AMD · consumer

Radeon RX 6600

Radeon RX 6600 has 8 GB of VRAM at 224 GB/s — about 7.44 GiB usable after driver and compositor overhead. 894 of 2118 indexed models fit at 128K context with q4_0 KV.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed
Memory
8 GB
GDDR6
Bandwidth
224 GB/s
128-bit bus
Tensor FP16
—
dense
TDP
132 W
$329 MSRP
Context4K8K16K32K64K128K
KV cachef16q8_0q4_0quantizing the KV cache is a ~2× lever on the dominant term at long context
text 723vision language 84audio tts 20audio asr 38video 7embedding 21image 1

What fits at 128K context

largest quantization that fits, per model · 894 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
GLM-4.6V-FlashIQ4_NL10.3B5.09 GiB1.41 GiB7.43 GiB0.01 GiB22±26.5%
GLM-Z1-9B-0414IQ4_NL9.4B5.09 GiB1.41 GiB7.43 GiB0.01 GiB22±26.5%
glm4.1v-9b-base-sftI1-IQ4_NL10.3B5.09 GiB1.41 GiB7.43 GiB0.01 GiB22±26.5%
GLM-4-9B-0414IQ4_NL9.4B5.09 GiB1.41 GiB7.43 GiB0.01 GiB22±26.5%
GLM-4.1V-9B-ThinkingIQ4_NL10.3B5.09 GiB1.41 GiB7.43 GiB0.01 GiB22±26.5%
Qwen3-VL-4B-Instruct-Unredacted-MAXI1-Q2_K_S4.4B1.46 GiB5.06 GiB7.43 GiB0.01 GiB22±26.5%
Qwen3-VL-4B-Thinking-Unredacted-MAXI1-Q2_K_S4.4B1.46 GiB5.06 GiB7.43 GiB0.01 GiB22±26.5%
Zubr1.0-VL-4BI1-Q2_K_S4.4B1.46 GiB5.06 GiB7.43 GiB0.01 GiB22±26.5%
Huihui-Qwen3-VL-4B-Instruct-abliteratedI1-Q2_K_S4.4B1.46 GiB5.06 GiB7.43 GiB0.01 GiB22±26.5%
Qwen3-VL-4B-Instruct-UncensoredI1-Q2_K_S4.4B1.46 GiB5.06 GiB7.43 GiB0.01 GiB22±26.5%
OpenCaption-4B-VL-SFT-v1.0I1-Q2_K_S4.4B1.46 GiB5.06 GiB7.43 GiB0.01 GiB22±26.5%
Parable-Qwen3-4B-Claude-Fable-5I1-Q2_K_S4.0B1.46 GiB5.06 GiB7.43 GiB0.01 GiB22±26.5%
Qwen3-4b-Z-Image-Turbo-AbliteratedV1I1-Q2_K_S4.0B1.46 GiB5.06 GiB7.43 GiB0.01 GiB22±26.5%
Neuron-4B-InstructI1-Q2_K_S4.0B1.46 GiB5.06 GiB7.43 GiB0.01 GiB22±26.5%
ChineseErrorCorrector4-4BI1-Q2_K_S4.0B1.46 GiB5.06 GiB7.43 GiB0.01 GiB22±26.5%
FastContext-1.0-4B-SFT-abliteratedI1-Q2_K_S4.0B1.46 GiB5.06 GiB7.43 GiB0.01 GiB22±26.5%
Qwen3-4B-Instruct_NSFW-V2.1I1-Q2_K_S4.0B1.46 GiB5.06 GiB7.43 GiB0.01 GiB22±26.5%
FastContext-1.0-4B-SFTI1-Q2_K_S4.0B1.46 GiB5.06 GiB7.43 GiB0.01 GiB22±26.5%
fable-traces-abliteratedI1-Q2_K_S4.0B1.46 GiB5.06 GiB7.43 GiB0.01 GiB22±26.5%
Nexa-AI-4B-InstructI1-Q2_K_S4.0B1.46 GiB5.06 GiB7.43 GiB0.01 GiB22±26.5%
Lumen-4B-InstructI1-Q2_K_S4.0B1.46 GiB5.06 GiB7.43 GiB0.01 GiB22±26.5%
Qwen3-HereticLM-4BI1-Q2_K_S4.0B1.46 GiB5.06 GiB7.43 GiB0.01 GiB22±26.5%
orpheus-3b-0.1-pretrainedQ5_K_S3.8B2.58 GiB3.94 GiB7.43 GiB0.01 GiB22±26.5%
zeta-2.1I1-IQ1_S8.3B1.98 GiB4.50 GiB7.42 GiB0.02 GiB22±26.5%
EVA-Yi-1.5-9B-32K-V1I1-Q2_K8.8B3.12 GiB3.38 GiB7.42 GiB0.02 GiB22±26.5%
Yi-Coder-9B-ChatQ2_K8.8B3.12 GiB3.38 GiB7.42 GiB0.02 GiB22±26.5%
Yi-1.5-9B-ChatQ2_K8.8B3.12 GiB3.38 GiB7.42 GiB0.02 GiB22±26.5%
rnj-1-instructUD-IQ1_S8.3B1.98 GiB4.50 GiB7.42 GiB0.02 GiB22±26.5%
DeepSeek-OCR-2MoEBF163.4B5.47 GiB1.05 GiB7.42 GiB0.02 GiB42±37%
DeepSeek-OCRMoEBF163.3B5.47 GiB1.05 GiB7.42 GiB0.02 GiB42±37%
Wan2.1-T2V-1.3BQ4_01.4B6.50 GiB0.00 GiB7.42 GiB0.02 GiB22±26.5%
