Intel · consumer

Arc A770M 16GB

Arc A770M 16GB has 16 GB of VRAM at 512 GB/s — about 14.88 GiB usable after driver and compositor overhead. 1685 of 2118 indexed models fit at 64K context with q8_0 KV.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed
Memory
16 GB
GDDR6
Bandwidth
512 GB/s
256-bit bus
Tensor FP16
—
dense
TDP
150 W
Context4K8K16K32K64K128K
KV cachef16q8_0q4_0quantizing the KV cache is a ~2× lever on the dominant term at long context
text 1431vision language 152audio asr 39video 15embedding 26audio tts 21image 1

What fits at 64K context

largest quantization that fits, per model · 1685 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
Pantheon-Reasoning-27BI1-IQ3_M27.8B11.89 GiB2.13 GiB14.88 GiB0.00 GiB20±30%
Qwen3.6-27B-uncensored-heretic-v2-Native-MTP-PreservedI1-IQ3_M27.4B11.89 GiB2.13 GiB14.88 GiB0.00 GiB20±30%
Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-MTPI1-IQ3_M27.8B11.89 GiB2.13 GiB14.88 GiB0.00 GiB20±30%
Qwen3.6-27B-Fable-5-ExperimentalI1-IQ3_M27.8B11.89 GiB2.13 GiB14.88 GiB0.00 GiB20±30%
Qwable-5-27B-CoderI1-IQ3_M27.8B11.89 GiB2.13 GiB14.88 GiB0.00 GiB20±30%
Qwen3.6-27B-AEON-Ultimate-Uncensored-BF16IQ3_M27.4B11.89 GiB2.13 GiB14.88 GiB0.00 GiB20±30%
EVE-27b-XENO-HAT-DeepSeek-V4-FlashI1-IQ3_M27.8B11.89 GiB2.13 GiB14.88 GiB0.00 GiB20±30%
EVE-27B-XENO-HATI1-IQ3_M27.8B11.89 GiB2.13 GiB14.88 GiB0.00 GiB20±30%
Godoter-27BI1-IQ3_M27.8B11.89 GiB2.13 GiB14.88 GiB0.00 GiB20±30%
Reasoning-Medical-27BI1-IQ3_M27.8B11.89 GiB2.13 GiB14.88 GiB0.00 GiB20±30%
Qwopus3.6-27B-v2-abliteratedI1-IQ3_M27.4B11.89 GiB2.13 GiB14.88 GiB0.00 GiB20±30%
Qwen3.6-27B-Uncensored-HauhauCS-Aggressive-BF16I1-IQ3_M27.8B11.89 GiB2.13 GiB14.88 GiB0.00 GiB20±30%
Reasoning-Medical0.1-27BI1-IQ3_M27.8B11.89 GiB2.13 GiB14.88 GiB0.00 GiB20±30%
Huihui-ThinkingCap-Qwen3.6-27B-abliteratedI1-IQ3_M27.4B11.89 GiB2.13 GiB14.88 GiB0.00 GiB20±30%
Semancer-27BI1-IQ3_M27.8B11.89 GiB2.13 GiB14.88 GiB0.00 GiB20±30%
Qwen3.6-27B-Omnimerge-v4IQ3_M27.8B11.89 GiB2.13 GiB14.88 GiB0.00 GiB20±30%
Darwin-28B-CoderI1-IQ3_M26.9B11.89 GiB2.13 GiB14.88 GiB0.00 GiB20±30%
Qwen3.6-35B-A3B-REAM-160-ru-agentMoEQ4_K_M23.6B13.41 GiB0.66 GiB14.88 GiB0.00 GiB67±37%
NuExtract-1.5Q2_K3.8B1.32 GiB12.75 GiB14.87 GiB0.01 GiB20±30%
Phi-3.5-mini-instructQ2_K3.8B1.32 GiB12.75 GiB14.87 GiB0.01 GiB20±30%
Phi-3.5-mini-instruct_UncensoredQ2_K3.8B1.32 GiB12.75 GiB14.87 GiB0.01 GiB20±30%
Phi-3-mini-128k-instructQ2_K3.8B1.32 GiB12.75 GiB14.87 GiB0.01 GiB20±30%
Phi-3-mini-4k-instructQ2_K3.8B1.32 GiB12.75 GiB14.87 GiB0.01 GiB20±30%
octo-netQ2_K3.8B1.32 GiB12.75 GiB14.87 GiB0.01 GiB20±30%
Voxtral-Small-24B-2507IQ3_XXS24.3B8.64 GiB5.31 GiB14.87 GiB0.01 GiB20±30%
Devstral-Small-2-24B-Instruct-2512IQ3_XXS24.0B8.64 GiB5.31 GiB14.87 GiB0.01 GiB20±30%
Transformed-Journey-24BI1-IQ3_XXS23.6B8.64 GiB5.31 GiB14.87 GiB0.01 GiB20±30%
Magistry-24B-v1.1I1-IQ3_XXS23.6B8.64 GiB5.31 GiB14.87 GiB0.01 GiB20±30%
Mergedonia-AETHER-24B-v1aI1-IQ3_XXS23.6B8.64 GiB5.31 GiB14.87 GiB0.01 GiB20±30%
Mergedonia-AETHER-24B-v1bI1-IQ3_XXS23.6B8.64 GiB5.31 GiB14.87 GiB0.01 GiB20±30%
