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. 1777 of 2118 indexed models fit at 8K 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 1528vision language 147audio asr 39audio tts 21video 14image 2embedding 26

What fits at 8K context

largest quantization that fits, per model · 1777 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
EuroLLM-22B-Instruct-2512IQ3_M22.6B9.72 GiB0.47 GiB11.16 GiB0.00 GiB27±26.5%
EVA-abliterated-TIES-Qwen2.5-14BI1-Q5_K_M14.8B9.79 GiB0.42 GiB11.16 GiB0.00 GiB26±26.5%
Neuron-V1-14B-InstructI1-Q5_K_M14.8B9.79 GiB0.42 GiB11.16 GiB0.00 GiB26±26.5%
Ektome-Qwen2.5-Coder-14B-Instruct-PristinelyUncensoredI1-Q5_K_M14.8B9.79 GiB0.42 GiB11.16 GiB0.00 GiB26±26.5%
Qwen2.5-14B-Instruct-1M-abliteratedI1-Q5_K_M14.8B9.79 GiB0.42 GiB11.16 GiB0.00 GiB26±26.5%
DeepCoder-14B-PreviewQ5_K_M14.8B9.79 GiB0.42 GiB11.16 GiB0.00 GiB26±26.5%
Deepseeker-Kunou-Qwen2.5-14bI1-Q5_K_M14.8B9.79 GiB0.42 GiB11.16 GiB0.00 GiB26±26.5%
SuperNova-MediusQ5_K_M14.8B9.79 GiB0.42 GiB11.16 GiB0.00 GiB26±26.5%
14B-Qwen2.5-Kunou-v1I1-Q5_K_M14.8B9.79 GiB0.42 GiB11.16 GiB0.00 GiB26±26.5%
Sugoi-14B-Ultra-HFI1-Q5_K_M14.8B9.79 GiB0.42 GiB11.16 GiB0.00 GiB26±26.5%
Qwen2.5-14B-Instruct-abliterated-v2Q5_K_M14.8B9.79 GiB0.42 GiB11.16 GiB0.00 GiB26±26.5%
Qwen2.5-14B-Instruct-UncensoredQ5_K_M14.8B9.79 GiB0.42 GiB11.16 GiB0.00 GiB26±26.5%
Qwen2.5-Coder-14B-Instruct-abliteratedQ5_K_M14.8B9.79 GiB0.42 GiB11.16 GiB0.00 GiB26±26.5%
OpenCodeReasoning-Nemotron-14BQ5_K_M14.8B9.79 GiB0.42 GiB11.16 GiB0.00 GiB26±26.5%
Qwen2.5-14B-InstructQ5_K_M14.8B9.79 GiB0.42 GiB11.16 GiB0.00 GiB26±26.5%
DeepSeek-R1-Distill-Qwen-14B-abliterated-v2I1-Q5_K_M14.8B9.79 GiB0.42 GiB11.16 GiB0.00 GiB26±26.5%
C1-TachuI1-Q5_K_M14.8B9.79 GiB0.42 GiB11.16 GiB0.00 GiB26±26.5%
DeepSeek-R1-Distill-Qwen-14B-abliteratedI1-Q5_K_M14.8B9.79 GiB0.42 GiB11.16 GiB0.00 GiB26±26.5%
0x-liteQ5_K_M14.8B9.79 GiB0.42 GiB11.16 GiB0.00 GiB26±26.5%
Tessera-4I1-Q5_K_M14.8B9.79 GiB0.42 GiB11.16 GiB0.00 GiB26±26.5%
AceReason-Nemotron-14BQ5_K_M14.8B9.79 GiB0.42 GiB11.16 GiB0.00 GiB26±26.5%
Qwen2.5-14B-InstructQ5_K_M14.8B9.79 GiB0.42 GiB11.16 GiB0.00 GiB26±26.5%
FinetunedQwen14BQ5_K_M14.8B9.79 GiB0.42 GiB11.16 GiB0.00 GiB26±26.5%
Tessera-4.1I1-Q5_K_M14.8B9.79 GiB0.42 GiB11.16 GiB0.00 GiB26±26.5%
Qwen2.5-14B-Instruct-1MQ5_K_M14.8B9.79 GiB0.42 GiB11.16 GiB0.00 GiB26±26.5%
Qwen2.5-Coder-14BQ5_K_M14.8B9.79 GiB0.42 GiB11.16 GiB0.00 GiB26±26.5%
DeepSeek-R1-Distill-Qwen-14BQ5_K_M14.8B9.79 GiB0.42 GiB11.16 GiB0.00 GiB26±26.5%
Strand-Rust-Coder-14B-v1Q5_K_M14.8B9.79 GiB0.42 GiB11.16 GiB0.00 GiB26±26.5%
UwU-14B-Math-v0.2I1-Q5_K_M14.8B9.79 GiB0.42 GiB11.16 GiB0.00 GiB26±26.5%
EVA-Qwen2.5-14B-v0.2I1-Q5_K_M14.8B9.79 GiB0.42 GiB11.16 GiB0.00 GiB26±26.5%
