Jensen has described the company’s new Blackwell AI processor as a “miracle” of computational power. He presented the company’s next-generation AI processor at the GTC 2024 conference on Monday, aiming to maintain the company’s leadership in the AI race.
The new Blackwell G200 is the successor to the Grace Hopper family of graphics processing units (GPUs), utilized in data centers and supercomputers to accelerate activities. Nvidia claims the new chip can support AI models with trillions of parameters while lowering operating costs and energy consumption.
According to a press statement, organizations such as Amazon Web Services, Dell, Google, Meta, Microsoft, OpenAI, Oracle, Tesla, and xAI are expected to use Blackwell.
“Hopper is fantastic, but we need bigger GPUs,” Huang said during his keynote talk, where he unveiled Blackwell as a new “very, very big GPU” that binds two silicon squares the size of their predecessor chip to work as a single unit.
Huang stated that Blackwell’s design enables it to simply replace its predecessor, the Grace Hopper chip. “You slip out Hopper and push in Blackwell. That explains why one of the ramping problems will be very efficient.
“Hopper installations can be found all over the world, and they may share the same infrastructure, design, power, electricity, thermals, and software. Huang noted that this is the Hopper version for the current SGX configuration.
Huang demonstrated a “quite expensive” fully functional board to the audience, which included two Blackwell chips and four Blackwell dyes connected to a Grace CPU with a “super fast chip-to-chip link.”
Read More: https://ciolook.com/