Slimaki-Tavern-24B-v1.3I1-IQ3_XXS23.6B8.64 GiB5.31 GiB14.87 GiB0.01 GiB20±30%
Maginum-Cydoms-24BI1-IQ3_XXS23.6B8.64 GiB5.31 GiB14.87 GiB0.01 GiB20±30%
Maginum-Cydoms-24B-absolute-heresyI1-IQ3_XXS23.6B8.64 GiB5.31 GiB14.87 GiB0.01 GiB20±30%
Morax-24B-v2IQ3_XXS23.6B8.64 GiB5.31 GiB14.87 GiB0.01 GiB20±30%
Mistral-Small-3.2-24B-Instruct-2506-ultra-uncensored-hereticI1-IQ3_XXS24.0B8.64 GiB5.31 GiB14.87 GiB0.01 GiB20±30%
Huihui-Mistral-Small-3.2-24B-Instruct-2506-abliterated-llamacppfixedI1-IQ3_XXS24.0B8.64 GiB5.31 GiB14.87 GiB0.01 GiB20±30%
Dans-PersonalityEngine-V1.2.0-24bI1-IQ3_XXS23.6B8.64 GiB5.31 GiB14.87 GiB0.01 GiB20±30%
Mistral-Small-3_2-24B-Instruct-2506-antislop.v2I1-IQ3_XXS24.0B8.64 GiB5.31 GiB14.87 GiB0.01 GiB20±30%
Dans-PersonalityEngine-V1.3.0-24bI1-IQ3_XXS23.6B8.64 GiB5.31 GiB14.87 GiB0.01 GiB20±30%
Cydonia_VistralIQ3_XXS23.6B8.64 GiB5.31 GiB14.87 GiB0.01 GiB20±30%
Dolphin3.0-Mistral-24BI1-IQ3_XXS23.6B8.64 GiB5.31 GiB14.87 GiB0.01 GiB20±30%
Goetia-24B-v1.1I1-IQ3_XXS23.6B8.64 GiB5.31 GiB14.87 GiB0.01 GiB20±30%
Devstral-Small-2505IQ3_XXS23.6B8.64 GiB5.31 GiB14.87 GiB0.01 GiB20±30%
Mistral-Small-3.2-24B-Instruct-2506IQ3_XXS24.0B8.64 GiB5.31 GiB14.87 GiB0.01 GiB20±30%
MS3.2-PaintedFantasy-v3-24BI1-IQ3_XXS23.6B8.64 GiB5.31 GiB14.87 GiB0.01 GiB20±30%
RP-Spectrum-24BI1-IQ3_XXS23.6B8.64 GiB5.31 GiB14.87 GiB0.01 GiB20±30%
MS3.2-PaintedFantasy-v4.1-24B-ultra-uncensored-heretic-v2I1-IQ3_XXS23.6B8.64 GiB5.31 GiB14.87 GiB0.01 GiB20±30%
Magidonia-24B-v4.3-heretic-v1.2I1-IQ3_XXS23.6B8.64 GiB5.31 GiB14.87 GiB0.01 GiB20±30%
Magidonia-24B-v4.3-absolute-heresyI1-IQ3_XXS23.6B8.64 GiB5.31 GiB14.87 GiB0.01 GiB20±30%
MagiSeek-Pro-V1I1-IQ3_XXS23.6B8.64 GiB5.31 GiB14.87 GiB0.01 GiB20±30%
Cogidonia-v2-24BI1-IQ3_XXS23.6B8.64 GiB5.31 GiB14.87 GiB0.01 GiB20±30%
Magidonia-24B-v4.3I1-IQ3_XXS—8.64 GiB5.31 GiB14.87 GiB0.01 GiB20±30%
Precog-24B-v1I1-IQ3_XXS—8.64 GiB5.31 GiB14.87 GiB0.01 GiB20±30%
experiment024bI1-IQ3_XXS23.6B8.64 GiB5.31 GiB14.87 GiB0.01 GiB20±30%
Magidonia-24B-v4.2.0IQ3_XXS23.6B8.64 GiB5.31 GiB14.87 GiB0.01 GiB20±30%
Berthier-Mistral-Military-24BI1-IQ3_XXS24.0B8.64 GiB5.31 GiB14.87 GiB0.01 GiB20±30%
MS-2501-DPE-QwQify-v0.1-24BIQ3_XXS23.6B8.64 GiB5.31 GiB14.87 GiB0.01 GiB20±30%
Mistral-Small-3.2-24B-Instruct-2506-llamacppfixedI1-IQ3_XXS24.0B8.64 GiB5.31 GiB14.87 GiB0.01 GiB20±30%
Cydonia-24B-v4.3-absolute-heresyI1-IQ3_XXS23.6B8.64 GiB5.31 GiB14.87 GiB0.01 GiB20±30%
Cydonia-24B-v4.3-heretic-v2I1-IQ3_XXS23.6B8.64 GiB5.31 GiB14.87 GiB0.01 GiB20±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 A770M 16GB run?
1685 of 2118 indexed open-weight models fit a Arc A770M 16GB at 65,536 context with q8_0 KV cache, the largest being Pantheon-Reasoning-27B at I1-IQ3_M. That covers text, vision-language, image, video and speech models.
How much usable memory does a Arc A770M 16GB actually have?
Its nameplate is 16 GB, but about 14.88 GiB is available to a model once driver and compositor overhead is accounted for.
Is a Arc A770M 16GB fast for local AI?
Its memory bandwidth is 512 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.