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. 359 of 2118 indexed models fit at 64K 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
35 W
Context4K8K16K32K64K128K
KV cachef16q8_0q4_0quantizing the KV cache is a ~2× lever on the dominant term at long context
text 265vision language 33audio asr 29video 2audio tts 17embedding 13

What fits at 64K context

largest quantization that fits, per model · 359 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%
Darwin-4B-ChimeraI1-IQ3_M4.0B1.83 GiB1.08 GiB3.72 GiB0.00 GiB21±30%
Dolphin3.0-Qwen2.5-3bQ4_K_S3.1B1.71 GiB1.20 GiB3.72 GiB0.00 GiB21±30%
Qwen2.5-Coder-3B-Instruct-abliteratedI1-Q4_K_S3.1B1.71 GiB1.20 GiB3.72 GiB0.00 GiB21±30%
GRM-Kerlin-3b-AbliteratedI1-Q4_K_S3.1B1.71 GiB1.20 GiB3.72 GiB0.00 GiB21±30%
Qwen2.5-Coder-3B-InstructQ4_K_S3.1B1.71 GiB1.20 GiB3.72 GiB0.00 GiB21±30%
Mythos-nanoI1-Q4_K_S3.1B1.71 GiB1.20 GiB3.72 GiB0.00 GiB21±30%
MATE-3BI1-Q4_K_S3.1B1.71 GiB1.20 GiB3.72 GiB0.00 GiB21±30%
Mythos-nano-OBLITERATEDI1-Q4_K_S3.1B1.71 GiB1.20 GiB3.72 GiB0.00 GiB21±30%
Qwen2.5-3B-Instruct-UncensoredI1-Q4_K_S3.1B1.71 GiB1.20 GiB3.72 GiB0.00 GiB21±30%
Nanonets-OCR-sQ4_K_S3.8B1.71 GiB1.20 GiB3.72 GiB0.00 GiB21±30%
Qwen2.5-3B-InstructQ4_K_S3.1B1.71 GiB1.20 GiB3.72 GiB0.00 GiB21±30%
Qwen2.5-Coder-3BQ4_K_S3.1B1.71 GiB1.20 GiB3.72 GiB0.00 GiB21±30%
raspberry-3BQ4_K_S3.1B1.71 GiB1.20 GiB3.72 GiB0.00 GiB21±30%
VibeThinker-3B-OBLITERATEDI1-Q4_K_S3.1B1.71 GiB1.20 GiB3.72 GiB0.00 GiB21±30%
VibeThinker-3BQ4_K_S3.1B1.71 GiB1.20 GiB3.72 GiB0.00 GiB21±30%
Fourier-Qwen2.5-VL-3B-0.67I1-Q4_K_S3.8B1.71 GiB1.20 GiB3.72 GiB0.00 GiB21±30%
Qwen2.5-VL-3B-InstructQ4_K_S3.8B1.71 GiB1.20 GiB3.72 GiB0.00 GiB21±30%
LFM2.5-8B-A1BMoEUD-IQ2_XXS8.5B2.52 GiB0.40 GiB3.71 GiB0.01 GiB39±37%
Qwen2.5-3BIQ4_NL3.1B1.70 GiB1.20 GiB3.71 GiB0.01 GiB21±30%
jina-embeddings-v4IQ4_NL3.8B1.70 GiB1.20 GiB3.71 GiB0.01 GiB21±30%
DeepSeek-OCRMoEQ4_K3.3B1.92 GiB1.00 GiB3.70 GiB0.02 GiB26±37%
Hunyuan-1.8B-InstructIQ3_XS1.8B0.78 GiB2.13 GiB3.70 GiB0.02 GiB21±30%
G9v3-3BQ2_K3.0B1.18 GiB1.73 GiB3.70 GiB0.02 GiB21±30%
FrickFritz-4BI1-Q2_K4.7B1.82 GiB1.06 GiB3.70 GiB0.02 GiB21±30%
qwen3.5-4b-agentic-coder-v4I1-Q2_K4.7B1.82 GiB1.06 GiB3.70 GiB0.02 GiB21±30%
Newton-bot-3-VLM-mini-4BQ2_K4.7B1.82 GiB1.06 GiB3.70 GiB0.02 GiB21±30%
Myth-4BI1-Q2_K4.3B1.82 GiB1.06 GiB3.70 GiB0.02 GiB21±30%
Qwen3.5-4B-UncensoredI1-Q2_K4.7B1.82 GiB1.06 GiB3.70 GiB0.02 GiB21±30%
JOSIE-2-4B-PreviewI1-Q2_K4.7B1.82 GiB1.06 GiB3.70 GiB0.02 GiB21±30%
Surogate-3.5-4BI1-Q2_K5.3B1.82 GiB1.06 GiB3.70 GiB0.02 GiB21±30%
Qwopus3.5-4B-v3Q2_K4.7B1.82 GiB1.06 GiB3.70 GiB0.02 GiB21±30%
umt5-xxlQ3_K_M5.7B2.85 GiB0.00 GiB3.70 GiB0.02 GiB21±30%
Qwen3.5-4BIQ2_M4.7B1.82 GiB1.06 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%
whisper-mediumF32764M2.85 GiB0.00 GiB3.69 GiB0.03 GiB21±30%
whisper-medium.enF32764M2.85 GiB0.00 GiB3.69 GiB0.03 GiB21±30%
InternVL3_5-8BIQ2_M8.5B2.84 GiB0.00 GiB3.69 GiB0.03 GiB21±30%
deepseek-coder-5.7bmqa-baseQ3_K_S5.7B2.33 GiB0.53 GiB3.69 GiB0.03 GiB21±30%
alduin-4b-it-baseI1-IQ4_XS4.3B2.12 GiB0.75 GiB3.69 GiB0.03 GiB21±30%
Qwen2.5-3B-Instruct-abliteratedI1-IQ1_M3.1B1.68 GiB1.20 GiB3.69 GiB0.03 GiB21±30%
EXAONE-4.0-1.2B-abliteratedI1-Q4_11.5B0.91 GiB1.99 GiB3.69 GiB0.03 GiB21±30%
HunyuanOCRMoEF321.1B2.01 GiB0.90 GiB3.68 GiB0.04 GiB12±37%
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-3.1-1b-a400m-instructMoEQ8_01.3B1.32 GiB1.59 GiB3.68 GiB0.04 GiB17±37%
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%
Vikhr-Gemma-2B-instructIQ2_M2.6B1.01 GiB1.85 GiB3.67 GiB0.05 GiB21±30%
Gemmasutra-Mini-2B-v1I1-IQ2_M2.6B1.01 GiB1.85 GiB3.67 GiB0.05 GiB21±30%
Holo-3.1-4BI1-IQ2_S5.2B1.79 GiB1.06 GiB3.67 GiB0.05 GiB21±30%
AfriqueQwen3.5-4BI1-IQ2_S5.2B1.79 GiB1.06 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 A350M 4GB run?
359 of 2118 indexed open-weight models fit a Arc A350M 4GB at 65,536 context with q8_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.
Arc A350M 4GB — what AI models can it run locally? — ossmodeldb