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. 1760 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
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 1511vision language 147audio tts 21audio asr 39image 2embedding 26

What fits at 16K context

largest quantization that fits, per model · 1760 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%
Hermes-4-14BQ5_014.8B9.58 GiB0.70 GiB11.15 GiB0.01 GiB18±30%
Darwin-35B-A3B-OpusMoEIQ2_S36.0B10.25 GiB0.09 GiB11.15 GiB0.01 GiB88±37%
Aurora-Code-1MoEIQ2_S34.7B10.25 GiB0.09 GiB11.15 GiB0.01 GiB88±37%
grug-35b-v2MoEIQ2_S35.1B10.25 GiB0.09 GiB11.15 GiB0.01 GiB88±37%
grug-35bMoEIQ2_S35.1B10.25 GiB0.09 GiB11.15 GiB0.01 GiB88±37%
WorldSim-Opus-3.6-35B-A3BMoEIQ2_S35.1B10.25 GiB0.09 GiB11.15 GiB0.01 GiB88±37%
Qwen3.6-35B-A3B-AnkoMoEIQ2_S35.1B10.25 GiB0.09 GiB11.15 GiB0.01 GiB88±37%
KAT-Coder-V2.5-DevMoEIQ2_S34.7B10.25 GiB0.09 GiB11.15 GiB0.01 GiB88±37%
Ornith-1.0-35BMoEIQ2_S34.7B10.25 GiB0.09 GiB11.15 GiB0.01 GiB88±37%
Nex-N2-miniMoEIQ2_S35.1B10.25 GiB0.09 GiB11.15 GiB0.01 GiB88±37%
Gemma-4-12B-StyleTuneI1-Q6_K13.0B9.88 GiB0.41 GiB11.14 GiB0.02 GiB18±30%
gemma-4-12b-heretic-styletune-headI1-Q6_K12.0B9.88 GiB0.41 GiB11.14 GiB0.02 GiB18±30%
syrian-gemma-12bI1-Q6_K13.0B9.88 GiB0.41 GiB11.14 GiB0.02 GiB18±30%
Qwen3.5-27B-Engineer-Deckard-GeminiI1-IQ3_XXS27.7B10.00 GiB0.28 GiB11.14 GiB0.02 GiB18±30%
Qwen3.5-27B-HERETIC-Polaris-Advanced-Thinking-Alpha-uncensoredI1-IQ3_XXS27.4B10.00 GiB0.28 GiB11.14 GiB0.02 GiB18±30%
Qwen3.5-27B-Deckard-PKD-Heretic-Uncensored-ThinkingI1-IQ3_XXS27.4B10.00 GiB0.28 GiB11.14 GiB0.02 GiB18±30%
Huihui-Qwen3.5-27B-abliteratedI1-IQ3_XXS27.8B10.00 GiB0.28 GiB11.14 GiB0.02 GiB18±30%
Qwen3.5-27B-Unredacted-MAXI1-IQ3_XXS27.4B10.00 GiB0.28 GiB11.14 GiB0.02 GiB18±30%
Qwen3.5-27B-hereticI1-IQ3_XXS27.4B10.00 GiB0.28 GiB11.14 GiB0.02 GiB18±30%
Qwen3.5-27B-DerestrictedI1-IQ3_XXS27.8B10.00 GiB0.28 GiB11.14 GiB0.02 GiB18±30%
Qwen3.5-27B-Claude-4.6-Opus-Reasoning-DistilledI1-IQ3_XXS27.8B10.00 GiB0.28 GiB11.14 GiB0.02 GiB18±30%
NSFW_13B_sftQ3_K_L13.3B6.77 GiB3.52 GiB11.13 GiB0.03 GiB18±30%
Goetia-26B-A4B-v1.4MoEI1-Q2_K26.0B10.08 GiB0.26 GiB11.13 GiB0.03 GiB18±30%
G4-Moonlight-Dusk-26B-A4B-hereticMoEI1-Q2_K26.5B10.08 GiB0.26 GiB11.13 GiB0.03 GiB18±30%
Pantheon-Reasoning-26B-A4B-1.1-hereticMoEI1-Q2_K26.5B10.08 GiB0.26 GiB11.13 GiB0.03 GiB18±30%
G4-Moonlight-Dusk-26B-A4BMoEI1-Q2_K26.5B10.08 GiB0.26 GiB11.13 GiB0.03 GiB18±30%
Chimera-X-26B-A4BMoEI1-Q2_K26.5B10.08 GiB0.26 GiB11.13 GiB0.03 GiB18±30%
