Intel · consumer

Arc A350M 4GB

Arc A350M 4GB has 4 GB of VRAM at 112 GB/s — about 3.72 GiB usable after driver and compositor overhead. 348 of 2118 indexed models fit at 128K context with q4_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
35 W
Context4K8K16K32K64K128K
KV cachef16q8_0q4_0quantizing the KV cache is a ~2× lever on the dominant term at long context
text 255vision language 32audio asr 29video 2audio tts 17embedding 13

What fits at 128K context

largest quantization that fits, per model · 348 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
Gemma-3-4b-it-Uncensored-DBL-XI1-IQ3_M4.7B2.01 GiB0.90 GiB3.72 GiB0.00 GiB21±30%
deepseek-coder-5.7bmqa-baseQ3_K_S5.7B2.33 GiB0.56 GiB3.72 GiB0.00 GiB21±30%
Vikhr-Gemma-2B-instructIQ2_M2.6B1.01 GiB1.89 GiB3.72 GiB0.00 GiB21±30%
Gemmasutra-Mini-2B-v1I1-IQ2_M2.6B1.01 GiB1.89 GiB3.72 GiB0.00 GiB21±30%
GLM-OCRI1-Q6_K1.3B0.68 GiB2.25 GiB3.72 GiB0.00 GiB20±30%
Darwin-4B-ChimeraIQ3_M4.0B1.91 GiB0.99 GiB3.71 GiB0.01 GiB21±30%
Qwen3.5-4B-NSFW-ARA-Heretic-LiteroticaI1-IQ3_XXS4.2B1.77 GiB1.13 GiB3.71 GiB0.01 GiB21±30%
Qwen3.5-4B-RpRMax-v1I1-IQ3_XXS4.7B1.77 GiB1.13 GiB3.71 GiB0.01 GiB21±30%
Holo-3.1-4B-uncensored-hereticI1-IQ3_XXS4.5B1.77 GiB1.13 GiB3.71 GiB0.01 GiB21±30%
GRaPE-2-MiniI1-IQ3_XXS4.7B1.77 GiB1.13 GiB3.71 GiB0.01 GiB21±30%
Qwen3.5-DPO-4B-2I1-IQ3_XXS4.2B1.77 GiB1.13 GiB3.71 GiB0.01 GiB21±30%
Huihui-Qwen3.5-4B-Claude-4.6-Opus-abliteratedI1-IQ3_XXS4.7B1.77 GiB1.13 GiB3.71 GiB0.01 GiB21±30%
Qwopus3.5-4B-v3-hereticI1-IQ3_XXS4.5B1.77 GiB1.13 GiB3.71 GiB0.01 GiB21±30%
Aureth-4B-Qwen3.5I1-IQ3_XXS4.5B1.77 GiB1.13 GiB3.71 GiB0.01 GiB21±30%
GRM-Kerlin-3b-AbliteratedIQ4_XS3.1B1.63 GiB1.27 GiB3.71 GiB0.01 GiB21±30%
MATE-3BIQ4_XS3.1B1.63 GiB1.27 GiB3.71 GiB0.01 GiB21±30%
Mythos-nano-OBLITERATEDIQ4_XS3.1B1.63 GiB1.27 GiB3.71 GiB0.01 GiB21±30%
Qwen2.5-3B-Instruct-UncensoredIQ4_XS3.1B1.63 GiB1.27 GiB3.71 GiB0.01 GiB21±30%
VibeThinker-3B-OBLITERATEDIQ4_XS3.1B1.63 GiB1.27 GiB3.71 GiB0.01 GiB21±30%
Nanonets-OCR-sIQ4_XS3.8B1.63 GiB1.27 GiB3.71 GiB0.01 GiB21±30%
EXAONE-4.0-1.2B-abliteratedIQ4_XS1.5B0.81 GiB2.11 GiB3.70 GiB0.02 GiB21±30%
Dolphin3.0-Qwen2.5-3bIQ4_XS3.1B1.62 GiB1.27 GiB3.70 GiB0.02 GiB21±30%
Qwen2.5-Coder-3B-Instruct-abliteratedI1-IQ4_XS3.1B1.62 GiB1.27 GiB3.70 GiB0.02 GiB21±30%
Qwen2.5-Coder-3B-InstructIQ4_XS3.1B1.62 GiB1.27 GiB3.70 GiB0.02 GiB21±30%
Mythos-nanoI1-IQ4_XS3.1B1.62 GiB1.27 GiB3.70 GiB0.02 GiB21±30%
Qwen2.5-3B-InstructIQ4_XS3.1B1.62 GiB1.27 GiB3.70 GiB0.02 GiB21±30%
Qwen2.5-3BIQ4_XS3.1B1.62 GiB1.27 GiB3.70 GiB0.02 GiB21±30%