Qwen3-Zero-Coder-Reasoning-V2-0.8BI1-Q6_K816M0.63 GiB5.91 GiB7.41 GiB0.03 GiB22±26.5%
Huihui-gemma-3n-E4B-it-abliteratedQ6_K_L7.8B5.96 GiB0.51 GiB7.41 GiB0.03 GiB22±26.5%
gemma-3n-E4B-itQ6_K_L7.8B5.96 GiB0.51 GiB7.41 GiB0.03 GiB22±26.5%
Qwythos-9B-v2Q4_K_S9.7B5.34 GiB1.13 GiB7.40 GiB0.04 GiB22±26.5%
Tess-4-9BQ4_K_S9.7B5.34 GiB1.13 GiB7.40 GiB0.04 GiB22±26.5%
dolphincoder-starcoder2-15bKV unresolvedI1-IQ1_M16.0B3.60 GiB2.81 GiB7.40 GiB0.04 GiB22±26.5%
starcoder2-15bKV unresolvedIQ1_M16.0B3.60 GiB2.81 GiB7.40 GiB0.04 GiB22±26.5%
Qwen3-4BUD-IQ2_M4.0B1.43 GiB5.06 GiB7.40 GiB0.04 GiB22±26.5%
Jan-nanoUD-IQ2_M4.0B1.43 GiB5.06 GiB7.40 GiB0.04 GiB22±26.5%
SuperGemma-4-12b-abliteratedI1-IQ2_M12.0B4.07 GiB2.38 GiB7.40 GiB0.04 GiB22±26.5%
gemma-4-12B-agentic-fable5-composer2.5-v2-3.5x-tau2-uncensored-hereticI1-IQ2_M12.0B4.07 GiB2.38 GiB7.40 GiB0.04 GiB22±26.5%
gemma-4-12B-coder-fable5-composer2.5-v1-uncensored-hereticI1-IQ2_M12.0B4.07 GiB2.38 GiB7.40 GiB0.04 GiB22±26.5%
gemma-4-12B-it-uncensored-hereticI1-IQ2_M12.0B4.07 GiB2.38 GiB7.40 GiB0.04 GiB22±26.5%
Grug-12BI1-IQ2_M12.0B4.07 GiB2.38 GiB7.40 GiB0.04 GiB22±26.5%
Aura-Medium-v1-BF16I1-IQ2_M12.0B4.07 GiB2.38 GiB7.40 GiB0.04 GiB22±26.5%
gemma-4-12B-it-Esper4I1-IQ2_M12.0B4.07 GiB2.38 GiB7.40 GiB0.04 GiB22±26.5%
gemma-4-12B-it-GuardpointI1-IQ2_M12.0B4.07 GiB2.38 GiB7.40 GiB0.04 GiB22±26.5%
Gemma-4-12B-it-AEON-Abliterated-K4-BF16I1-IQ2_M12.0B4.07 GiB2.38 GiB7.40 GiB0.04 GiB22±26.5%
gemma-4-12B-it-Tachibana-AgentI1-IQ2_M12.0B4.07 GiB2.38 GiB7.40 GiB0.04 GiB22±26.5%
gemma-4-12b-marvin-gutenberg-rp-v2I1-IQ2_M12.0B4.07 GiB2.38 GiB7.40 GiB0.04 GiB22±26.5%
gemma-4-12b-crownelius-writerI1-IQ2_M12.0B4.07 GiB2.38 GiB7.40 GiB0.04 GiB22±26.5%
Huihui-gemma-4-12B-coder-fable5-composer2.5-v1-abliteratedI1-IQ2_M12.0B4.07 GiB2.38 GiB7.40 GiB0.04 GiB22±26.5%
gemma-4-12b-asterion-agenticI1-IQ2_M12.0B4.07 GiB2.38 GiB7.40 GiB0.04 GiB22±26.5%
Huihui-gemma-4-12B-agentic-fable5-abliteratedI1-IQ2_M12.0B4.07 GiB2.38 GiB7.40 GiB0.04 GiB22±26.5%
g4-12b-it-trismegistusI1-IQ2_M12.0B4.07 GiB2.38 GiB7.40 GiB0.04 GiB22±26.5%
gemma4-12b-it-asimovI1-IQ2_M12.0B4.07 GiB2.38 GiB7.40 GiB0.04 GiB22±26.5%
FabGemmaI1-IQ2_M12.0B4.07 GiB2.38 GiB7.40 GiB0.04 GiB22±26.5%
Huihui-gemma-4-12B-it-qat-q4_0-unquantized-abliteratedI1-IQ2_M12.0B4.07 GiB2.38 GiB7.40 GiB0.04 GiB22±26.5%
gemma-4-12B-it-abliterated-uncensoredI1-IQ2_M12.0B4.07 GiB2.38 GiB7.40 GiB0.04 GiB22±26.5%
●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 generation2.06 it/s0.90–3.3091
Benchmarked· n=91

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 Radeon RX 6600 run?
894 of 2118 indexed open-weight models fit a Radeon RX 6600 at 131,072 context with q4_0 KV cache, the largest being GLM-4.6V-Flash at IQ4_NL. That covers text, vision-language, image, video and speech models.
How much usable memory does a Radeon RX 6600 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 Radeon RX 6600 fast for local AI?
Its memory bandwidth is 224 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.