AMD · consumer

Radeon RX 9070 GRE

Radeon RX 9070 GRE has 12 GB of VRAM at 432 GB/s — about 11.16 GiB usable after driver and compositor overhead. 1755 of 2118 indexed models fit at 16K context with q4_0 KV.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed
Memory
12 GB
GDDR6
Bandwidth
432 GB/s
192-bit bus
Tensor FP16
—
dense
TDP
220 W
Context4K8K16K32K64K128K
KV cachef16q8_0q4_0quantizing the KV cache is a ~2× lever on the dominant term at long context
text 1507vision language 146audio tts 21video 14audio asr 39image 2embedding 26

What fits at 16K context

largest quantization that fits, per model · 1755 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
Marco-Mini-InstructMoEI1-Q4_K_M17.3B9.78 GiB0.49 GiB11.15 GiB0.01 GiB100±37%
Qwen3-15B-A2B-BaseMoEQ5_K_S15.6B10.04 GiB0.21 GiB11.15 GiB0.01 GiB98±37%
grug-27bIQ2_M27.4B9.90 GiB0.28 GiB11.15 GiB0.01 GiB27±26.5%
Carnice-V2-27bIQ2_M27.4B9.90 GiB0.28 GiB11.15 GiB0.01 GiB27±26.5%
Fara1.5-27BIQ2_M27.4B9.90 GiB0.28 GiB11.15 GiB0.01 GiB27±26.5%
Magistry-24B-v1.1IQ3_XS23.6B9.43 GiB0.70 GiB11.15 GiB0.01 GiB27±26.5%
Llama3.2-30B-A3B-II-Dark-Champion-INSTRUCT-Heretic-Abliterated-UncensoredMoEI1-IQ2_M30.0B9.41 GiB0.83 GiB11.14 GiB0.02 GiB59±37%
DeepSeek-V2-Lite-ChatMoEQ5_015.7B10.10 GiB0.13 GiB11.14 GiB0.02 GiB86±37%
Tiger-Gemma-12B-v3Q6_K12.8B9.77 GiB0.41 GiB11.13 GiB0.03 GiB27±26.5%
AfriqueGemma-12BI1-Q6_K12.2B9.77 GiB0.41 GiB11.13 GiB0.03 GiB27±26.5%
gemma-2-27b-itQ2_K_S27.2B9.06 GiB1.04 GiB11.12 GiB0.04 GiB27±26.5%
Llama3.2-24B-A3B-II-Dark-Champion-INSTRUCT-Heretic-Abliterated-UncensoredMoEI1-Q4_K_S18.0B9.72 GiB0.49 GiB11.12 GiB0.04 GiB65±37%
Grug-12BQ6_K_L12.0B9.76 GiB0.41 GiB11.12 GiB0.04 GiB27±26.5%
gemma-4-12B-it-Esper4Q6_K_L12.0B9.76 GiB0.41 GiB11.12 GiB0.04 GiB27±26.5%
gemma-4-12B-itQ6_K_L12.0B9.76 GiB0.41 GiB11.12 GiB0.04 GiB27±26.5%
Apriel-1.6-15b-ThinkerI1-Q5_K_S14.9B9.33 GiB0.84 GiB11.12 GiB0.04 GiB27±26.5%
Salience-1.5-FlashMoEI1-Q2_K_S31.1B9.80 GiB0.42 GiB11.11 GiB0.05 GiB88±37%
Huihui-Qwen3-VL-30B-A3B-Instruct-abliteratedMoEI1-Q2_K_S31.1B9.80 GiB0.42 GiB11.11 GiB0.05 GiB88±37%
Qwen3-30B-A3B-Gemini-Pro-High-Reasoning-2507-ABLITERATED-UNCENSOREDMoEI1-Q2_K_S30.5B9.80 GiB0.42 GiB11.11 GiB0.05 GiB88±37%
MiroThinker-v1.0-30BMoEI1-Q2_K_S30.5B9.80 GiB0.42 GiB11.11 GiB0.05 GiB88±37%
Qwen3-30B-A3B-YOYO-V5MoEI1-Q2_K_S30.5B9.80 GiB0.42 GiB11.11 GiB0.05 GiB88±37%
Qwen3-30B-A3B-Thinking-2507-Claude-4.5-Sonnet-High-Reasoning-DistillMoEI1-Q2_K_S30.5B9.80 GiB0.42 GiB11.11 GiB0.05 GiB88±37%
Huihui-Qwen3-30B-A3B-Thinking-2507-abliteratedMoEI1-Q2_K_S30.5B9.80 GiB0.42 GiB11.11 GiB0.05 GiB88±37%
Huihui-Qwen3-30B-A3B-Instruct-2507-abliteratedMoEI1-Q2_K_S30.5B9.80 GiB0.42 GiB11.11 GiB0.05 GiB88±37%
gemma-4-26B-A4B-itMoEIQ2_M26.5B9.97 GiB0.26 GiB11.11 GiB0.05 GiB26±26.5%
Huihui-Qwen3-Coder-30B-A3B-Instruct-abliteratedMoEI1-Q2_K_S30.5B9.80 GiB0.42 GiB11.11 GiB0.05 GiB88±37%
Qwen3-Coder-30B-A3B-Instruct-RTPurboMoEI1-Q2_K_S30.5B9.80 GiB0.42 GiB11.11 GiB0.05 GiB88±37%
Mellum2-12B-A2.5B-ThinkingMoEQ6_K12.1B10.13 GiB0.08 GiB11.11 GiB0.05 GiB76±37%
reka-flash-3.1I1-IQ3_M20.9B9.55 GiB0.58 GiB11.11 GiB0.05 GiB27±26.5%
