AMD · workstation

Radeon Pro W6600

Radeon Pro W6600 has 8 GB of VRAM at 224 GB/s — about 7.44 GiB usable after driver and compositor overhead. 1398 of 2118 indexed models fit at 8K context with q8_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
130 W
$649 MSRP
Context4K8K16K32K64K128K
KV cachef16q8_0q4_0quantizing the KV cache is a ~2× lever on the dominant term at long context
text 1203vision language 101image 2embedding 26video 7audio tts 21audio asr 38

What fits at 8K context

largest quantization that fits, per model · 1398 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
INTELLECT-1-InstructI1-Q4_K_M10.2B5.80 GiB0.70 GiB7.44 GiB0.00 GiB22±26.5%
dolphin-2.9.2-Phi-3-MediumKV unresolvedQ3_K_S14.0B5.65 GiB0.83 GiB7.44 GiB0.00 GiB22±26.5%
Phi-3-medium-128k-instructQ3_K_S14.0B5.65 GiB0.83 GiB7.44 GiB0.00 GiB22±26.5%
Phi-3-medium-4k-instructI1-IQ3_S14.0B5.65 GiB0.83 GiB7.44 GiB0.00 GiB22±26.5%
Rocinante-XL-16B-v1IQ2_S16.1B5.59 GiB0.90 GiB7.44 GiB0.00 GiB22±26.5%
stable-code-3bF162.8B5.21 GiB1.33 GiB7.43 GiB0.01 GiB22±26.5%
Nous-Hermes-2-SOLAR-10.7BQ4_K_S10.7B5.70 GiB0.80 GiB7.43 GiB0.01 GiB22±26.5%
SOLAR-10.7B-Instruct-v1.0I1-Q4_K_S10.7B5.70 GiB0.80 GiB7.43 GiB0.01 GiB22±26.5%
LFM2.5-Queen-Opus-4.7-8B-A1BMoEI1-Q6_K8.5B6.48 GiB0.05 GiB7.43 GiB0.01 GiB63±37%
LFM2.5-8B-A1B-KO-SFTMoEI1-Q6_K8.5B6.48 GiB0.05 GiB7.43 GiB0.01 GiB63±37%
LFM2.5-8B-A1B-hereticMoEI1-Q6_K8.5B6.48 GiB0.05 GiB7.43 GiB0.01 GiB63±37%
LFM2.5-8B-A1B-SOMPOA-heresyMoEI1-Q6_K8.5B6.48 GiB0.05 GiB7.43 GiB0.01 GiB63±37%
Huihui-LFM2.5-8B-A1B-abliteratedMoEI1-Q6_K8.5B6.48 GiB0.05 GiB7.43 GiB0.01 GiB63±37%
Supertron2.1-8B-A1BMoEI1-Q6_K8.5B6.48 GiB0.05 GiB7.43 GiB0.01 GiB63±37%
LFM2.5-8B-A1BMoEQ6_K8.5B6.48 GiB0.05 GiB7.43 GiB0.01 GiB63±37%
Phi-4-reasoning-plusQ2_K_L14.7B5.63 GiB0.83 GiB7.42 GiB0.02 GiB22±26.5%
Phi-4-reasoningQ2_K_L14.7B5.63 GiB0.83 GiB7.42 GiB0.02 GiB22±26.5%
phi-4Q2_K_L14.7B5.63 GiB0.83 GiB7.42 GiB0.02 GiB22±26.5%
granite-4.1-8bQ5_K_M8.8B5.82 GiB0.66 GiB7.42 GiB0.02 GiB22±26.5%
Qwen3.6-35B-A3B-REAM-160-ru-agentMoEIQ2_XXS23.6B6.43 GiB0.08 GiB7.42 GiB0.02 GiB93±37%
SOLAR-10.7B-Instruct-v1.0-uncensoredQ4_K_S10.7B5.69 GiB0.80 GiB7.42 GiB0.02 GiB22±26.5%
Wan2.1-T2V-1.3BQ4_01.4B6.50 GiB0.00 GiB7.42 GiB0.02 GiB22±26.5%
Ling-mini-2.0MoEIQ3_XS16.3B6.35 GiB0.17 GiB7.41 GiB0.03 GiB97±37%
Qwythos-9B-v2Q4_K_L9.7B6.34 GiB0.13 GiB7.40 GiB0.04 GiB22±26.5%
Tess-4-9BQ4_K_L9.7B6.34 GiB0.13 GiB7.40 GiB0.04 GiB22±26.5%
salamandra-7b-instruct-2606I1-Q6_K7.8B5.94 GiB0.53 GiB7.40 GiB0.04 GiB22±26.5%
Qwen3.5-9B-BaseQ5_19.7B6.33 GiB0.13 GiB7.40 GiB0.04 GiB22±26.5%
Teuken-7B-instruct-research-v0.4Q6_K_L7.5B6.33 GiB0.13 GiB7.40 GiB0.04 GiB22±26.5%
LocateAnything-3BBF163.8B6.34 GiB0.15 GiB7.40 GiB0.04 GiB22±26.5%
deepseek-math-7b-instructQ5_K_S6.9B4.48 GiB1.99 GiB7.40 GiB0.04 GiB22±26.5%
