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. 833 of 2118 indexed models fit at 4K context with q8_0 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 690vision language 59audio asr 37video 2embedding 25audio tts 19image 1

What fits at 4K context

largest quantization that fits, per model · 833 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
gemma-4-E2B-it-ultra-uncensored-hereticI1-Q3_K_S5.1B2.90 GiB0.03 GiB3.72 GiB0.00 GiB21±30%
Barcenas-E2BI1-Q3_K_S5.1B2.90 GiB0.03 GiB3.72 GiB0.00 GiB21±30%
gemma-4-E2B-it-Uncensored-MAXI1-Q3_K_S5.1B2.90 GiB0.03 GiB3.72 GiB0.00 GiB21±30%
Firefly-v4I1-Q3_K_S5.1B2.90 GiB0.03 GiB3.72 GiB0.00 GiB21±30%
gemma-4-E2B-it-abliteratedI1-Q3_K_S5.1B2.90 GiB0.03 GiB3.72 GiB0.00 GiB21±30%
gemma-4-E2B-it-uncensoredQ3_K_S5.1B2.90 GiB0.03 GiB3.72 GiB0.00 GiB21±30%
gemma-4-E2B-it-heretic-araQ3_K_S5.1B2.90 GiB0.03 GiB3.72 GiB0.00 GiB21±30%
rp-model_E2B_v5.1I1-Q3_K_S5.1B2.90 GiB0.03 GiB3.72 GiB0.00 GiB21±30%
BartaLens-E2BI1-Q3_K_S5.1B2.90 GiB0.03 GiB3.72 GiB0.00 GiB21±30%
gemma-4-E2BQ3_K_S5.1B2.90 GiB0.03 GiB3.72 GiB0.00 GiB21±30%
SambaLingo-Japanese-ChatI1-IQ2_XXS6.9B1.83 GiB1.06 GiB3.71 GiB0.01 GiB21±30%
Aya-Medikal-V2I1-IQ2_XS8.0B2.60 GiB0.27 GiB3.71 GiB0.01 GiB21±30%
gemma-4-E2B-it-qat-q4_0-unquantized-hereticI1-IQ3_S5.1B2.89 GiB0.03 GiB3.71 GiB0.01 GiB21±30%
Huihui-gemma-4-E2B-it-qat-q4_0-unquantized-abliteratedI1-IQ3_S5.1B2.89 GiB0.03 GiB3.71 GiB0.01 GiB21±30%
Gemma4_E2B_Abliterated_Baked_HF_ReadyI1-IQ3_S5.1B2.89 GiB0.03 GiB3.71 GiB0.01 GiB21±30%
gemma-4-E2B-itIQ3_XS5.1B2.89 GiB0.03 GiB3.71 GiB0.01 GiB21±30%
Fara1.5-4BQ4_K_L4.5B2.83 GiB0.07 GiB3.71 GiB0.01 GiB21±30%
AREX-TurboQ4_K_L4.5B2.83 GiB0.07 GiB3.71 GiB0.01 GiB21±30%
OLMoE-1B-7B-0924-InstructMoEI1-IQ3_XS6.9B2.67 GiB0.27 GiB3.71 GiB0.01 GiB44±37%
granite-speech-4.1-2b-plusQ4_K2.1B2.75 GiB0.17 GiB3.70 GiB0.02 GiB21±30%
EVA-Yi-1.5-9B-32K-V1I1-IQ2_S8.8B2.68 GiB0.20 GiB3.70 GiB0.02 GiB21±30%
Miril-Drone-2B-1IQ3_XS5.1B2.88 GiB0.03 GiB3.70 GiB0.02 GiB21±30%
Olmo-3-7B-InstructUD-IQ1_S7.3B1.81 GiB1.06 GiB3.70 GiB0.02 GiB21±30%
Olmo-3-7B-ThinkUD-IQ1_S7.3B1.81 GiB1.06 GiB3.70 GiB0.02 GiB21±30%
umt5-xxlQ3_K_M5.7B2.85 GiB0.00 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%
MARTHA-LXVII.8B_QWEN-3.5-3.6_prune_9b-3.6_baseI1-IQ2_XS8.1B2.80 GiB0.06 GiB3.69 GiB0.03 GiB21±30%
granite-speech-4.1-2bQ4_K2.3B2.74 GiB0.17 GiB3.69 GiB0.03 GiB21±30%
LFM2-8B-A1BMoEQ2_K8.3B2.87 GiB0.02 GiB3.69 GiB0.03 GiB59±37%
InternVL3_5-8BIQ2_M8.5B2.84 GiB0.00 GiB3.69 GiB0.03 GiB21±30%
