Skip to content

AMD Radeon RX 6600 XT

AMD · 8GBGDDR6 · Can run 57 models

BuyAmazon
ManufacturerAMD
VRAM8 GB
Memory TypeGDDR6
ArchitectureRDNA 2
Stream Procs2,048
Bandwidth256 GB/s
TDP160W
MSRP$379
ReleasedAug 11, 2021

AI Notes

The RX 6600 XT offers 8GB VRAM at a budget-friendly price on the used market. It can run 7B models with Q4 quantization, though the lower memory bandwidth (256 GB/s) means slower token generation compared to higher-end cards. ROCm support is available for RDNA 2 but requires some setup effort.

Compatible Models

ModelParametersBest QuantVRAM UsedFitEst. Speed
Qwen 3 0.6B600MQ4_K_M2.5 GBRuns~102 tok/s
Qwen 3.5 0.8B800MQ4_K_M1.5 GBRuns~171 tok/s
Gemma 3 1B1BQ8_02 GBRuns~128 tok/s
Llama 3.2 1B1BQ8_03 GBRuns~85 tok/s
DeepSeek R1 1.5B1.5BQ8_03 GBRuns~85 tok/s
SmolLM2 1.7B1.7BQ8_02.7 GBRuns~95 tok/s
Gemma 2 2B2BQ8_04 GBRuns~64 tok/s
Gemma 3n E2B2BQ4_K_M3.3 GBRuns~78 tok/s
Gemma 4 E2B2BQ4_K_M4 GBRuns~64 tok/s
Qwen 3.5 2B2BQ4_K_M3 GBRuns~85 tok/s
Llama 3.2 3B3BQ8_05 GBRuns~51 tok/s
StarCoder2 3B3BQ4_K_M3.5 GBRuns~73 tok/s
Phi-3 Mini 3.8B3.8BQ8_05.8 GBRuns~44 tok/s
Phi-4 Mini 3.8B3.8BQ4_K_M4.5 GBRuns~57 tok/s
Gemma 3 4B4BQ4_K_M5 GBRuns~51 tok/s
Gemma 3n E4B4BQ4_K_M4.5 GBRuns~57 tok/s
Gemma 4 E4B4BQ4_K_M6 GBRuns~43 tok/s
Qwen 3 4B4BQ4_K_M4.5 GBRuns~57 tok/s
Qwen 3.5 4B4BQ4_K_M4.5 GBRuns~57 tok/s
Yi 1.5 6B6BQ4_K_M5 GBRuns~51 tok/s
Falcon 3 7B7BQ4_K_M6.8 GBRuns~38 tok/s
InternLM 2.5 7B7BQ4_K_M5.5 GBRuns~47 tok/s
StarCoder2 7B7BQ4_K_M5.5 GBRuns~47 tok/s
Aya Expanse 8B8BQ4_K_M6.5 GBRuns~39 tok/s
Dolphin 3 8B8BQ4_K_M6 GBRuns~43 tok/s
Nous Hermes 2 8B8BQ4_K_M6 GBRuns~43 tok/s
Yi 1.5 9B9BQ4_K_M6.5 GBRuns~39 tok/s
Codestral Mamba 7B7BQ4_K_M6.9 GBRuns (tight)~37 tok/s
OpenChat 3.5 7B7BQ4_K_M6.9 GBRuns (tight)~37 tok/s
Qwen 2.5 VL 7B7BQ4_K_M7 GBRuns (tight)~37 tok/s
WizardLM 2 7B7BQ4_K_M6.9 GBRuns (tight)~37 tok/s
Cogito 8B8BQ4_K_M7.5 GBRuns (tight)~34 tok/s
DeepSeek R1 8B8BQ4_K_M7.5 GBRuns (tight)~34 tok/s
Nemotron 3 Nano 8B8BQ4_K_M7.5 GBRuns (tight)~34 tok/s
Qwen 3 8B8BQ4_K_M7.5 GBRuns (tight)~34 tok/s
Qwen 3.5 9B9BQ4_K_M7.5 GBRuns (tight)~34 tok/s
DeepSeek R1 7B7BQ8_09 GBCPU Offload~8 tok/s
Mistral 7B7BQ8_09 GBCPU Offload~8 tok/s
Qwen 2.5 7B7BQ8_09 GBCPU Offload~8 tok/s
Qwen 2.5 Coder 7B7BQ8_09 GBCPU Offload~8 tok/s
Granite 3.3 8B8BQ8_010 GBCPU Offload~8 tok/s
Llama 3.1 8B8BQ8_010 GBCPU Offload~8 tok/s
Gemma 2 9B9BQ8_011 GBCPU Offload~7 tok/s
Yi Coder 9B9BQ4_K_M8 GBCPU Offload~10 tok/s
Falcon 3 10B10BQ4_K_M8.5 GBCPU Offload~9 tok/s
Llama 3.2 Vision 11B11BQ4_K_M8.5 GBCPU Offload~9 tok/s
Gemma 3 12B12BQ4_K_M10.5 GBCPU Offload~7 tok/s
Mistral Nemo 12B12BQ4_K_M9.5 GBCPU Offload~8 tok/s
DeepSeek R1 14B14BQ4_K_M9.9 GBCPU Offload~8 tok/s
Phi-4 14B14BQ4_K_M9.9 GBCPU Offload~8 tok/s
Phi-4 Reasoning 14B14BQ4_K_M11 GBCPU Offload~7 tok/s
Qwen 2.5 14B14BQ4_K_M9.9 GBCPU Offload~8 tok/s
Qwen 2.5 Coder 14B14BQ4_K_M12 GBCPU Offload~6 tok/s
Qwen 3 14B14BQ4_K_M12 GBCPU Offload~6 tok/s
StarCoder2 15B15BQ4_K_M10.5 GBCPU Offload~7 tok/s
InternLM 2.5 20B20BQ4_K_M12 GBCPU Offload~6 tok/s
Qwen 3.5 35B A3B35BQ4_K_M12 GBCPU Offload~6 tok/s
57 model(s) are too large for this hardware.