Intel · consumer

Arc A730M 12GB

Arc A730M 12GB has 12 GB of VRAM at 336 GB/s — about 11.16 GiB usable after driver and compositor overhead. 1714 of 2118 indexed models fit at 32K context with q4_0 KV.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed
Memory
12 GB
GDDR6
Bandwidth
336 GB/s
192-bit bus
Tensor FP16
—
dense
TDP
120 W
Context4K8K16K32K64K128K
KV cachef16q8_0q4_0quantizing the KV cache is a ~2× lever on the dominant term at long context
video 14text 1468vision language 144audio asr 39audio tts 21image 2embedding 26

What fits at 32K context

largest quantization that fits, per model · 1714 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
Wan2.1-FLF2V-14B-720PQ4_116.4B10.32 GiB0.00 GiB11.16 GiB0.00 GiB18±30%
Wan2.1-I2V-14B-480PQ4_116.4B10.32 GiB0.00 GiB11.15 GiB0.01 GiB18±30%
Wan2.1-I2V-14B-720PQ4_116.4B10.32 GiB0.00 GiB11.15 GiB0.01 GiB18±30%
Gemma-3-27B-MeditronFOI1-IQ2_M28.8B9.41 GiB0.87 GiB11.15 GiB0.01 GiB18±30%
gemma-4-19B-A4B-it-INSTRUCT-Heretic-UncensoredMoEI1-Q4_019.0B9.92 GiB0.43 GiB11.14 GiB0.02 GiB18±30%
gemma-4-19B-A4B-it-The-DECKARD-Heretic-Uncensored-ThinkingMoEI1-Q4_019.0B9.92 GiB0.43 GiB11.14 GiB0.02 GiB18±30%
gemma-4-19b-a4b-it-REAP-hereticMoEI1-Q4_019.0B9.92 GiB0.43 GiB11.14 GiB0.02 GiB18±30%
Gemma-4-19BMoEI1-Q4_019.0B9.92 GiB0.43 GiB11.14 GiB0.02 GiB18±30%
DeepSeek-R1-Distill-Llama-8B-AbliteratedI1-Q4_K_M8.0B9.17 GiB1.13 GiB11.13 GiB0.03 GiB18±30%
Llama-3.1-Nemotron-Nano-8B-v1Q4_K_M8.0B9.17 GiB1.13 GiB11.13 GiB0.03 GiB18±30%
GLM-4-32B-0414-Korean-CultureI1-IQ2_S32.6B9.70 GiB0.54 GiB11.13 GiB0.03 GiB18±30%
GLM-Z1-32B-0414IQ2_S32.6B9.70 GiB0.54 GiB11.13 GiB0.03 GiB18±30%
GLM-4-32B-0414IQ2_S32.6B9.70 GiB0.54 GiB11.13 GiB0.03 GiB18±30%
Bielik-11B-v2.3-InstructQ6_K11.2B8.53 GiB1.76 GiB11.13 GiB0.03 GiB18±30%
Ling-liteMoEQ4_116.8B9.84 GiB0.49 GiB11.13 GiB0.03 GiB49±37%
Gemma-4-31B-Isometry-RPI1-IQ2_XXS32.7B8.51 GiB1.74 GiB11.13 GiB0.03 GiB18±30%
Gemma-4-Dark-Gemistry-31BI1-IQ2_XXS32.7B8.51 GiB1.74 GiB11.13 GiB0.03 GiB18±30%
Prosopon-31BI1-IQ2_XXS32.7B8.51 GiB1.74 GiB11.13 GiB0.03 GiB18±30%
Gemma-4-Novelist-Eclipse-31BI1-IQ2_XXS32.7B8.51 GiB1.74 GiB11.13 GiB0.03 GiB18±30%
Giftige-Blume-31B-v1-StyleSwapI1-IQ2_XXS32.7B8.51 GiB1.74 GiB11.13 GiB0.03 GiB18±30%
G4-MeroMero-31B-StyleSwapI1-IQ2_XXS32.7B8.51 GiB1.74 GiB11.13 GiB0.03 GiB18±30%
Gemma-4-31B-StyleTune-heretic-araI1-IQ2_XXS32.7B8.51 GiB1.74 GiB11.13 GiB0.03 GiB18±30%
Pantheon-Reasoning-31B-1.1I1-IQ2_XXS32.7B8.51 GiB1.74 GiB11.13 GiB0.03 GiB18±30%
Gemma-4-31B-StyleTuneI1-IQ2_XXS32.7B8.51 GiB1.74 GiB11.13 GiB0.03 GiB18±30%
Barcenas-StyleTune-31B-FableI1-IQ2_XXS32.1B8.51 GiB1.74 GiB11.13 GiB0.03 GiB18±30%
Goetia-26B-A4B-v1.3-Absolute-Heretic-ARAMoEI1-Q2_K_S25.8B9.89 GiB0.43 GiB11.12 GiB0.04 GiB18±30%
Frank-26B-A4BMoEI1-Q2_K_S26.5B9.89 GiB0.43 GiB11.12 GiB0.04 GiB18±30%
G4-MeroMero-26B-A4B-it-uncensored-hereticMoEI1-Q2_K_S25.8B9.89 GiB0.43 GiB11.12 GiB0.04 GiB18±30%
EVE-26b-XENO-HATMoEI1-Q2_K_S25.8B9.89 GiB0.43 GiB11.12 GiB0.04 GiB18±30%
Gemma-4-26B-A4B-Animus-V14.1-FFT-hereticMoEI1-Q2_K_S25.8B9.89 GiB0.43 GiB11.12 GiB0.04 GiB18±30%
gemma-4-26B-A4B-it-qat-q4_0-unquantized-hereticMoEI1-Q2_K_S25.8B9.89 GiB0.43 GiB11.12 GiB0.04 GiB18±30%
