NVIDIA · consumer

GeForce RTX 2070

GeForce RTX 2070 has 8 GB of VRAM at 448 GB/s — about 7.44 GiB usable after driver and compositor overhead. 1421 of 2118 indexed models fit at 4K 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
60 TF
dense
TDP
175 W
$499 MSRP
Context4K8K16K32K64K128K
KV cachef16q8_0q4_0quantizing the KV cache is a ~2× lever on the dominant term at long context
text 1223vision language 103video 8embedding 26audio tts 21image 2audio asr 38

What fits at 4K context

largest quantization that fits, per model · 1421 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
Aurora-Code-1MoEI1-IQ1_M34.7B6.59 GiB0.04 GiB7.44 GiB0.00 GiB244±37%
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%
UncensoredLM-DeepSeek-R1-Distill-Qwen-14BIQ3_M14.2B6.20 GiB0.38 GiB7.43 GiB0.01 GiB48±12.9%
medgemma-27b-itUD-IQ1_S28.8B6.06 GiB0.49 GiB7.43 GiB0.01 GiB48±12.9%
gemma-3-27b-itUD-IQ1_S27.4B6.06 GiB0.49 GiB7.43 GiB0.01 GiB48±12.9%
medgemma-27b-text-itUD-IQ1_S27.0B6.06 GiB0.49 GiB7.43 GiB0.01 GiB48±12.9%
gemma-4-E4B-it-hereticQ6_K8.0B6.55 GiB0.07 GiB7.43 GiB0.01 GiB48±12.9%
GLM-4.7-Flash-REAP-23B-A3B-absolute-heresyMoEI1-IQ2_S23.0B6.51 GiB0.11 GiB7.43 GiB0.01 GiB163±37%
Ministral-3-3B-Instruct-2512BF163.8B6.40 GiB0.22 GiB7.42 GiB0.02 GiB48±12.9%
Ministral-3-3B-Reasoning-2512BF164.3B6.40 GiB0.22 GiB7.42 GiB0.02 GiB48±12.9%
Grug-12BQ3_K_L12.0B6.20 GiB0.38 GiB7.42 GiB0.02 GiB48±12.9%
gemma-4-12B-it-Esper4Q3_K_L12.0B6.20 GiB0.38 GiB7.42 GiB0.02 GiB48±12.9%
gemma-4-12B-itQ3_K_L12.0B6.20 GiB0.38 GiB7.42 GiB0.02 GiB48±12.9%
LFM2.5-8B-A1BMoEUD-Q6_K8.5B6.60 GiB0.02 GiB7.42 GiB0.02 GiB140±37%
Ministral-3-3B-Instruct-2512-BF16BF164.3B6.39 GiB0.22 GiB7.42 GiB0.02 GiB48±12.9%
Amaretto-3BF164.3B6.39 GiB0.22 GiB7.42 GiB0.02 GiB48±12.9%
spoomplesmaxx-mini-14BI1-IQ3_S14.8B6.23 GiB0.33 GiB7.42 GiB0.02 GiB48±12.9%
vanilla-cn-roleplay-0.2I1-IQ3_S14.8B6.23 GiB0.33 GiB7.42 GiB0.02 GiB48±12.9%
Claria-14bI1-IQ3_S14.8B6.23 GiB0.33 GiB7.42 GiB0.02 GiB48±12.9%
NTX-2.1-ProI1-IQ3_S14.8B6.23 GiB0.33 GiB7.42 GiB0.02 GiB48±12.9%
Qwen3-14B-UncensoredI1-IQ3_S14.8B6.23 GiB0.33 GiB7.42 GiB0.02 GiB48±12.9%
FrogMini-14B-2510I1-IQ3_S—6.23 GiB0.33 GiB7.42 GiB0.02 GiB48±12.9%
Qwen3-14B-abliteratedI1-IQ3_S14.8B6.23 GiB0.33 GiB7.42 GiB0.02 GiB48±12.9%
Hermes-4-14BIQ3_S14.8B6.23 GiB0.33 GiB7.42 GiB0.02 GiB48±12.9%
Slava-Qwen3-14B-SerbianI1-IQ3_S14.8B6.23 GiB0.33 GiB7.42 GiB0.02 GiB48±12.9%
Huihui-Qwen3-14B-abliterated-v2I1-IQ3_S14.8B6.23 GiB0.33 GiB7.42 GiB0.02 GiB48±12.9%
zeta-2.1I1-Q6_K8.3B6.31 GiB0.27 GiB7.42 GiB0.02 GiB48±12.9%
HomunculusQ3_K_L12.5B6.23 GiB0.33 GiB7.41 GiB0.03 GiB48±12.9%
Tess-4-9BQ5_K_M9.7B6.51 GiB0.07 GiB7.41 GiB0.03 GiB48±12.9%
glm-4v-9bQ5_K_M13.9B6.57 GiB0.00 GiB7.41 GiB0.03 GiB48±12.9%
Ministral-3-14B-Instruct-2512-BF16-abliteratedI1-Q3_K_M13.9B6.22 GiB0.33 GiB7.41 GiB0.03 GiB48±12.9%
Ministral-3-14B-abliteratedQ3_K_M13.9B6.22 GiB0.33 GiB7.41 GiB0.03 GiB48±12.9%
