AMD · consumer

Radeon RX 9060

Radeon RX 9060 has 8 GB of VRAM at 288 GB/s — about 7.44 GiB usable after driver and compositor overhead. 615 of 2118 indexed models fit at 64K context with f16 KV.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed
Memory
8 GB
GDDR6
Bandwidth
288 GB/s
128-bit bus
Tensor FP16
—
dense
TDP
132 W
Context4K8K16K32K64K128K
KV cachef16q8_0q4_0quantizing the KV cache is a ~2× lever on the dominant term at long context
vision language 71text 465embedding 18audio asr 35video 7image 1audio tts 18

What fits at 64K context

largest quantization that fits, per model · 615 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
Qwen3.5-9BQ3_K_M9.7B4.50 GiB2.00 GiB7.44 GiB0.00 GiB28±26.5%
Crow-9B-HERETIC-4.6I1-Q3_K_L9.4B4.49 GiB2.00 GiB7.43 GiB0.01 GiB28±26.5%
Qwen3.5-9B-Claude-4.6-OS-Auto-Variable-HERETIC-UNCENSORED-THINKINGI1-Q3_K_L9.4B4.49 GiB2.00 GiB7.43 GiB0.01 GiB28±26.5%
Qwen3.5-9B-Claude-4.6-HighIQ-INSTRUCT-HERETIC-UNCENSOREDI1-Q3_K_L9.4B4.49 GiB2.00 GiB7.43 GiB0.01 GiB28±26.5%
Qwen3.5-9B-Claude-4.6-OS-HERETIC-UNCENSORED-INSTRUCTI1-Q3_K_L9.4B4.49 GiB2.00 GiB7.43 GiB0.01 GiB28±26.5%
Qwen3.5-9B-Claude-4.6-HighIQ-THINKING-HERETIC-UNCENSOREDI1-Q3_K_L9.4B4.49 GiB2.00 GiB7.43 GiB0.01 GiB28±26.5%
NaNovel-9BI1-Q3_K_L9.7B4.49 GiB2.00 GiB7.43 GiB0.01 GiB28±26.5%
Qwen3.5-9B-Unredacted-MAXI1-Q3_K_L9.4B4.49 GiB2.00 GiB7.43 GiB0.01 GiB28±26.5%
Qwen3.5-9B-abliteratedI1-Q3_K_L9.4B4.49 GiB2.00 GiB7.43 GiB0.01 GiB28±26.5%
Ken3.5-9BI1-Q3_K_L9.7B4.49 GiB2.00 GiB7.43 GiB0.01 GiB28±26.5%
Huihui-Qwen3.5-9B-abliteratedQ3_K_L9.7B4.49 GiB2.00 GiB7.43 GiB0.01 GiB28±26.5%
Qwen3.5-9B-abliteratedQ3_K_L9.0B4.49 GiB2.00 GiB7.43 GiB0.01 GiB28±26.5%
Qwen3.5-9B-BaseQ3_K_L9.7B4.49 GiB2.00 GiB7.43 GiB0.01 GiB28±26.5%
Qwen3.5-9B-gemini-3.1-opus-4.6-reasoningI1-Q3_K_L9.4B4.49 GiB2.00 GiB7.43 GiB0.01 GiB28±26.5%
SmolLM3-3BQ5_K_S3.1B2.01 GiB4.50 GiB7.42 GiB0.02 GiB28±26.5%
Wan2.1-T2V-1.3BQ4_01.4B6.50 GiB0.00 GiB7.42 GiB0.02 GiB28±26.5%
Huihui-gemma-3n-E4B-it-abliteratedQ5_K_L7.8B5.55 GiB0.93 GiB7.42 GiB0.02 GiB28±26.5%
gemma-3n-E4B-itQ5_K_L7.8B5.55 GiB0.93 GiB7.42 GiB0.02 GiB28±26.5%
Parable-Granite-4.1-3B-Claude-Fable-5I1-IQ3_M3.4B1.51 GiB5.00 GiB7.41 GiB0.03 GiB28±26.5%
granite-4.0-microIQ3_M3.4B1.51 GiB5.00 GiB7.41 GiB0.03 GiB28±26.5%
EXAONE-4.0-1.2B-abliteratedF161.5B2.78 GiB3.75 GiB7.41 GiB0.03 GiB28±26.5%
Qwythos-9B-v2Q3_K_S9.7B4.48 GiB2.00 GiB7.41 GiB0.03 GiB28±26.5%
Tess-4-9BQ3_K_S9.7B4.48 GiB2.00 GiB7.41 GiB0.03 GiB28±26.5%
granite-3.1-3b-a800m-instructMoEQ6_K_L3.3B2.54 GiB4.00 GiB7.41 GiB0.03 GiB22±37%
Qwen2.5-VL-7B-InstructUD-IQ3_XXS8.3B2.95 GiB3.50 GiB7.40 GiB0.04 GiB28±26.5%
snowflake-arctic-embed-l-v2.0Q6_K_L568M0.52 GiB6.00 GiB7.40 GiB0.04 GiB28±26.5%
Qwen2.5-3B-Instruct-abliteratedI1-Q5_K_S3.1B4.24 GiB2.25 GiB7.40 GiB0.04 GiB28±26.5%
Qwen3.6-35B-A3B-REAM-160-ru-agentMoEIQ1_S23.6B5.24 GiB1.25 GiB7.40 GiB0.04 GiB54±37%
