Skip to content

NVIDIA GeForce RTX 2060 Super

NVIDIA · 8GBGDDR6 · Can run 57 models

BuyAmazon
ManufacturerNVIDIA
VRAM8 GB
Memory TypeGDDR6
ArchitectureTuring
CUDA Cores2,176
Tensor Cores272
Bandwidth448 GB/s
TDP175W
MSRP$399
ReleasedJul 9, 2019

AI Notes

The RTX 2060 Super offers 8GB VRAM, enough to run 7B models comfortably with Q4 quantization. It's a common card on the used market at attractive prices. Tensor cores provide a small inference boost over the GTX series. A decent entry point for local AI on a budget.

Compatible Models

ModelParametersBest QuantVRAM UsedFitEst. Speed
Qwen 3 0.6B600MQ4_K_M2.5 GBRuns~179 tok/s
Qwen 3.5 0.8B800MQ4_K_M1.5 GBRuns~299 tok/s
Gemma 3 1B1BQ8_02 GBRuns~224 tok/s
Llama 3.2 1B1BQ8_03 GBRuns~149 tok/s
DeepSeek R1 1.5B1.5BQ8_03 GBRuns~149 tok/s
SmolLM2 1.7B1.7BQ8_02.7 GBRuns~166 tok/s
Gemma 2 2B2BQ8_04 GBRuns~112 tok/s
Gemma 3n E2B2BQ4_K_M3.3 GBRuns~136 tok/s
Gemma 4 E2B2BQ4_K_M4 GBRuns~112 tok/s
Qwen 3.5 2B2BQ4_K_M3 GBRuns~149 tok/s
Llama 3.2 3B3BQ8_05 GBRuns~90 tok/s
StarCoder2 3B3BQ4_K_M3.5 GBRuns~128 tok/s
Phi-3 Mini 3.8B3.8BQ8_05.8 GBRuns~77 tok/s
Phi-4 Mini 3.8B3.8BQ4_K_M4.5 GBRuns~100 tok/s
Gemma 3 4B4BQ4_K_M5 GBRuns~90 tok/s
Gemma 3n E4B4BQ4_K_M4.5 GBRuns~100 tok/s
Gemma 4 E4B4BQ4_K_M6 GBRuns~75 tok/s
Qwen 3 4B4BQ4_K_M4.5 GBRuns~100 tok/s
Qwen 3.5 4B4BQ4_K_M4.5 GBRuns~100 tok/s
Yi 1.5 6B6BQ4_K_M5 GBRuns~90 tok/s
Falcon 3 7B7BQ4_K_M6.8 GBRuns~66 tok/s
InternLM 2.5 7B7BQ4_K_M5.5 GBRuns~81 tok/s
StarCoder2 7B7BQ4_K_M5.5 GBRuns~81 tok/s
Aya Expanse 8B8BQ4_K_M6.5 GBRuns~69 tok/s
Dolphin 3 8B8BQ4_K_M6 GBRuns~75 tok/s
Nous Hermes 2 8B8BQ4_K_M6 GBRuns~75 tok/s
Yi 1.5 9B9BQ4_K_M6.5 GBRuns~69 tok/s
Codestral Mamba 7B7BQ4_K_M6.9 GBRuns (tight)~65 tok/s
OpenChat 3.5 7B7BQ4_K_M6.9 GBRuns (tight)~65 tok/s
Qwen 2.5 VL 7B7BQ4_K_M7 GBRuns (tight)~64 tok/s
WizardLM 2 7B7BQ4_K_M6.9 GBRuns (tight)~65 tok/s
Cogito 8B8BQ4_K_M7.5 GBRuns (tight)~60 tok/s
DeepSeek R1 8B8BQ4_K_M7.5 GBRuns (tight)~60 tok/s
Nemotron 3 Nano 8B8BQ4_K_M7.5 GBRuns (tight)~60 tok/s
Qwen 3 8B8BQ4_K_M7.5 GBRuns (tight)~60 tok/s
Qwen 3.5 9B9BQ4_K_M7.5 GBRuns (tight)~60 tok/s
DeepSeek R1 7B7BQ8_09 GBCPU Offload~15 tok/s
Mistral 7B7BQ8_09 GBCPU Offload~15 tok/s
Qwen 2.5 7B7BQ8_09 GBCPU Offload~15 tok/s
Qwen 2.5 Coder 7B7BQ8_09 GBCPU Offload~15 tok/s
Granite 3.3 8B8BQ8_010 GBCPU Offload~14 tok/s
Llama 3.1 8B8BQ8_010 GBCPU Offload~14 tok/s
Gemma 2 9B9BQ8_011 GBCPU Offload~12 tok/s
Yi Coder 9B9BQ4_K_M8 GBCPU Offload~17 tok/s
Falcon 3 10B10BQ4_K_M8.5 GBCPU Offload~16 tok/s
Llama 3.2 Vision 11B11BQ4_K_M8.5 GBCPU Offload~16 tok/s
Gemma 3 12B12BQ4_K_M10.5 GBCPU Offload~13 tok/s
Mistral Nemo 12B12BQ4_K_M9.5 GBCPU Offload~14 tok/s
DeepSeek R1 14B14BQ4_K_M9.9 GBCPU Offload~14 tok/s
Phi-4 14B14BQ4_K_M9.9 GBCPU Offload~14 tok/s
Phi-4 Reasoning 14B14BQ4_K_M11 GBCPU Offload~12 tok/s
Qwen 2.5 14B14BQ4_K_M9.9 GBCPU Offload~14 tok/s
Qwen 2.5 Coder 14B14BQ4_K_M12 GBCPU Offload~11 tok/s
Qwen 3 14B14BQ4_K_M12 GBCPU Offload~11 tok/s
StarCoder2 15B15BQ4_K_M10.5 GBCPU Offload~13 tok/s
InternLM 2.5 20B20BQ4_K_M12 GBCPU Offload~11 tok/s
Qwen 3.5 35B A3B35BQ4_K_M12 GBCPU Offload~11 tok/s
57 model(s) are too large for this hardware.