Big AI models no longer need supercomputers. Discover how everyday laptops and mini-PCs can run and fine-tune them faster, cheaper, and with less power.
AI workloads are growing faster than ever but most notebooks, desktops and mini-PCs struggle with the memory and compute demands of large models. Phison Electronics’ upgraded aiDAPTIV+ tackles this problem by moving memory heavy operations from expensive DRAM to cost effective NAND flash. This reduces bottlenecks, lowers power use and makes inference and fine tuning of large AI models feasible on everyday devices.
For tasks like Mixture of Experts (MoE) inference, aiDAPTIV+ allows models with 120 B parameters to run on just 32 GB of DRAM compared with 96 GB needed using traditional methods. It also stores tokens that do not fit in the key value cache so they can be reused for future prompts. In tests, this approach speeds up inference by up to 10 times while reducing energy consumption and improves Time to First Token on portable devices.
This opens advanced AI to a wider audience. Students, developers and smaller organizations can now run or fine tune large LLMs directly on notebooks without relying on expensive workstations or cloud servers. For OEMs, resellers and system integrators, aiDAPTIV+ provides a practical end to end solution that overcomes GPU VRAM limits, supports scalable AI workloads and maintains data privacy all on familiar lower cost platforms.
By combining aiDAPTIV+ with modern processors and GPUs such as Intel Core Ultra Series 3 with Intel Arc graphics, even large models like 70B parameter LLMs can be trained on notebooks. This enables high performance AI where it was previously impossible, bringing desktop class AI capabilities to portable devices.
“As AI models grow into tens and hundreds of billions of parameters, the industry keeps hitting the same wall with GPU memory limitations,” said Michael Wu, President and GM, Phison US. “By expanding GPU memory with high-capacity, flash-based architecture in aiDAPTIV+, we offer everyone, from consumers and SMBs to large enterprises, the ability to train and run large-scale models on affordable hardware. In effect, we are turning everyday devices into supercomputers.”
“Our engineering collaboration enables Phison’s aiDAPTIV+ technology to accommodate and accelerate large models such as gpt-oss-120b on an Acer laptop with just 32GB of memory,” said Mark Yang, AVP, Compute Software Technology at Acer. “This can significantly enhance the user experience interacting with on-device Agentic AI, for actions ranging from simple search to intelligent inquiries that support productivity and creativity.”
