Intel · consumer

Arc A310 4GB

Arc A310 4GB has 4 GB of VRAM at 124 GB/s — about 3.72 GiB usable after driver and compositor overhead. 285 of 2118 indexed models fit at 128K context with q8_0 KV.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed
Memory
4 GB
GDDR6
Bandwidth
124 GB/s
64-bit bus
Tensor FP16
—
dense
TDP
75 W
$110 MSRP
Context4K8K16K32K64K128K
KV cachef16q8_0q4_0quantizing the KV cache is a ~2× lever on the dominant term at long context
text 203vision language 23audio asr 29embedding 13video 2audio tts 15

What fits at 128K context

largest quantization that fits, per model · 285 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
LFM2-1.2BQ5_K_S1.2B0.79 GiB2.13 GiB3.72 GiB0.00 GiB23±30%
granite-4.0-7B-A1B-Creative-v0.1MoEI1-IQ3_XXS6.7B2.42 GiB0.53 GiB3.71 GiB0.01 GiB41±37%
SEX_ROLEPLAY-3.2-1BI1-Q3_K_L1.5B0.79 GiB2.13 GiB3.71 GiB0.01 GiB23±30%
Llama-3.2-1B-Instruct-abliteratedI1-Q3_K_L1.5B0.79 GiB2.13 GiB3.71 GiB0.01 GiB23±30%
Novaciano-3.2-1BI1-Q3_K_L1.5B0.79 GiB2.13 GiB3.71 GiB0.01 GiB23±30%
Imp-RPG.System-1BI1-Q3_K_L1.5B0.79 GiB2.13 GiB3.71 GiB0.01 GiB23±30%
gemma-4-E2B-itIQ2_M5.1B2.44 GiB0.48 GiB3.71 GiB0.01 GiB23±30%
gemma-3n-E2B-itIQ3_M5.4B2.08 GiB0.82 GiB3.71 GiB0.01 GiB23±30%
Dolphin3.0-Qwen2.5-1.5BQ5_K_M1.5B1.05 GiB1.86 GiB3.71 GiB0.01 GiB23±30%
Qwen2.5-1.5B-Instruct-abliteratedI1-Q5_K_M1.5B1.05 GiB1.86 GiB3.71 GiB0.01 GiB23±30%
Qwen2.5-1.5B-VibeThinker-heretic-uncensored-abliteratedI1-Q5_K_M1.5B1.05 GiB1.86 GiB3.71 GiB0.01 GiB23±30%
NEXUS-Coder-OBLITERATEDI1-Q5_K_M1.5B1.05 GiB1.86 GiB3.71 GiB0.01 GiB23±30%
NEXUS-Coder-AbliteratedI1-Q5_K_M1.5B1.05 GiB1.86 GiB3.71 GiB0.01 GiB23±30%
Qwen2.5-1.5B-hereticI1-Q5_K_M1.5B1.05 GiB1.86 GiB3.71 GiB0.01 GiB23±30%
Qwen2.5-Coder-1.5B-Unsensored-DPOI1-Q5_K_M1.5B1.05 GiB1.86 GiB3.71 GiB0.01 GiB23±30%
Qwen2.5-Math-1.5B-InstructQ5_K_M1.5B1.05 GiB1.86 GiB3.71 GiB0.01 GiB23±30%
ShellWhisperer-1.5BQ5_K_M1.5B1.05 GiB1.86 GiB3.71 GiB0.01 GiB23±30%
Qwen2.5-1.5BQ5_K_M1.5B1.05 GiB1.86 GiB3.71 GiB0.01 GiB23±30%
PiCo-1BI1-Q5_K_M1.5B1.05 GiB1.86 GiB3.71 GiB0.01 GiB23±30%
Fourier-Qwen2-VL-2B-0.67I1-Q5_K_M2.2B1.05 GiB1.86 GiB3.71 GiB0.01 GiB23±30%
Qwen2-VL-2B-InstructQ5_K_M2.2B1.05 GiB1.86 GiB3.71 GiB0.01 GiB23±30%
FableForge-1.5BI1-Q5_K_M1.5B1.05 GiB1.86 GiB3.71 GiB0.01 GiB23±30%
Qwen2-1.5B-InstructQ5_K_M1.5B1.05 GiB1.86 GiB3.71 GiB0.01 GiB23±30%
NEXUS-MedicalQ5_K_M1.5B1.05 GiB1.86 GiB3.71 GiB0.01 GiB23±30%
NEXUS-FinanceQ5_K_M1.5B1.05 GiB1.86 GiB3.71 GiB0.01 GiB23±30%
NEXUS-SecurityQ5_K_M1.5B1.05 GiB1.86 GiB3.71 GiB0.01 GiB23±30%
Qwen2-1.5BQ5_K1.5B1.05 GiB1.86 GiB3.71 GiB0.01 GiB23±30%
Qwen2.5-Coder-1.5BQ5_K_M1.5B1.05 GiB1.86 GiB3.71 GiB0.01 GiB23±30%
NEXUS-ScienceQ5_K_M1.5B1.05 GiB1.86 GiB3.71 GiB0.01 GiB23±30%