deepseek-llm-7b-chatQ5_06.9B4.48 GiB1.99 GiB7.40 GiB0.04 GiB22±26.5%
Janus-Pro-7BI1-Q5_K_S7.4B4.48 GiB1.99 GiB7.40 GiB0.04 GiB22±26.5%
deepseek-coder-7b-instruct-v1.5I1-Q5_K_S6.9B4.48 GiB1.99 GiB7.40 GiB0.04 GiB22±26.5%
HomunculusQ3_K_M12.5B5.79 GiB0.66 GiB7.40 GiB0.04 GiB22±26.5%
gemma-4-12B-it-hereticIQ4_XS12.0B5.94 GiB0.51 GiB7.40 GiB0.04 GiB22±26.5%
Qwen2.5-3B-Instruct-abliteratedF163.1B6.33 GiB0.15 GiB7.40 GiB0.04 GiB22±26.5%
GRM-Kerlin-3bF163.4B6.33 GiB0.15 GiB7.40 GiB0.04 GiB22±26.5%
Smilodon-9B-v1I1-Q4_K_M10.2B5.37 GiB1.09 GiB7.39 GiB0.05 GiB22±26.5%
bella-bartender-v2I1-Q4_K_M9.2B5.37 GiB1.09 GiB7.39 GiB0.05 GiB22±26.5%
Gemma-The-Writer-9B-HERETIC-Uncensored-AbliteratedI1-Q4_K_M9.2B5.37 GiB1.09 GiB7.39 GiB0.05 GiB22±26.5%
Dirty-Muse-Writer-v01-Uncensored-Erotica-NSFWI1-Q4_K_M9.2B5.37 GiB1.09 GiB7.39 GiB0.05 GiB22±26.5%
Gemma-2-9B-It-SPPO-Iter3I1-Q4_K_M9.2B5.37 GiB1.09 GiB7.39 GiB0.05 GiB22±26.5%
Gemma-SEA-LION-v3-9B-ITI1-Q4_K_M9.2B5.37 GiB1.09 GiB7.39 GiB0.05 GiB22±26.5%
G2-Darkest-Writer-Dirty-Shirley-9B-v2I1-Q4_K_M9.2B5.37 GiB1.09 GiB7.39 GiB0.05 GiB22±26.5%
G2-Darkest-Writer-9B-v1I1-Q4_K_M9.2B5.37 GiB1.09 GiB7.39 GiB0.05 GiB22±26.5%
Tiger-Gemma-9B-v3I1-Q4_K_M9.2B5.37 GiB1.09 GiB7.39 GiB0.05 GiB22±26.5%
gemma-2-9bQ4_K_M9.2B5.37 GiB1.09 GiB7.39 GiB0.05 GiB22±26.5%
gemma-2-9b-it-abliteratedQ4_K_M9.2B5.37 GiB1.09 GiB7.39 GiB0.05 GiB22±26.5%
gemma-2-9b-itQ4_K_M9.2B5.37 GiB1.09 GiB7.39 GiB0.05 GiB22±26.5%
Tiger-Gemma-9B-v1Q4_K_M9.2B5.37 GiB1.09 GiB7.39 GiB0.05 GiB22±26.5%
magnum-v4-9bQ4_K_M9.2B5.37 GiB1.09 GiB7.39 GiB0.05 GiB22±26.5%
Garnet-OCR-3B-0422F164.1B6.33 GiB0.15 GiB7.39 GiB0.05 GiB22±26.5%
DeepSeek-R1-Distill-Llama-8B-AbliteratedI1-Q2_K8.0B5.92 GiB0.53 GiB7.39 GiB0.05 GiB22±26.5%
deepseek-coder-6.7b-instructQ5_06.7B4.33 GiB2.13 GiB7.38 GiB0.06 GiB22±26.5%
deepseek-coder-6.7b-baseQ5_06.7B4.33 GiB2.13 GiB7.38 GiB0.06 GiB22±26.5%
deepseek-coder-6.7B-kexerI1-Q5_K_S6.7B4.33 GiB2.13 GiB7.38 GiB0.06 GiB22±26.5%
Magicoder-S-DS-6.7BI1-Q5_K_S6.7B4.33 GiB2.13 GiB7.38 GiB0.06 GiB22±26.5%
MathCoder2-CodeLlama-7BQ5_K_S6.7B4.33 GiB2.13 GiB7.38 GiB0.06 GiB22±26.5%
CodeLlama-7b-instruct-hfQ5_06.7B4.33 GiB2.13 GiB7.38 GiB0.06 GiB22±26.5%
CodeLlama-7b-hfQ5_06.7B4.33 GiB2.13 GiB7.38 GiB0.06 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.

Questions people ask

What AI models can a Radeon Pro W6600 run?
1398 of 2118 indexed open-weight models fit a Radeon Pro W6600 at 8,192 context with q8_0 KV cache, the largest being INTELLECT-1-Instruct at I1-Q4_K_M. That covers text, vision-language, image, video and speech models.
How much usable memory does a Radeon Pro W6600 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 Pro W6600 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.