Pantheon-Reasoning-26B-A4B-1.1MoEI1-Q2_K26.5B10.08 GiB0.26 GiB11.13 GiB0.03 GiB18±30%
Gemma-4-26B-A4B-StyleTune-V2MoEI1-Q2_K26.5B10.08 GiB0.26 GiB11.13 GiB0.03 GiB18±30%
Gemma-4-26B-A4B-StyleTuneMoEI1-Q2_K26.5B10.08 GiB0.26 GiB11.13 GiB0.03 GiB18±30%
gemma-4-26b-a4b-heretic-styletune-v2-headMoEI1-Q2_K25.8B10.08 GiB0.26 GiB11.13 GiB0.03 GiB18±30%
Trinity-MiniMoEIQ3_XS26.1B10.23 GiB0.10 GiB11.13 GiB0.03 GiB70±37%
L3-DARKEST-PLANET-16.5BQ4_K_S16.5B9.03 GiB1.25 GiB11.12 GiB0.04 GiB18±30%
NousCoder-14BQ5_K_S14.8B9.56 GiB0.70 GiB11.12 GiB0.04 GiB18±30%
spoomplesmaxx-mini-14BI1-Q5_K_S14.8B9.56 GiB0.70 GiB11.12 GiB0.04 GiB18±30%
vanilla-cn-roleplay-0.2I1-Q5_K_S14.8B9.56 GiB0.70 GiB11.12 GiB0.04 GiB18±30%
Claria-14bI1-Q5_K_S14.8B9.56 GiB0.70 GiB11.12 GiB0.04 GiB18±30%
qwen3-14b-code-reasoning-conversationalQ5_K_S14.8B9.56 GiB0.70 GiB11.12 GiB0.04 GiB18±30%
NTX-2.1-ProI1-Q5_K_S14.8B9.56 GiB0.70 GiB11.12 GiB0.04 GiB18±30%
Qwen3-14B-UncensoredI1-Q5_K_S14.8B9.56 GiB0.70 GiB11.12 GiB0.04 GiB18±30%
Qwen3-14BQ5_K_S14.8B9.56 GiB0.70 GiB11.12 GiB0.04 GiB18±30%
FrogMini-14B-2510I1-Q5_K_S—9.56 GiB0.70 GiB11.12 GiB0.04 GiB18±30%
Qwen3-14B-abliteratedQ5_K_S14.8B9.56 GiB0.70 GiB11.12 GiB0.04 GiB18±30%
Josiefied-Qwen3-14B-abliterated-v3Q5_K_S14.8B9.56 GiB0.70 GiB11.12 GiB0.04 GiB18±30%
Slava-Qwen3-14B-SerbianI1-Q5_K_S14.8B9.56 GiB0.70 GiB11.12 GiB0.04 GiB18±30%
Qwen3-14B-BaseQ5_K_S14.8B9.56 GiB0.70 GiB11.12 GiB0.04 GiB18±30%
Huihui-Qwen3-14B-abliterated-v2I1-Q5_K_S14.8B9.56 GiB0.70 GiB11.12 GiB0.04 GiB18±30%
Qwen3.6-27B-Heretic2-ThinkingI1-Q2_K27.4B9.98 GiB0.28 GiB11.12 GiB0.04 GiB18±30%
Qwen3.6-27B-Uncensored-AggressiveI1-Q2_K27.4B9.98 GiB0.28 GiB11.12 GiB0.04 GiB18±30%
Qwen-3.5-Opus-GLM-27BI1-Q2_K26.9B9.98 GiB0.28 GiB11.12 GiB0.04 GiB18±30%
Qwen3.6-27B-abliteratedI1-Q2_K27.4B9.98 GiB0.28 GiB11.12 GiB0.04 GiB18±30%
KoQweopus-3.5-27B-experimentalI1-Q2_K27.8B9.98 GiB0.28 GiB11.12 GiB0.04 GiB18±30%
Webcoda-AI-27BI1-Q2_K27.4B9.98 GiB0.28 GiB11.12 GiB0.04 GiB18±30%
Qwen3.5-27B-imabari-v2I1-Q2_K27.8B9.98 GiB0.28 GiB11.12 GiB0.04 GiB18±30%
Qwen3.6-27B-AEON-Ultimate-Uncensored-BF16Q2_K27.4B9.98 GiB0.28 GiB11.12 GiB0.04 GiB18±30%
Qwen3.5-27B-uncensored-heretic-v1I1-Q2_K27.4B9.98 GiB0.28 GiB11.12 GiB0.04 GiB18±30%
Carnice-V2-27bI1-Q2_K27.4B9.98 GiB0.28 GiB11.12 GiB0.04 GiB18±30%
●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?
1760 of 2118 indexed open-weight models fit a Arc A730M 12GB at 16,384 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.