Nvidia, in collaboration with Foxconn, will build a new class of data centers, widely referred to as ‘AI factories'. The aim of this new venture is to provide advanced supercomputing capabilities necessary for the acceleration of autonomous vehicles, industrial robotics, and self-driving car development.
Jensen Huang, Nvidia CEO, and Foxconn CEO, Young Liu, declared this partnership on a Foxconn event, Hon Hai Tech Day, held in Taiwan recently. Essential to note is that the AI factory will leverage Nvidia's GPU computing infrastructure specifically built to process and transform huge chunks of data into valuable AI models.
The AI Factory: An End-to-End System
Standing front and center stage during the event, Huang detailed that the AI factory ties in perfectly with their intentions to build an advanced EV car with an embedded AI brain. Indicating that the car will continuously compile data throughout its lifecycle, and this collected data will be reverted to the AI factory and be used to enhance its software continually.
In January, Nvidia and Foxconn announced their partnership aimed at advancing developments in autonomous vehicle platforms. That former agreement championed Foxconn to be a significant supplier of Electronic Control Units (ECUs) for autombile makers. The ECUs will utilize Nvidia's Drive Orin system-on-a-chip (SoC), a high-level AI platform that supports the functionality of autonomous driving.
The Intersecting Paths of Nvidia's AI Factory and Tesla's Dojo Supercomputer
It is worth noting that these AI factories pose a significant rivalry to Tesla's Dojo supercomputer. Unlike Tesla, who utilize Nvidia GPU-based supercomputers, the new Foxconn-Nvidia AI factories will be based on Nvidia's GH200 Grace Hopper Superchip and AI Enterprise software.
In a quest to transition “from a manufacturing service company to a platform solutions company,” Foxconn plans to scale out the AI factories across various sectors. Foxconn is initially setting its focus on three primary platforms: Smart EVs, smart cities, and smart manufacturing. Given the final remarks from Huang and Liu, it appears every sector will have an AI factory in the future.
NVIDIA TensorRT Comes to Personal Windows PCs
In other Nvidia AI news today, the company revealed its TensorRT-LLM is now available for personal Windows. NVIDIA says that Windows computers with NVIDIA GeForce RTX GPUs can run LLMs up to four times faster. This will make LLM applications, such as writing and coding assistants, more responsive and effective. They can generate multiple unique auto-complete options at once, giving users more choices and efficiency.
NVIDIA showed an example of how TensorRT-LLM can improve LLM performance. A standard Meta LLaMa 2 LLM could not answer the question, “How does NVIDIA ACE generate emotional responses?“. But with TensorRT-LLM and a vector library or database, the LLM gave a correct answer faster.