reka-flash-3IQ3_M20.9B9.55 GiB0.58 GiB11.11 GiB0.05 GiB27±26.5%
Fallen-Gemma3-27B-v1Q2_K27.4B9.78 GiB0.39 GiB11.11 GiB0.05 GiB27±26.5%
Skyfall-31B-v4.2-hereticI1-IQ2_S31.4B9.14 GiB0.95 GiB11.11 GiB0.05 GiB27±26.5%
Skyfall-31B-v4.2I1-IQ2_S31.4B9.14 GiB0.95 GiB11.11 GiB0.05 GiB27±26.5%
Goetia-26B-A4B-v1.4MoEI1-IQ2_M26.0B9.96 GiB0.26 GiB11.11 GiB0.05 GiB26±26.5%
G4-Moonlight-Dusk-26B-A4B-hereticMoEI1-IQ2_M26.5B9.96 GiB0.26 GiB11.11 GiB0.05 GiB26±26.5%
Pantheon-Reasoning-26B-A4B-1.1-hereticMoEI1-IQ2_M26.5B9.96 GiB0.26 GiB11.11 GiB0.05 GiB26±26.5%
G4-Moonlight-Dusk-26B-A4BMoEI1-IQ2_M26.5B9.96 GiB0.26 GiB11.11 GiB0.05 GiB26±26.5%
Chimera-X-26B-A4BMoEI1-IQ2_M26.5B9.96 GiB0.26 GiB11.11 GiB0.05 GiB26±26.5%
Pantheon-Reasoning-26B-A4B-1.1MoEI1-IQ2_M26.5B9.96 GiB0.26 GiB11.11 GiB0.05 GiB26±26.5%
Gemma-4-26B-A4B-StyleTune-V2MoEI1-IQ2_M26.5B9.96 GiB0.26 GiB11.11 GiB0.05 GiB26±26.5%
Gemma-4-26B-A4B-StyleTuneMoEI1-IQ2_M26.5B9.96 GiB0.26 GiB11.11 GiB0.05 GiB26±26.5%
gemma-4-26b-a4b-heretic-styletune-v2-headMoEI1-IQ2_M25.8B9.96 GiB0.26 GiB11.11 GiB0.05 GiB26±26.5%
InternVL3_5-30B-A3BQ2_K30.8B10.16 GiB0.00 GiB11.10 GiB0.06 GiB27±26.5%
LFM2-24B-A2BMoEQ3_K_M23.8B10.10 GiB0.09 GiB11.10 GiB0.06 GiB105±37%
ERNIE-4.5-21B-A3B-ThinkingQ3_K_L21.8B9.93 GiB0.25 GiB11.10 GiB0.06 GiB27±26.5%
ERNIE-4.5-21B-A3B-PTQ3_K_L21.9B9.93 GiB0.25 GiB11.10 GiB0.06 GiB27±26.5%
MythoMax-L2-Kimiko-v2-13bQ3_K_L13.0B6.63 GiB3.52 GiB11.09 GiB0.07 GiB27±26.5%
MythoMax-L2-13bI1-Q3_K_L13.0B6.63 GiB3.52 GiB11.09 GiB0.07 GiB27±26.5%
IQuest-Coder-V1-40B-InstructI1-IQ1_M39.8B8.68 GiB1.41 GiB11.08 GiB0.08 GiB27±26.5%
v6-Finch-7B-HFQ8_07.6B7.88 GiB2.25 GiB11.08 GiB0.08 GiB27±26.5%
rwkv-6-world-7bQ8_07.6B7.88 GiB2.25 GiB11.08 GiB0.08 GiB27±26.5%
Muse-Glimmer-30BUD-IQ2_XXS29.8B10.01 GiB0.08 GiB11.08 GiB0.08 GiB27±26.5%
gemma-4-19B-A4B-it-INSTRUCT-Heretic-UncensoredMoEI1-Q4_019.0B9.92 GiB0.26 GiB11.06 GiB0.10 GiB27±26.5%
gemma-4-19B-A4B-it-The-DECKARD-Heretic-Uncensored-ThinkingMoEI1-Q4_019.0B9.92 GiB0.26 GiB11.06 GiB0.10 GiB27±26.5%
gemma-4-19b-a4b-it-REAP-hereticMoEI1-Q4_019.0B9.92 GiB0.26 GiB11.06 GiB0.10 GiB27±26.5%
Gemma-4-19BMoEI1-Q4_019.0B9.92 GiB0.26 GiB11.06 GiB0.10 GiB27±26.5%
dolphin-2.6-mistral-7bQ5_K_M7.2B9.56 GiB0.56 GiB11.06 GiB0.10 GiB27±26.5%
DeepSeek-R1-Distill-Llama-8B-AbliteratedI1-Q4_18.0B9.56 GiB0.56 GiB11.06 GiB0.10 GiB27±26.5%
OpenHermes-2.5-Mistral-7BKV unresolvedQ5_K_M7.2B9.56 GiB0.56 GiB11.06 GiB0.10 GiB27±26.5%
Darwin-35B-A3B-OpusMoEIQ2_XS36.0B10.06 GiB0.09 GiB11.05 GiB0.11 GiB131±37%
●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 9070 GRE run?
1755 of 2118 indexed open-weight models fit a Radeon RX 9070 GRE at 16,384 context with q4_0 KV cache, the largest being Marco-Mini-Instruct at I1-Q4_K_M. That covers text, vision-language, image, video and speech models.
How much usable memory does a Radeon RX 9070 GRE actually have?
Its nameplate is 12 GB, but about 11.16 GiB is available to a model once driver and compositor overhead is accounted for.
Is a Radeon RX 9070 GRE fast for local AI?
Its memory bandwidth is 432 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.