granite-3.3-2b-instructQ4_12.5B1.50 GiB5.00 GiB7.39 GiB0.05 GiB28±26.5%
granite-3.2-2b-instructQ4_12.5B1.50 GiB5.00 GiB7.39 GiB0.05 GiB28±26.5%
granite-vision-3.2-2bQ4_13.0B1.50 GiB5.00 GiB7.39 GiB0.05 GiB28±26.5%
GLM-4.6V-FlashUD-IQ3_XXS10.3B3.94 GiB2.50 GiB7.38 GiB0.06 GiB28±26.5%
GLM-Z1-9B-0414UD-IQ3_XXS9.4B3.94 GiB2.50 GiB7.38 GiB0.06 GiB28±26.5%
GLM-4-9B-0414UD-IQ3_XXS9.4B3.94 GiB2.50 GiB7.38 GiB0.06 GiB28±26.5%
GLM-4.1V-9B-ThinkingUD-IQ3_XXS10.3B3.94 GiB2.50 GiB7.38 GiB0.06 GiB28±26.5%
granite-speech-4.1-2bQ4_K_M2.3B1.49 GiB5.00 GiB7.38 GiB0.06 GiB28±26.5%
granite-4.0-1b-speechQ4_K_M2.3B1.49 GiB5.00 GiB7.38 GiB0.06 GiB28±26.5%
granite-3.1-2b-instructQ4_12.5B1.48 GiB5.00 GiB7.38 GiB0.06 GiB28±26.5%
glm4.1v-9b-base-sftI1-IQ3_XXS10.3B3.94 GiB2.50 GiB7.37 GiB0.07 GiB28±26.5%
GrammarCoder-7B-BaseI1-IQ3_XXS7.6B2.91 GiB3.50 GiB7.36 GiB0.08 GiB28±26.5%
granite-4.1-3bQ3_K_S3.4B1.46 GiB5.00 GiB7.36 GiB0.08 GiB28±26.5%
granite-4.0-micro-baseQ3_K_S3.4B1.46 GiB5.00 GiB7.36 GiB0.08 GiB28±26.5%
GLM-ASR-Nano-2512BF162.3B2.97 GiB3.50 GiB7.36 GiB0.08 GiB28±26.5%
InternVL3_5-14BIQ3_M15.1B6.41 GiB0.00 GiB7.36 GiB0.08 GiB28±26.5%
DeepHat-V1-7B-Heretic-AbliteratedI1-IQ3_XXS7.6B2.90 GiB3.50 GiB7.35 GiB0.09 GiB28±26.5%
ShizhenGPT-7B-VLI1-IQ3_XXS8.3B2.90 GiB3.50 GiB7.35 GiB0.09 GiB28±26.5%
HuatuoGPT-o1-7BI1-IQ3_XXS7.6B2.90 GiB3.50 GiB7.35 GiB0.09 GiB28±26.5%
MathSmith-DS-Qwen-7B-LongCoTI1-IQ3_XXS7.6B2.90 GiB3.50 GiB7.35 GiB0.09 GiB28±26.5%
AstraGPTCoder-7BI1-IQ3_XXS7.6B2.90 GiB3.50 GiB7.35 GiB0.09 GiB28±26.5%
Qwen2.5-Coder-7B-Instruct-Ghidra-v2I1-IQ3_XXS7.6B2.90 GiB3.50 GiB7.35 GiB0.09 GiB28±26.5%
EsDrac-v1-7BI1-IQ3_XXS7.6B2.90 GiB3.50 GiB7.35 GiB0.09 GiB28±26.5%
Hemlock-Apothecary-7B-GRPO-e3I1-IQ3_XXS7.6B2.90 GiB3.50 GiB7.35 GiB0.09 GiB28±26.5%
openhands-lm-7b-v0.1I1-IQ3_XXS7.6B2.90 GiB3.50 GiB7.35 GiB0.09 GiB28±26.5%
Hemlock2-Coder-7B-GRPOI1-IQ3_XXS7.6B2.90 GiB3.50 GiB7.35 GiB0.09 GiB28±26.5%
shellwhiz-7bI1-IQ3_XXS7.6B2.90 GiB3.50 GiB7.35 GiB0.09 GiB28±26.5%
Qwen2.5-Coder-7B-Instruct-abliteratedI1-IQ3_XXS7.6B2.90 GiB3.50 GiB7.35 GiB0.09 GiB28±26.5%
Qwen2.5-Coder-7B-Instruct-OBLITERATED-advancedI1-IQ3_XXS7.6B2.90 GiB3.50 GiB7.35 GiB0.09 GiB28±26.5%
Qwen-STEM-Specialist-7BI1-IQ3_XXS7.6B2.90 GiB3.50 GiB7.35 GiB0.09 GiB28±26.5%
DeepHat-V1-7BIQ3_XXS7.6B2.90 GiB3.50 GiB7.35 GiB0.09 GiB28±26.5%
VulnLLM-R-7BI1-IQ3_XXS7.6B2.90 GiB3.50 GiB7.35 GiB0.09 GiB28±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.

Questions people ask

What AI models can a Radeon RX 9060 run?
615 of 2118 indexed open-weight models fit a Radeon RX 9060 at 65,536 context with f16 KV cache, the largest being Qwen3.5-9B at Q3_K_M. That covers text, vision-language, image, video and speech models.
How much usable memory does a Radeon RX 9060 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 9060 fast for local AI?
Its memory bandwidth is 288 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.