Intel · workstation

Arc Pro A40 6GB

Arc Pro A40 6GB has 6 GB of VRAM at 192 GB/s — about 5.58 GiB usable after driver and compositor overhead. 1250 of 2118 indexed models fit at 4K context with q4_0 KV.

Spec sheet· bandwidth, theoreticalFrom the file· fit from summed bytesPredicted· speed
Memory
6 GB
GDDR6
Bandwidth
192 GB/s
96-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 1070vision language 93audio asr 38audio tts 19embedding 26image 1video 3

What fits at 4K context

largest quantization that fits, per model · 1250 of 2118 indexed
ModelBest quantParamsWeights●KV●Total in memory◐Headroom◐tok/s◐
Mistral-NeMo-Minitron-8B-InstructQ4_K_S8.4B4.57 GiB0.18 GiB5.58 GiB0.00 GiB22±30%
dolphin-2.9.2-Phi-3-MediumKV unresolvedQ2_K_S14.0B4.50 GiB0.22 GiB5.58 GiB0.00 GiB22±30%
Darwin-36B-OpusMoEIQ2_M34.7B4.75 GiB0.02 GiB5.58 GiB0.00 GiB107±37%
gemma-4-E4B-itQ4_18.0B4.73 GiB0.03 GiB5.58 GiB0.00 GiB22±30%
gemma-4-E4B-itQ4_18.0B4.73 GiB0.03 GiB5.58 GiB0.00 GiB22±30%
Gemma-4-E4B-it-Minecraft-MT-en-zh-v0.1I1-IQ4_XS8.0B4.72 GiB0.03 GiB5.57 GiB0.01 GiB22±30%
gemma-4-E4B-Queen-it-qat-q4_0-unquantizedI1-IQ4_XS8.0B4.72 GiB0.03 GiB5.57 GiB0.01 GiB22±30%
Gemma-4-E4B-Luchador-RudoI1-IQ4_XS8.0B4.72 GiB0.03 GiB5.57 GiB0.01 GiB22±30%
supergemma4-e4b-abliteratedI1-IQ4_XS7.5B4.72 GiB0.03 GiB5.57 GiB0.01 GiB22±30%
Gemma-4-E4B-AbliteratedI1-IQ4_XS8.0B4.72 GiB0.03 GiB5.57 GiB0.01 GiB22±30%
gemma-4-E4B-it-The-DECKARD-Claude-Opus-Expresso-Universe-HERETIC-UNCENSORED-ThinkingI1-IQ4_XS8.0B4.72 GiB0.03 GiB5.57 GiB0.01 GiB22±30%
gemma-4-E4B-it-The-DECKARD-Expresso-Universe-HERETIC-UNCENSORED-ThinkingI1-IQ4_XS8.0B4.72 GiB0.03 GiB5.57 GiB0.01 GiB22±30%
gemma-4-E4B-it-hereticI1-IQ4_XS8.0B4.72 GiB0.03 GiB5.57 GiB0.01 GiB22±30%
gemma-4-E4B-it-Claude-Opus-4.5-HERETIC-UNCENSORED-ThinkingI1-IQ4_XS8.0B4.72 GiB0.03 GiB5.57 GiB0.01 GiB22±30%
Huihui-gemma-4-E4B-it-abliteratedI1-IQ4_XS8.0B4.72 GiB0.03 GiB5.57 GiB0.01 GiB22±30%
gemma-4-E4B-it-Uncensored-MAXI1-IQ4_XS8.0B4.72 GiB0.03 GiB5.57 GiB0.01 GiB22±30%
Darkidol-Gemma-4-E4B-itI1-IQ4_XS8.0B4.72 GiB0.03 GiB5.57 GiB0.01 GiB22±30%
gemma-4-E4B-it-abliteratedI1-IQ4_XS8.0B4.72 GiB0.03 GiB5.57 GiB0.01 GiB22±30%
OpenMedResearch-Gemma-4E4NI1-IQ4_XS8.0B4.72 GiB0.03 GiB5.57 GiB0.01 GiB22±30%
Reasoning-Medical0.1-E4B-sftI1-IQ4_XS8.0B4.72 GiB0.03 GiB5.57 GiB0.01 GiB22±30%
Qwen3.5-9B-CoderI1-Q3_K_L9.7B4.70 GiB0.04 GiB5.57 GiB0.01 GiB22±30%
Qwythos-9B-Claude-Mythos-5-1M-MTPI1-Q3_K_L9.7B4.70 GiB0.04 GiB5.57 GiB0.01 GiB22±30%
Huihui-Qwythos-9B-Claude-Mythos-5-1M-abliteratedI1-Q3_K_L9.7B4.70 GiB0.04 GiB5.57 GiB0.01 GiB22±30%
Qwen3.5-9B-Fable-5-v1I1-Q3_K_L9.7B4.70 GiB0.04 GiB5.57 GiB0.01 GiB22±30%
Qwythos-9B-v2I1-Q3_K_L9.7B4.70 GiB0.04 GiB5.57 GiB0.01 GiB22±30%
PINQWEN-3.5-9B-1M-BF16I1-Q3_K_L9.7B4.70 GiB0.04 GiB5.57 GiB0.01 GiB22±30%
Openprose-2-FlashI1-Q3_K_L9.7B4.70 GiB0.04 GiB5.57 GiB0.01 GiB22±30%
Qwen3.5-9B-Nikusui-v1I1-Q3_K_L9.7B4.70 GiB0.04 GiB5.57 GiB0.01 GiB22±30%
Ornstein-3.5-9B-V1.5I1-Q3_K_L9.7B4.70 GiB0.04 GiB5.57 GiB0.01 GiB22±30%
