiB1.06 GiB7.43 GiB0.01 GiB48±12.9%
zephyr-7b-betaKV unresolvedQ6_K7.2B5.53 GiB1.06 GiB7.43 GiB0.01 GiB48±12.9%
japanese-stablelm-instruct-gamma-7bKV unresolvedQ6_K7.2B5.53 GiB1.06 GiB7.43 GiB0.01 GiB48±12.9%
Kunoichi-DPO-v2-7BKV unresolvedQ6_K7.2B5.53 GiB1.06 GiB7.43 GiB0.01 GiB48±12.9%
dolphin-2.0-mistral-7bKV unresolvedQ6_K7.2B5.53 GiB1.06 GiB7.43 GiB0.01 GiB48±12.9%
SciPhi-Mistral-7B-32kKV unresolvedQ6_K7.2B5.53 GiB1.06 GiB7.43 GiB0.01 GiB48±12.9%
Kimiko-Mistral-7BKV unresolvedQ6_K7.2B5.53 GiB1.06 GiB7.43 GiB0.01 GiB48±12.9%
MathCoder2-CodeLlama-7BQ2_K6.7B2.36 GiB4.25 GiB7.43 GiB0.01 GiB48±12.9%
WizardLM-7B-UncensoredI1-Q2_K6.7B2.36 GiB4.25 GiB7.43 GiB0.01 GiB48±12.9%
Llama-2-7B-32K-InstructI1-Q2_K6.7B2.36 GiB4.25 GiB7.43 GiB0.01 GiB48±12.9%
Luna-AI-Llama2-UncensoredI1-Q2_K6.7B2.36 GiB4.25 GiB7.43 GiB0.01 GiB48±12.9%
Swallow-7b-NVE-instruct-hfI1-Q2_K6.7B2.36 GiB4.25 GiB7.43 GiB0.01 GiB48±12.9%
llava-v1.5-7bQ2_K6.7B2.36 GiB4.25 GiB7.43 GiB0.01 GiB48±12.9%
NVIDIA-Nemotron-Nano-9B-v2IQ3_XXS8.9B4.73 GiB1.86 GiB7.43 GiB0.01 GiB48±12.9%
LFM2-24B-A2BMoEIQ2_S23.8B6.45 GiB0.17 GiB7.43 GiB0.01 GiB175±37%
EVA-abliterated-TIES-Qwen2.5-14BI1-IQ2_M14.8B4.99 GiB1.59 GiB7.43 GiB0.01 GiB48±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 generation7.02 it/s5.47–8.41287
Benchmarked· n=287

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 2070 SUPER run?
1357 of 2118 indexed open-weight models fit a GeForce RTX 2070 SUPER at 16,384 context with q8_0 KV cache, the largest being dolphin-2.9.3-mistral-7B-32k at I1-Q6_K. That covers text, vision-language, image, video and speech models.
How much usable memory does a GeForce RTX 2070 SUPER actually have?
Its nameplate is 8 GB, but about 7.44 GiB is available to a model once driver and compositor overhead is accounted for.
Is a GeForce RTX 2070 SUPER fast for local AI?
Its memory bandwidth is 448 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.