Qwen2.5-Coder-3BIQ4_XS3.1B1.62 GiB1.27 GiB3.70 GiB0.02 GiB21±30%
raspberry-3BIQ4_XS3.1B1.62 GiB1.27 GiB3.70 GiB0.02 GiB21±30%
Fourier-Qwen2.5-VL-3B-0.67I1-IQ4_XS3.8B1.62 GiB1.27 GiB3.70 GiB0.02 GiB21±30%
Qwen2.5-VL-3B-InstructIQ4_XS3.8B1.62 GiB1.27 GiB3.70 GiB0.02 GiB21±30%
jina-embeddings-v4IQ4_XS3.8B1.62 GiB1.27 GiB3.70 GiB0.02 GiB21±30%
umt5-xxlQ3_K_M5.7B2.85 GiB0.00 GiB3.70 GiB0.02 GiB21±30%
Hunyuan-1.8B-InstructIQ2_M1.8B0.65 GiB2.25 GiB3.69 GiB0.03 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%
Gemma-3-4B-VL-it-Gemini-Pro-Heretic-Uncensored-ThinkingIQ4_XS4.3B2.12 GiB0.75 GiB3.69 GiB0.03 GiB21±30%
gemma-3-4b-it-roleplay-tuned-v1IQ4_XS4.3B2.12 GiB0.75 GiB3.69 GiB0.03 GiB21±30%
gemma-3-4b-it-roleplay-tuned-v2IQ4_XS4.3B2.12 GiB0.75 GiB3.69 GiB0.03 GiB21±30%
medgemma-1.5-4b-itIQ4_XS4.3B2.12 GiB0.75 GiB3.69 GiB0.03 GiB21±30%
gemma-3-4b-it-heretic-uncensored-abliterated-ExtremeIQ4_XS4.3B2.12 GiB0.75 GiB3.69 GiB0.03 GiB21±30%
gemma-3-4b-it-abliteratedIQ4_XS4.3B2.12 GiB0.75 GiB3.69 GiB0.03 GiB21±30%
Gemma3-4B-CodeCenturionIQ4_XS4.3B2.12 GiB0.75 GiB3.69 GiB0.03 GiB21±30%
InternVL3_5-8BIQ2_M8.5B2.84 GiB0.00 GiB3.69 GiB0.03 GiB21±30%
LFM2-8B-A1BMoEIQ2_M8.3B2.47 GiB0.42 GiB3.69 GiB0.03 GiB38±37%
LocateAnything-3BQ3_K_M3.8B1.61 GiB1.27 GiB3.69 GiB0.03 GiB21±30%
GRM-Kerlin-3bI1-Q3_K_M3.4B1.61 GiB1.27 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%
Garnet-OCR-3B-0422I1-Q3_K_M4.1B1.61 GiB1.27 GiB3.68 GiB0.04 GiB21±30%
granite-4.0-h-3b-arI1-Q6_K3.4B2.60 GiB0.28 GiB3.68 GiB0.04 GiB21±30%
alduin-4b-it-baseI1-IQ4_XS4.3B2.12 GiB0.75 GiB3.68 GiB0.04 GiB21±30%
medgemma-4b-itIQ4_XS4.3B2.11 GiB0.75 GiB3.68 GiB0.04 GiB21±30%
amoral-gemma3-4B-v1IQ4_XS4.3B2.11 GiB0.75 GiB3.68 GiB0.04 GiB21±30%
ArrowMint-Gemma3-4B-YUKI-v0.1I1-IQ4_XS4.3B2.11 GiB0.75 GiB3.68 GiB0.04 GiB21±30%
gemma-3-4b-itIQ4_XS4.3B2.11 GiB0.75 GiB3.68 GiB0.04 GiB21±30%
starcoder2-3bKV unresolvedQ4_13.0B1.80 GiB1.05 GiB3.67 GiB0.05 GiB21±30%
VibeThinker-3BQ3_K_L3.1B1.59 GiB1.27 GiB3.67 GiB0.05 GiB21±30%
Wan2.2-TI2V-5B-TurboQ4_05.0B2.83 GiB0.00 GiB3.66 GiB0.06 GiB21±30%
DeepSeek-OCR-2MoEQ4_K_M3.4B1.82 GiB1.05 GiB3.66 GiB0.06 GiB26±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 A350M 4GB run?
348 of 2118 indexed open-weight models fit a Arc A350M 4GB at 131,072 context with q4_0 KV cache, the largest being Gemma-3-4b-it-Uncensored-DBL-X at I1-IQ3_M. That covers text, vision-language, image, video and speech models.
How much usable memory does a Arc A350M 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 A350M 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.