Ornith-1.0-9B-heretic-MTPI1-Q3_K_L9.4B4.70 GiB0.04 GiB5.57 GiB0.01 GiB22±30%
Tess-4-9BI1-Q3_K_L9.7B4.70 GiB0.04 GiB5.57 GiB0.01 GiB22±30%
dotwebs-1I1-Q3_K_L9.7B4.70 GiB0.04 GiB5.57 GiB0.01 GiB22±30%
liftQ3_K_L9.7B4.70 GiB0.04 GiB5.57 GiB0.01 GiB22±30%
Hemlock-Qwopus3.5-9B-CoderI1-Q3_K_L9.7B4.70 GiB0.04 GiB5.57 GiB0.01 GiB22±30%
Qwen3.5-9BQ3_K_L9.7B4.70 GiB0.04 GiB5.57 GiB0.01 GiB22±30%
canary-qwen-2.5bBF162.6B4.73 GiB0.00 GiB5.57 GiB0.01 GiB22±30%
EXAONE-Deep-7.8BQ4_K_L7.8B4.73 GiB0.00 GiB5.57 GiB0.01 GiB22±30%
EXAONE-3.5-7.8B-InstructQ4_K_L7.8B4.73 GiB0.00 GiB5.57 GiB0.01 GiB22±30%
Qwen3-TTS-12Hz-0.6B-BaseQ4_K_M915M4.72 GiB0.00 GiB5.57 GiB0.01 GiB22±30%
VoxCPM2F162.3B4.72 GiB0.00 GiB5.57 GiB0.01 GiB22±30%
Qwen3-16B-A3BMoEIQ2_S16.0B4.67 GiB0.11 GiB5.57 GiB0.01 GiB58±37%
Assistant_Pepe_8BQ4_K_M—4.59 GiB0.14 GiB5.57 GiB0.01 GiB22±30%
Ministral-8B-Instruct-2410Q4_K_M8.0B4.57 GiB0.16 GiB5.56 GiB0.02 GiB22±30%
SuperGemma-4-12b-abliteratedI1-IQ3_XXS12.0B4.52 GiB0.20 GiB5.56 GiB0.02 GiB22±30%
gemma-4-12B-agentic-fable5-composer2.5-v2-3.5x-tau2-uncensored-hereticI1-IQ3_XXS12.0B4.52 GiB0.20 GiB5.56 GiB0.02 GiB22±30%
gemma-4-12B-coder-fable5-composer2.5-v1-uncensored-hereticI1-IQ3_XXS12.0B4.52 GiB0.20 GiB5.56 GiB0.02 GiB22±30%
gemma-4-12B-it-uncensored-hereticI1-IQ3_XXS12.0B4.52 GiB0.20 GiB5.56 GiB0.02 GiB22±30%
Grug-12BI1-IQ3_XXS12.0B4.52 GiB0.20 GiB5.56 GiB0.02 GiB22±30%
Aura-Medium-v1-BF16I1-IQ3_XXS12.0B4.52 GiB0.20 GiB5.56 GiB0.02 GiB22±30%
gemma-4-12B-it-Esper4I1-IQ3_XXS12.0B4.52 GiB0.20 GiB5.56 GiB0.02 GiB22±30%
gemma-4-12B-it-GuardpointI1-IQ3_XXS12.0B4.52 GiB0.20 GiB5.56 GiB0.02 GiB22±30%
Gemma-4-12B-it-AEON-Abliterated-K4-BF16I1-IQ3_XXS12.0B4.52 GiB0.20 GiB5.56 GiB0.02 GiB22±30%
gemma-4-12B-it-Tachibana-AgentI1-IQ3_XXS12.0B4.52 GiB0.20 GiB5.56 GiB0.02 GiB22±30%
gemma-4-12b-marvin-gutenberg-rp-v2I1-IQ3_XXS12.0B4.52 GiB0.20 GiB5.56 GiB0.02 GiB22±30%
gemma-4-12b-crownelius-writerI1-IQ3_XXS12.0B4.52 GiB0.20 GiB5.56 GiB0.02 GiB22±30%
Huihui-gemma-4-12B-coder-fable5-composer2.5-v1-abliteratedI1-IQ3_XXS12.0B4.52 GiB0.20 GiB5.56 GiB0.02 GiB22±30%
gemma-4-12b-asterion-agenticI1-IQ3_XXS12.0B4.52 GiB0.20 GiB5.56 GiB0.02 GiB22±30%
Huihui-gemma-4-12B-agentic-fable5-abliteratedI1-IQ3_XXS12.0B4.52 GiB0.20 GiB5.56 GiB0.02 GiB22±30%
g4-12b-it-trismegistusI1-IQ3_XXS12.0B4.52 GiB0.20 GiB5.56 GiB0.02 GiB22±30%
gemma4-12b-it-asimovI1-IQ3_XXS12.0B4.52 GiB0.20 GiB5.56 GiB0.02 GiB22±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 Pro A40 6GB run?
1250 of 2118 indexed open-weight models fit a Arc Pro A40 6GB at 4,096 context with q4_0 KV cache, the largest being Mistral-NeMo-Minitron-8B-Instruct at Q4_K_S. That covers text, vision-language, image, video and speech models.
How much usable memory does a Arc Pro A40 6GB actually have?
Its nameplate is 6 GB, but about 5.58 GiB is available to a model once driver and compositor overhead is accounted for.
Is a Arc Pro A40 6GB fast for local AI?
Its memory bandwidth is 192 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.