EVA-Qwen2.5-14B-v0.0I1-Q5_K_M14.8B9.79 GiB0.42 GiB11.16 GiB0.00 GiB26±26.5%
EVA-Qwen2.5-14B-v0.1I1-Q5_K_M14.8B9.79 GiB0.42 GiB11.16 GiB0.00 GiB26±26.5%
oxy-1-smallQ5_K_M14.8B9.79 GiB0.42 GiB11.16 GiB0.00 GiB26±26.5%
Impish_QWEN_14B-1MI1-Q5_K_M14.8B9.79 GiB0.42 GiB11.16 GiB0.00 GiB26±26.5%
Qwen2.5-14BQ5_K_M14.8B9.78 GiB0.42 GiB11.15 GiB0.01 GiB26±26.5%
Lamarck-14B-v0.7I1-Q5_K_M14.8B9.78 GiB0.42 GiB11.15 GiB0.01 GiB26±26.5%
QwenStock-14BI1-Q5_K_M14.8B9.78 GiB0.42 GiB11.15 GiB0.01 GiB26±26.5%
DeepSeek-R1-Distill-Qwen-14B-UncensoredI1-Q5_K_M14.8B9.78 GiB0.42 GiB11.15 GiB0.01 GiB26±26.5%
Kimi-Linear-48B-A3B-InstructMoEIQ1_M49.1B10.17 GiB0.07 GiB11.15 GiB0.01 GiB26±26.5%
Goetia-26B-A4B-v1.4MoEI1-Q2_K26.0B10.08 GiB0.17 GiB11.14 GiB0.02 GiB26±26.5%
G4-Moonlight-Dusk-26B-A4B-hereticMoEI1-Q2_K26.5B10.08 GiB0.17 GiB11.14 GiB0.02 GiB26±26.5%
Pantheon-Reasoning-26B-A4B-1.1-hereticMoEI1-Q2_K26.5B10.08 GiB0.17 GiB11.14 GiB0.02 GiB26±26.5%
G4-Moonlight-Dusk-26B-A4BMoEI1-Q2_K26.5B10.08 GiB0.17 GiB11.14 GiB0.02 GiB26±26.5%
Chimera-X-26B-A4BMoEI1-Q2_K26.5B10.08 GiB0.17 GiB11.14 GiB0.02 GiB26±26.5%
Pantheon-Reasoning-26B-A4B-1.1MoEI1-Q2_K26.5B10.08 GiB0.17 GiB11.14 GiB0.02 GiB26±26.5%
Gemma-4-26B-A4B-StyleTune-V2MoEI1-Q2_K26.5B10.08 GiB0.17 GiB11.14 GiB0.02 GiB26±26.5%
Gemma-4-26B-A4B-StyleTuneMoEI1-Q2_K26.5B10.08 GiB0.17 GiB11.14 GiB0.02 GiB26±26.5%
gemma-4-26b-a4b-heretic-styletune-v2-headMoEI1-Q2_K25.8B10.08 GiB0.17 GiB11.14 GiB0.02 GiB26±26.5%
Gemma-4-Gembrain-X-Core-31BI1-IQ2_S31.3B9.46 GiB0.68 GiB11.12 GiB0.04 GiB27±26.5%
Gemma-4-Gembrain-X-31BI1-IQ2_S31.3B9.46 GiB0.68 GiB11.12 GiB0.04 GiB27±26.5%
Gemma-4-31B-Isometry-Fabled-PersonaI1-IQ2_S31.3B9.46 GiB0.68 GiB11.12 GiB0.04 GiB27±26.5%
Versipellis-31BI1-IQ2_S31.3B9.46 GiB0.68 GiB11.12 GiB0.04 GiB27±26.5%
Gemma4-Gutenberg-31BI1-IQ2_S31.3B9.46 GiB0.68 GiB11.12 GiB0.04 GiB27±26.5%
G4-MeroMero-31B-uncensored-hereticI1-IQ2_S31.3B9.46 GiB0.68 GiB11.12 GiB0.04 GiB27±26.5%
Gemma-4-Novelist-31BI1-IQ2_S31.3B9.46 GiB0.68 GiB11.12 GiB0.04 GiB27±26.5%
Wanabi-Gemma4-31BI1-IQ2_S31.3B9.46 GiB0.68 GiB11.12 GiB0.04 GiB27±26.5%
G4-Alice-v1.2-31BI1-IQ2_S31.3B9.46 GiB0.68 GiB11.12 GiB0.04 GiB27±26.5%
Agares-31B-v1I1-IQ2_S30.7B9.46 GiB0.68 GiB11.12 GiB0.04 GiB27±26.5%
Gemma4-Gutenberg-31B-HereticI1-IQ2_S31.3B9.46 GiB0.68 GiB11.12 GiB0.04 GiB27±26.5%
gemma-4-Ortenzya-The-Creative-Wordsmith-31B-it-uncensored-hereticI1-IQ2_S31.3B9.46 GiB0.68 GiB11.12 GiB0.04 GiB27±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 9070 GRE run?
1777 of 2118 indexed open-weight models fit a Radeon RX 9070 GRE at 8,192 context with q4_0 KV cache, the largest being EuroLLM-22B-Instruct-2512 at IQ3_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.