Ministral-3-14B-Instruct-2512-BF16Q3_K_M13.9B6.22 GiB0.33 GiB7.41 GiB0.03 GiB48±12.9%
Ministral-3-14B-Instruct-2512Q3_K_M13.9B6.22 GiB0.33 GiB7.41 GiB0.03 GiB48±12.9%
Ministral-3-14B-Reasoning-2512-UncensoredI1-Q3_K_M13.9B6.22 GiB0.33 GiB7.41 GiB0.03 GiB48±12.9%
Ministral-3-14B-Reasoning-2512Q3_K_M13.9B6.22 GiB0.33 GiB7.41 GiB0.03 GiB48±12.9%
granite-3.3-8b-instructQ6_K8.2B6.24 GiB0.33 GiB7.41 GiB0.03 GiB48±12.9%
granite-3.1-8b-instructQ6_K_M8.2B6.24 GiB0.33 GiB7.41 GiB0.03 GiB48±12.9%
granite-3.2-8b-instructQ6_K8.2B6.24 GiB0.33 GiB7.41 GiB0.03 GiB48±12.9%
gemma-2-27b-itIQ1_S27.2B5.71 GiB0.76 GiB7.41 GiB0.03 GiB49±12.9%
Qwen3-VL-8B-Instruct-HereticI1-IQ3_XXS8.8B6.28 GiB0.30 GiB7.41 GiB0.03 GiB48±12.9%
SuperGemma-4-12b-abliteratedI1-IQ4_XS12.0B6.18 GiB0.38 GiB7.41 GiB0.03 GiB48±12.9%
gemma-4-12B-agentic-fable5-composer2.5-v2-3.5x-tau2-uncensored-hereticI1-IQ4_XS12.0B6.18 GiB0.38 GiB7.41 GiB0.03 GiB48±12.9%
gemma-4-12B-coder-fable5-composer2.5-v1-uncensored-hereticI1-IQ4_XS12.0B6.18 GiB0.38 GiB7.41 GiB0.03 GiB48±12.9%
gemma-4-12B-it-uncensored-hereticI1-IQ4_XS12.0B6.18 GiB0.38 GiB7.41 GiB0.03 GiB48±12.9%
Aura-Medium-v1-BF16I1-IQ4_XS12.0B6.18 GiB0.38 GiB7.41 GiB0.03 GiB48±12.9%
gemma-4-12B-it-GuardpointI1-IQ4_XS12.0B6.18 GiB0.38 GiB7.41 GiB0.03 GiB48±12.9%
Gemma-4-12B-it-AEON-Abliterated-K4-BF16I1-IQ4_XS12.0B6.18 GiB0.38 GiB7.41 GiB0.03 GiB48±12.9%
gemma-4-12B-it-Tachibana-AgentI1-IQ4_XS12.0B6.18 GiB0.38 GiB7.41 GiB0.03 GiB48±12.9%
gemma-4-12b-marvin-gutenberg-rp-v2I1-IQ4_XS12.0B6.18 GiB0.38 GiB7.41 GiB0.03 GiB48±12.9%
gemma-4-12b-crownelius-writerI1-IQ4_XS12.0B6.18 GiB0.38 GiB7.41 GiB0.03 GiB48±12.9%
Huihui-gemma-4-12B-coder-fable5-composer2.5-v1-abliteratedI1-IQ4_XS12.0B6.18 GiB0.38 GiB7.41 GiB0.03 GiB48±12.9%
gemma-4-12b-asterion-agenticI1-IQ4_XS12.0B6.18 GiB0.38 GiB7.41 GiB0.03 GiB48±12.9%
Huihui-gemma-4-12B-agentic-fable5-abliteratedI1-IQ4_XS12.0B6.18 GiB0.38 GiB7.41 GiB0.03 GiB48±12.9%
g4-12b-it-trismegistusI1-IQ4_XS12.0B6.18 GiB0.38 GiB7.41 GiB0.03 GiB48±12.9%
gemma4-12b-it-asimovI1-IQ4_XS12.0B6.18 GiB0.38 GiB7.41 GiB0.03 GiB48±12.9%
FabGemmaI1-IQ4_XS12.0B6.18 GiB0.38 GiB7.41 GiB0.03 GiB48±12.9%
Huihui-gemma-4-12B-it-qat-q4_0-unquantized-abliteratedI1-IQ4_XS12.0B6.18 GiB0.38 GiB7.41 GiB0.03 GiB48±12.9%
gemma-4-12B-it-abliterated-uncensoredI1-IQ4_XS12.0B6.18 GiB0.38 GiB7.41 GiB0.03 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 generation5.75 it/s4.02–6.72293
Benchmarked· n=293

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 run?
1421 of 2118 indexed open-weight models fit a GeForce RTX 2070 at 4,096 context with q8_0 KV cache, the largest being Aurora-Code-1 at I1-IQ1_M. That covers text, vision-language, image, video and speech models.
How much usable memory does a GeForce RTX 2070 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 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.