Intel · consumer

Arc A370M 4GB

Arc A370M 4GB has 4 GB of VRAM at 112 GB/s — about 3.72 GiB usable after driver and compositor overhead. 285 of 2118 indexed models fit at 64K context with f16 KV.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed
Memory
4 GB
GDDR6
Bandwidth
112 GB/s
64-bit bus
Tensor FP16
—
dense
TDP
50 W
Context4K8K16K32K64K128K
KV cachef16q8_0q4_0quantizing the KV cache is a ~2× lever on the dominant term at long context
text 203vision language 23audio asr 29video 2audio tts 15embedding 13

What fits at 64K context

largest quantization that fits, per model · 285 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
LFM2.5-Queen-Opus-4.7-8B-A1BMoEI1-IQ2_XXS8.5B2.16 GiB0.75 GiB3.71 GiB0.01 GiB30±37%
LFM2.5-8B-A1B-KO-SFTMoEI1-IQ2_XXS8.5B2.16 GiB0.75 GiB3.71 GiB0.01 GiB30±37%
LFM2.5-8B-A1B-hereticMoEI1-IQ2_XXS8.5B2.16 GiB0.75 GiB3.71 GiB0.01 GiB30±37%
LFM2.5-8B-A1B-SOMPOA-heresyMoEI1-IQ2_XXS8.5B2.16 GiB0.75 GiB3.71 GiB0.01 GiB30±37%
Huihui-LFM2.5-8B-A1B-abliteratedMoEI1-IQ2_XXS8.5B2.16 GiB0.75 GiB3.71 GiB0.01 GiB30±37%
Supertron2.1-8B-A1BMoEI1-IQ2_XXS8.5B2.16 GiB0.75 GiB3.71 GiB0.01 GiB30±37%
Dolphin3.0-Llama3.2-1BQ5_K_L1.2B0.91 GiB2.00 GiB3.71 GiB0.01 GiB21±30%
OneLLM-Doey-ChatQA-V1-Llama-3.2-1BQ5_K_L1.2B0.91 GiB2.00 GiB3.71 GiB0.01 GiB21±30%
Llama-3.2-1B-InstructQ5_K_L1.2B0.91 GiB2.00 GiB3.71 GiB0.01 GiB21±30%
Gemma-3-4b-it-Uncensored-DBL-XI1-IQ1_M4.7B1.20 GiB1.69 GiB3.70 GiB0.02 GiB21±30%
gemma-4-E2B-itIQ2_M5.1B2.44 GiB0.46 GiB3.70 GiB0.02 GiB21±30%
umt5-xxlQ3_K_M5.7B2.85 GiB0.00 GiB3.70 GiB0.02 GiB21±30%
medgemma-1.5-4b-itUD-IQ2_M4.3B1.46 GiB1.42 GiB3.70 GiB0.02 GiB21±30%
medgemma-4b-itUD-IQ2_M4.3B1.46 GiB1.42 GiB3.70 GiB0.02 GiB21±30%
LFM2-1.2BQ6_K_M1.2B0.90 GiB2.00 GiB3.70 GiB0.02 GiB21±30%
whisper-mediumF32764M2.85 GiB0.00 GiB3.69 GiB0.03 GiB21±30%
whisper-medium.enF32764M2.85 GiB0.00 GiB3.69 GiB0.03 GiB21±30%
Trinity-Nano-PreviewMoEIQ2_M6.1B1.94 GiB0.98 GiB3.69 GiB0.03 GiB28±37%
SEX_ROLEPLAY-3.2-1BI1-Q4_K_M1.5B0.89 GiB2.00 GiB3.69 GiB0.03 GiB21±30%
Llama-3.2-1B-Instruct-abliteratedI1-Q4_K_M1.5B0.89 GiB2.00 GiB3.69 GiB0.03 GiB21±30%
Novaciano-3.2-1BI1-Q4_K_M1.5B0.89 GiB2.00 GiB3.69 GiB0.03 GiB21±30%
Imp-RPG.System-1BI1-Q4_K_M1.5B0.89 GiB2.00 GiB3.69 GiB0.03 GiB21±30%
Miril-Drone-2B-1IQ2_M5.1B2.43 GiB0.46 GiB3.69 GiB0.03 GiB21±30%
InternVL3_5-8BIQ2_M8.5B2.84 GiB0.00 GiB3.69 GiB0.03 GiB21±30%
gemma-3n-E2B-itIQ3_M5.4B2.08 GiB0.80 GiB3.69 GiB0.03 GiB21±30%
EXAONE-Deep-7.8BQ2_K7.8B2.84 GiB0.00 GiB3.68 GiB0.04 GiB21±30%
EXAONE-3.5-7.8B-InstructQ2_K7.8B2.84 GiB0.00 GiB3.68 GiB0.04 GiB21±30%
granite-4.0-7B-A1B-Creative-v0.1MoEI1-IQ3_XXS6.7B2.42 GiB0.50 GiB3.68 GiB0.04 GiB38±37%
alduin-4b-it-baseI1-IQ2_M4.3B1.44 GiB1.42 GiB3.68 GiB0.04 GiB21±30%
SmolVLM2-500M-Video-InstructQ8_0507M0.41 GiB2.50 GiB3.68 GiB0.04 GiB21±30%