Huihui-gemma-4-26B-A4B-it-qat-q4_0-unquantized-abliteratedMoEI1-Q2_K_S26.5B9.89 GiB0.43 GiB11.12 GiB0.04 GiB18±30%
G4-MeroMero-26B-A4BMoEI1-Q2_K_S25.8B9.89 GiB0.43 GiB11.12 GiB0.04 GiB18±30%
G4-Dark-Soul-26B-A4BMoEI1-Q2_K_S25.8B9.89 GiB0.43 GiB11.12 GiB0.04 GiB18±30%
gemma-4-26B-A4B-it-local-abliterated-sota-internal-t34MoEI1-Q2_K_S25.8B9.89 GiB0.43 GiB11.12 GiB0.04 GiB18±30%
gemma-4-26B-A4B-it-SOMPOA-heresyMoEI1-Q2_K_S25.8B9.89 GiB0.43 GiB11.12 GiB0.04 GiB18±30%
gemma-4-26B-A4B-it-hereticMoEI1-Q2_K_S25.8B9.89 GiB0.43 GiB11.12 GiB0.04 GiB18±30%
gemma-4-26B-A4B-it-abliterixMoEI1-Q2_K_S25.8B9.89 GiB0.43 GiB11.12 GiB0.04 GiB18±30%
gemma-4-26B-A4B-it-heretic-ara-v2MoEI1-Q2_K_S25.8B9.89 GiB0.43 GiB11.12 GiB0.04 GiB18±30%
Gemma-4-26B-A4B-it-heretic-antislopMoEI1-Q2_K_S25.8B9.89 GiB0.43 GiB11.12 GiB0.04 GiB18±30%
gemma-4-26B-A4B-it-Claude-Opus-Distill-v2MoEI1-Q2_K_S26.5B9.89 GiB0.43 GiB11.12 GiB0.04 GiB18±30%
gemma-4-26B-A4B-Heretic-StableMoEI1-Q2_K_S25.8B9.89 GiB0.43 GiB11.12 GiB0.04 GiB18±30%
gemma-4-26B-A4B-it-Uncensored-MAXMoEI1-Q2_K_S25.8B9.89 GiB0.43 GiB11.12 GiB0.04 GiB18±30%
gemma-4-26B-A4B-it-ultra-uncensored-hereticMoEI1-Q2_K_S25.8B9.89 GiB0.43 GiB11.12 GiB0.04 GiB18±30%
gemma-4-26B-A4B-it-ara-abliteratedMoEI1-Q2_K_S25.8B9.89 GiB0.43 GiB11.12 GiB0.04 GiB18±30%
Huihui-gemma-4-26B-A4B-it-abliteratedMoEI1-Q2_K_S26.5B9.89 GiB0.43 GiB11.12 GiB0.04 GiB18±30%
Gemma-4-26B-A4B-AbliteratedMoEI1-Q2_K_S25.8B9.89 GiB0.43 GiB11.12 GiB0.04 GiB18±30%
gemma4-26b-fiction-bf16MoEI1-Q2_K_S25.8B9.89 GiB0.43 GiB11.12 GiB0.04 GiB18±30%
gemma-4-26B-A4B-it-heretic-araMoEI1-Q2_K_S25.8B9.89 GiB0.43 GiB11.12 GiB0.04 GiB18±30%
ERNIE-4.5-21B-A3B-ThinkingQ3_K_M21.8B9.80 GiB0.49 GiB11.11 GiB0.05 GiB18±30%
ERNIE-4.5-21B-A3B-PTQ3_K_M21.9B9.80 GiB0.49 GiB11.11 GiB0.05 GiB18±30%
dolphin-2.6-mixtral-8x7bMoEI1-IQ1_S46.7B9.15 GiB1.13 GiB11.11 GiB0.05 GiB27±37%
xLAM-8x7b-rMoEIQ1_S46.7B9.15 GiB1.13 GiB11.11 GiB0.05 GiB27±37%
Salience-1.5-FlashMoEI1-IQ2_M31.1B9.47 GiB0.84 GiB11.11 GiB0.05 GiB49±37%
Huihui-Qwen3-VL-30B-A3B-Instruct-abliteratedMoEI1-IQ2_M31.1B9.47 GiB0.84 GiB11.11 GiB0.05 GiB49±37%
Qwen3-30B-A3B-Gemini-Pro-High-Reasoning-2507-ABLITERATED-UNCENSOREDMoEI1-IQ2_M30.5B9.47 GiB0.84 GiB11.11 GiB0.05 GiB49±37%
MiroThinker-v1.0-30BMoEI1-IQ2_M30.5B9.47 GiB0.84 GiB11.11 GiB0.05 GiB49±37%
Qwen3-30B-A3B-YOYO-V5MoEI1-IQ2_M30.5B9.47 GiB0.84 GiB11.11 GiB0.05 GiB49±37%
Qwen3-30B-A3B-Thinking-2507-Claude-4.5-Sonnet-High-Reasoning-DistillMoEI1-IQ2_M30.5B9.47 GiB0.84 GiB11.11 GiB0.05 GiB49±37%
Huihui-Qwen3-30B-A3B-Thinking-2507-abliteratedMoEI1-IQ2_M30.5B9.47 GiB0.84 GiB11.11 GiB0.05 GiB49±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 Arc A730M 12GB run?
1714 of 2118 indexed open-weight models fit a Arc A730M 12GB at 32,768 context with q4_0 KV cache, the largest being Wan2.1-FLF2V-14B-720P at Q4_1. That covers text, vision-language, image, video and speech models.
How much usable memory does a Arc A730M 12GB 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 Arc A730M 12GB fast for local AI?
Its memory bandwidth is 336 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.