EXAONE-4.0-1.2B-abliteratedF161.5B2.78 GiB0.12 GiB3.69 GiB0.03 GiB21±30%
Voxtral-Mini-3B-2507Q5_K_S4.7B2.63 GiB0.25 GiB3.69 GiB0.03 GiB21±30%
Qwen3.5-4BQ4_K_M4.7B2.81 GiB0.07 GiB3.68 GiB0.04 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%
Llama-3.2-3B-Instruct-uncensoredQ5_K_L3.6B2.64 GiB0.23 GiB3.68 GiB0.04 GiB21±30%
Ministral-3-8B-Instruct-2512-BF16-abliteratedI1-IQ2_XS8.9B2.56 GiB0.28 GiB3.68 GiB0.04 GiB21±30%
Amaretto-8BI1-IQ2_XS8.9B2.56 GiB0.28 GiB3.68 GiB0.04 GiB21±30%
Foundation-Sec-8B-InstructI1-IQ2_S8.0B2.57 GiB0.27 GiB3.67 GiB0.05 GiB21±30%
Foundation-Sec-8B-Instruct-hereticI1-IQ2_S8.0B2.57 GiB0.27 GiB3.67 GiB0.05 GiB21±30%
Meta-Llama-3-8BIQ2_S8.0B2.57 GiB0.27 GiB3.67 GiB0.05 GiB21±30%
llama3.1-heretic-unsensoredI1-IQ2_S8.0B2.57 GiB0.27 GiB3.67 GiB0.05 GiB21±30%
Anubis-Mini-8B-v1I1-IQ2_S8.0B2.57 GiB0.27 GiB3.67 GiB0.05 GiB21±30%
dolphin-2.9-llama3-8bIQ2_S8.0B2.57 GiB0.27 GiB3.67 GiB0.05 GiB21±30%
Meta-Llama-3-8BIQ2_S8.0B2.57 GiB0.27 GiB3.67 GiB0.05 GiB21±30%
L3.1-Dark-Reasoning-LewdPlay-evo-Hermes-R1-Uncensored-8B-hereticI1-IQ2_S8.0B2.57 GiB0.27 GiB3.67 GiB0.05 GiB21±30%
L3.1-Dark-Reasoning-LewdPlay-evo-Hermes-R1-Uncensored-8BI1-IQ2_S8.0B2.57 GiB0.27 GiB3.67 GiB0.05 GiB21±30%
Gluon-8BI1-IQ2_S8.0B2.57 GiB0.27 GiB3.67 GiB0.05 GiB21±30%
Llama3.3-8B-Instruct-Thinking-Heretic-Uncensored-Claude-4.5-Opus-High-ReasoningI1-IQ2_S8.0B2.57 GiB0.27 GiB3.67 GiB0.05 GiB21±30%
Llama3.3-8B-Instruct-Thinking-Claude-4.5-Opus-High-ReasoningI1-IQ2_S8.0B2.57 GiB0.27 GiB3.67 GiB0.05 GiB21±30%
Hypnos-i1-8BI1-IQ2_S8.0B2.57 GiB0.27 GiB3.67 GiB0.05 GiB21±30%
grok-oss-Apollyon-8B-hereticI1-IQ2_S8.0B2.57 GiB0.27 GiB3.67 GiB0.05 GiB21±30%
Llama-3.3-8B-Instruct-128K-JbliteratedI1-IQ2_S8.0B2.57 GiB0.27 GiB3.67 GiB0.05 GiB21±30%
PsyCoPref-Llama3-8BI1-IQ2_S8.0B2.57 GiB0.27 GiB3.67 GiB0.05 GiB21±30%
Llama3.1-GptDeluxe-8BI1-IQ2_S8.0B2.57 GiB0.27 GiB3.67 GiB0.05 GiB21±30%
SelfCite-8B-CC-SFTI1-IQ2_S8.0B2.57 GiB0.27 GiB3.67 GiB0.05 GiB21±30%
ai-girlfriend-v2I1-IQ2_S8.0B2.57 GiB0.27 GiB3.67 GiB0.05 GiB21±30%
Llama-3.3-8B-Instruct-128K_AbliteratedI1-IQ2_S8.0B2.57 GiB0.27 GiB3.67 GiB0.05 GiB21±30%
Aisha-Uncensored-v2I1-IQ2_S8.0B2.57 GiB0.27 GiB3.67 GiB0.05 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?
833 of 2118 indexed open-weight models fit a Arc A370M 4GB at 4,096 context with q8_0 KV cache, the largest being gemma-4-E2B-it-ultra-uncensored-heretic at I1-Q3_K_S. 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.