Nvidia Launches New Processor to Strengthen AI Leadership
Nvidia Corp. has unveiled an enhanced AI processor, aiming to solidify its leading position in a rapidly growing market by significantly boosting the chip’s capacity and speed.
The Grace Hopper Superchip, a combination graphics chip and processor, gets a boost from a new type of memory, Nvidia said at the Siggraph conference in Los Angeles. The product is based on High Bandwidth Memory 3 or HBM3e, which can process data at 5 terabytes per second.
The superchip, known as the GH200, will enter production in the second quarter of 2024, Nvidia said. It’s part of a new lineup of hardware and software to be announced at the event, a computer graphics expo where CEO Jensen Huang will speak.
Nvidia has built an early lead in the market for so-called AI accelerators, chips that excel at crunching data in the AI software development process. That has helped propel the company’s value to more than $1 trillion this year, making it the world’s most valuable chip maker. The latest processors show that Nvidia is working to make it harder for rivals such as Advanced Micro Devices Inc. and Intel Corp. to catch up.
Huang has turned the 30-year-old graphics chip business into a top-selling vendor of AI model training equipment. The process involves sifting through huge amounts of data. Now that AI tools like ChatGPT and Google Bard are reaching consumers and businesses alike, companies are racing to add Nvidia technology to handle workloads.
The superchip serves as the heart of a new server computer design that can process larger amounts of data and access it faster—a key advantage given the tsunami of information flowing through AI models. Artificial intelligence training gains momentum if the chip is able to load the model at once and update it without having to load parts into slower memory formats. This saves power and speeds up the whole process.
In servers, two chips can be deployed together, providing more than 3.5 times the capacity of the current model, Nvidia said. In this way, customers can use fewer machines or do the work much faster.