NEXUS-LegalQ5_K_M1.5B1.05 GiB1.86 GiB3.71 GiB0.01 GiB23±30%
NEXUS-CoderQ5_K_M1.5B1.05 GiB1.86 GiB3.71 GiB0.01 GiB23±30%
SmolVLM2-500M-Video-InstructQ4_K_M507M0.28 GiB2.66 GiB3.71 GiB0.01 GiB23±30%
TinyLlama-1.1B-Chat-v1.0Q5_K_M1.1B1.46 GiB1.46 GiB3.71 GiB0.01 GiB23±30%
gemma-2bQ5_02.5B1.68 GiB1.20 GiB3.71 GiB0.01 GiB23±30%
granite-4.0-h-tiny-baseMoEQ2_K6.9B2.41 GiB0.53 GiB3.71 GiB0.01 GiB41±37%
Gemma-3-4b-it-Uncensored-DBL-XI1-IQ1_M4.7B1.20 GiB1.69 GiB3.71 GiB0.01 GiB23±30%
Miril-Drone-2B-1IQ2_M5.1B2.43 GiB0.48 GiB3.70 GiB0.02 GiB23±30%
DeepScaleR-1.5B-PreviewQ4_K_M1.8B1.04 GiB1.86 GiB3.70 GiB0.02 GiB23±30%
Qwen2.5-Coder-1.5B-Instruct-abliteratedQ4_K_M1.8B1.04 GiB1.86 GiB3.70 GiB0.02 GiB23±30%
Qwen2.5-1.5B-Instruct-uncensoredQ4_K_M1.8B1.04 GiB1.86 GiB3.70 GiB0.02 GiB23±30%
VibeThinker-1.5BQ4_K_M1.8B1.04 GiB1.86 GiB3.70 GiB0.02 GiB23±30%
DeepSeek-R1-Distill-Qwen-1.5B-uncensoredI1-Q4_K_M1.8B1.04 GiB1.86 GiB3.70 GiB0.02 GiB23±30%
DeepSeek-R1-Distill-Qwen-1.5BQ4_K_M1.8B1.04 GiB1.86 GiB3.70 GiB0.02 GiB23±30%
Nemotron-Research-Reasoning-Qwen-1.5BQ4_K_M1.8B1.04 GiB1.86 GiB3.70 GiB0.02 GiB23±30%
Qwen2.5-Coder-1.5B-InstructQ4_K_M1.5B1.04 GiB1.86 GiB3.70 GiB0.02 GiB23±30%
Qwen2.5-1.5B-InstructQ4_K_M1.5B1.04 GiB1.86 GiB3.70 GiB0.02 GiB23±30%
deepseek-r1-distill-qwen-1.5b-unsloth-bnb-4bitQ4_K_M1.8B1.04 GiB1.86 GiB3.70 GiB0.02 GiB23±30%
dots.ocrI1-Q4_K_M3.0B1.04 GiB1.86 GiB3.70 GiB0.02 GiB23±30%
Dolphin3.0-Llama3.2-1BQ4_11.2B0.77 GiB2.13 GiB3.70 GiB0.02 GiB23±30%
Llama-3.2-1B-Instruct-hereticI1-Q4_11.2B0.77 GiB2.13 GiB3.70 GiB0.02 GiB23±30%
Llama-3.2-1B-InstructQ4_11.2B0.77 GiB2.13 GiB3.70 GiB0.02 GiB23±30%
DeepSeek-R1-Distill-Qwen-1.5B-Fully-UncensoredQ4_K_M1.8B1.04 GiB1.86 GiB3.70 GiB0.02 GiB23±30%
Trinity-Nano-PreviewMoEIQ2_M6.1B1.94 GiB0.98 GiB3.70 GiB0.02 GiB31±37%
umt5-xxlQ3_K_M5.7B2.85 GiB0.00 GiB3.70 GiB0.02 GiB23±30%
medgemma-1.5-4b-itUD-IQ2_M4.3B1.46 GiB1.42 GiB3.70 GiB0.02 GiB23±30%
medgemma-4b-itUD-IQ2_M4.3B1.46 GiB1.42 GiB3.70 GiB0.02 GiB23±30%
smollm-360M-instruct-add-basicsQ5_K362M0.27 GiB2.66 GiB3.69 GiB0.03 GiB23±30%
SmolLM2-360M-InstructQ5_K_L362M0.27 GiB2.66 GiB3.69 GiB0.03 GiB23±30%
whisper-mediumF32764M2.85 GiB0.00 GiB3.69 GiB0.03 GiB23±30%
whisper-medium.enF32764M2.85 GiB0.00 GiB3.69 GiB0.03 GiB23±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 A310 4GB run?
285 of 2118 indexed open-weight models fit a Arc A310 4GB at 131,072 context with q8_0 KV cache, the largest being LFM2-1.2B at Q5_K_S. That covers text, vision-language, image, video and speech models.
How much usable memory does a Arc A310 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 A310 4GB fast for local AI?
Its memory bandwidth is 124 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.