SmolVLM-500M-InstructQ8_0507M0.41 GiB2.50 GiB3.68 GiB0.04 GiB21±30%
granite-4.0-h-tiny-baseMoEQ2_K6.9B2.41 GiB0.50 GiB3.68 GiB0.04 GiB38±37%
LFM2-2.6B-ExpQ5_K_L2.6B1.87 GiB1.00 GiB3.68 GiB0.04 GiB21±30%
FableForge-1.5BQ5_K_M1.5B1.12 GiB1.75 GiB3.67 GiB0.05 GiB21±30%
gemma-2bQ5_K2.5B1.71 GiB1.13 GiB3.67 GiB0.05 GiB21±30%
gemma-3-4b-it-roleplay-tuned-v1I1-IQ2_M4.3B1.43 GiB1.42 GiB3.67 GiB0.05 GiB21±30%
gemma-3-4b-it-roleplay-tuned-v2I1-IQ2_M4.3B1.43 GiB1.42 GiB3.67 GiB0.05 GiB21±30%
gemma-3-4b-it-heretic-uncensored-abliterated-ExtremeI1-IQ2_M4.3B1.43 GiB1.42 GiB3.67 GiB0.05 GiB21±30%
amoral-gemma3-4B-v1IQ2_M4.3B1.43 GiB1.42 GiB3.67 GiB0.05 GiB21±30%
gemma-3-4b-it-abliteratedIQ2_M4.3B1.43 GiB1.42 GiB3.67 GiB0.05 GiB21±30%
Gemma3-4B-CodeCenturionI1-IQ2_M4.3B1.43 GiB1.42 GiB3.67 GiB0.05 GiB21±30%
ArrowMint-Gemma3-4B-YUKI-v0.1I1-IQ2_M4.3B1.43 GiB1.42 GiB3.67 GiB0.05 GiB21±30%
gemma-3-4b-itIQ2_M4.3B1.43 GiB1.42 GiB3.67 GiB0.05 GiB21±30%
Wan2.2-TI2V-5B-TurboQ4_05.0B2.83 GiB0.00 GiB3.66 GiB0.06 GiB21±30%
t5-v1_1-xxlQ4_04.8B2.82 GiB0.00 GiB3.66 GiB0.06 GiB21±30%
Wan2.2-TI2V-5BQ4_05.0B2.82 GiB0.00 GiB3.66 GiB0.06 GiB21±30%
Qwen2.5-Omni-7BQ2_K10.7B2.81 GiB0.00 GiB3.65 GiB0.07 GiB21±30%
Dolphin3.0-Qwen2.5-1.5BQ5_K_L1.5B1.10 GiB1.75 GiB3.65 GiB0.07 GiB21±30%
Qwen2.5-Math-1.5B-InstructQ5_K_L1.5B1.10 GiB1.75 GiB3.65 GiB0.07 GiB21±30%
Qwen2.5-Coder-1.5B-InstructQ5_K_L1.5B1.10 GiB1.75 GiB3.65 GiB0.07 GiB21±30%
Qwen2.5-1.5B-InstructQ5_K_L1.5B1.10 GiB1.75 GiB3.65 GiB0.07 GiB21±30%
Qwen2-VL-2B-InstructQ5_K_L2.2B1.10 GiB1.75 GiB3.65 GiB0.07 GiB21±30%
Qwen2.5-1.5BQ5_11.5B1.10 GiB1.75 GiB3.65 GiB0.07 GiB21±30%
Llama-3.2-1B-Instruct-hereticI1-Q5_K_M1.2B0.85 GiB2.00 GiB3.65 GiB0.07 GiB21±30%
Llama-3.2-1BQ5_K_M1.2B0.85 GiB2.00 GiB3.65 GiB0.07 GiB21±30%
Llama-3.2-1B-Instruct-UncensoredI1-Q5_K_M1.2B0.85 GiB2.00 GiB3.65 GiB0.07 GiB21±30%
s2-proMoEQ3_K4.6B2.82 GiB0.00 GiB3.64 GiB0.08 GiB12±37%
privacy-filter-multilingualMoEKV unresolvedF161.4B2.62 GiB0.25 GiB3.64 GiB0.08 GiB64±37%
privacy-filter-nemotronMoEKV unresolvedF161.4B2.62 GiB0.25 GiB3.64 GiB0.08 GiB64±37%
gemma-4-E2B-it-abliteratedI1-IQ2_M5.1B2.38 GiB0.46 GiB3.64 GiB0.08 GiB21±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 A370M 4GB run?
285 of 2118 indexed open-weight models fit a Arc A370M 4GB at 65,536 context with f16 KV cache, the largest being LFM2.5-Queen-Opus-4.7-8B-A1B at I1-IQ2_XXS. That covers text, vision-language, image, video and speech models.
How much usable memory does a Arc A370M 4GB actually have?
Its nameplate is 4 GB, but about 3.72 GiB is available to a model once driver and compositor overhead is accounted for.
Is a Arc A370M 4GB fast for local AI?
Its memory bandwidth is 112 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.
Arc A370M 4GB — what AI models can it run locally? — ossmodeldb