<\/p>\n
The NVIDIA GTC conference is all about AI <\/a>and Phison was there to introduce exciting new developments in its aiDAPTIV+ products and solutions. With more than 39,000 on-site attendees and another 300,000 experiencing the show online, the event was a resounding success for the AI industry and Phison alike. It truly was the Super Bowl of AI.<\/p>\n Read on for the highlights of Phison\u2019s showing at GTC.<\/p>\n <\/p>\n <\/p>\n In partnership with MAINGEAR<\/a>, Phison showed to conference attendees the first-ever LLM training laptop PC concept demo supporting models up to 8 billion-parameters. The MAINGEAR AI laptop includes Phison aiDAPTIVCache SSDs, aiDAPTIVLink middleware and an aiDAPTIV+ Pro Suite GUI-based, all-in-one feature set for LLM operations\u2014including ingest, fine-tuning, retrieval-augmented generation (RAG), monitoring, validation and inference.<\/p>\n The laptop is ideal for aspiring AI developers and data scientists, as well as any organization that wants to train AI models on its own data without having to rely on the cloud or multimillion-dollar on-premises infrastructure.<\/p>\n \u201caiDAPTIV+ is now equivalent to having an expert on your own data in your backpack at all times,\u201d said Michael Wu, GM and President at Phison US in the company\u2019s recent press release<\/a>. \u201cNot only do you get to train and do inferencing on your own fine-tuned or RAG-enabled LLMs, but then you reap the rewards of insights.\u201d<\/p>\n <\/p>\n <\/p>\n In the same press release, MAINGEAR CEO Wallace Santos said the proof-of-concept laptop \u201cenables both powerful edge AI inference and on-device model training from anywhere\u201d and gives \u201cbusinesses, educators, and students a cost-effective tool to develop AI models<\/a> without the complexity of traditional data center infrastructure.\u201d<\/p>\n Reviewer Sayan Sen, reporting for Neowin.net<\/a>, said the Phison-MAINGEAR AI laptop \u201chelps in dynamic caching by expanding on the available HBM (high-bandwidth memory)\/GDDR and combining it with NAND flash-based aiDAPTIVCache.\u201d The article<\/a> includes a chart that shows the MAINGEAR AI PC with aiDAPTIV+ far outperforming a MAINGEAR setup without aiDAPTIV+.<\/p>\n The MAINGEAR AI PC with aiDAPTIV+ technology will be available Q2 2025 in three preconfigured models. Customers can now register online<\/a> to get more details about availability and ordering.<\/p>\n <\/p>\n <\/p>\n Phison announced that aiDAPTIV+ technology<\/a> is now validated to run on NVIDIA Jetson-based devices. Using aiDAPTIVCache SSDs enhances inference and Low-Rank Adaptation (LoRA)-based LLM training capabilities. This opens up a whole new world of AI possibilities in edge computing and robotics use cases. It will also be beneficial in self-driving vehicles, industrial automation, healthcare diagnostics, retail customer analytics, telecommunications, agriculture, environmental monitoring, intelligent surveillance and much more.<\/p>\n <\/p>\n <\/p>\n <\/p>\n Phison has updated its aiDAPTIVLink middleware and set availability for April of this year. aiDAPTIVLink 3.0 delivers enhanced inference capabilities through faster recall response and more accuracy in answers to prompts. It does this by allowing longer token lengths, which enables much more complex prompts.<\/p>\n <\/p>\n An article by GuruFocus<\/a> illustrated this increase in performance:<\/p>\n \u201cGenerative AI adoption at edge devices is often limited by DRAM capacity constraints when running large language models (LLM). The [Adlink] DLAP Supreme series, featuring NVIDIA Jetson AGX Orin, elevates AI power with Phison aiDAPTIV+ technology, effectively overcoming token length limitations and enhancing inference speed. For instance, using the DLAP-411 Orin Supreme equipped with an NVIDIA Jetson AGX Orin 64GB configuration alongside the Llama3.1-70B model, the inference [time to first token] TTFT is improved by 52 times, and token length increased by 4 times.\u201d<\/em><\/p>\n<\/blockquote>\n <\/p>\n Phison\u2019s announcements were included in a GTC summary article from Forbes:<\/a><\/p>\n \u201cPhison announced an array of expanded capabilities on aiDAPTIV+, a more affordable AI training and inference solution for on-premise infrastructure. aiDAPTIV+ is being integrated into AI laptop PCs as well as on edge computing devices running the Nvidia Jetson platform.\u201d<\/em><\/p>\n<\/blockquote>\n <\/p>\n TweakTown<\/a>, on the MAINGEAR-Phison AI PC laptop:<\/p>\n \u201cPhison is setting out to reduce the cost of running AI models by offloading the processing, which is currently on high-cost GPU memory, to low-cost NAND flash storage, and by doing this the operational cost of the AI model shrinks, enabling larger models to be trained on cheaper devices. By utilizing the power of the company’s aiDAPTIV+ software, users will now be able to customize an AI model with their own data, all within a device that can fit in your backpack.\u201d<\/em><\/p>\n<\/blockquote>\n <\/p>\n TweakTown<\/a>, on how aiDAPTIV+ solutions drastically cut the cut of AI training:<\/p>\n \u201cPhison expects the world’s first 1 trillion parameter model will be unveiled before the end of 2025. \u2026 Phison expects the operational costs of a 1 trillion parameter model will be approximately $3 million in raw GPU power, but the company has outlined a plan to alleviate the cost down to just $100,000. How? A combination of SSDs, Phison’s aiDAPTIV+ software, and an NVIDIA GH200 Superchip.<\/em><\/p>\n <\/p>\n By offloading the processing from the GPUs onto SSDs through the use of Phison’s intuitive aiDAPTIV+ software, the company is able to significantly reduce the number of GPUs required to successfully run a large AI model. Phison anticipates it will be able to reduce the cost of training the largest-sized model by 1\/30th of the price, or less than 4% of the currently estimated operating cost.\u201d<\/em><\/p>\n<\/blockquote>\n <\/p>\n![\"\"](\"https:\/\/phisonblog.com\/wp-content\/uploads\/2025\/03\/Maingear-Phison-AI-Laptop-1024x640.png\")
<\/p>\nIntroducing the industry\u2019s first AI laptop PC capable of LLMOps<\/h3>\n
![\"\"](\"https:\/\/phisonblog.com\/wp-content\/uploads\/2024\/03\/964_2464965148.jpg\")
<\/a>![\"\"](\"https:\/\/phisonblog.com\/wp-content\/uploads\/2025\/03\/aidaptiv-cache-1024x836.png\")
<\/p>\nEnable AI inferencing at the edge with aiDAPTIV+<\/h3>\n
![\"\"](\"https:\/\/phisonblog.com\/wp-content\/uploads\/2025\/03\/aidaptiv-link-graph.png\")
<\/p>\nUpdated aiDAPTIVLink middleware enables greater speed and more complex prompts<\/h3>\n
\n
See what other media had to say<\/h3>\n
\n
\n
\n