March 20, 2024: Nvidia, the US chipmaker with a market worth of over $2 trillion, has prolonged its lead within the synthetic intelligence sector with the announcement of a number of groundbreaking applied sciences at its annual developer convention on Monday.
The corporate unveiled the “Blackwell” sequence of AI chips, designed to energy the costly datacentres that practice cutting-edge AI fashions similar to the most recent generations of GPT, Claude, and Gemini.
The Blackwell B200 chip is a big improve over Nvidia’s earlier H100 AI chip, decreasing the variety of chips and energy required to coach an enormous AI mannequin like GPT-4 by 75%.
The corporate additionally launched the GB200 “superchip,” which mixes two B200 chips and Nvidia’s Grace CPU on a single board, providing “30x the efficiency” for server farms that run chatbots like Claude or ChatGPT, whereas decreasing vitality consumption by as much as 25 occasions.
Nvidia CEO Jensen Huang showcased even bigger configurations, such because the GB200 NVL72, a single server rack with 72 B200 chips related by almost two miles of cabling, and the DGX Superpod, which mixes eight of those racks right into a shipping-container-sized AI datacentre in a field.
Along with the brand new AI chips, Nvidia introduced Undertaking GR00T, a basis mannequin for controlling humanoid robots.
This mannequin, paired with the Jetson Thor system-on-a-chip, goals to create autonomous machines that may be instructed utilizing pure language to hold out basic duties, even these they haven’t been particularly skilled for.
“Undertaking GR00T will assist robots perceive pure language and emulate actions by observing human actions – rapidly studying coordination, dexterity, and different expertise with the intention to navigate, adapt, and work together with the true world,” Huang defined.
Nvidia additionally entered the quantum computing sector with the introduction of a cloud service that makes use of its AI chips to simulate quantum computer systems.
Whereas the service is not going to initially be related to an actual quantum pc, it’s going to enable researchers to check their concepts with out the expense of accessing uncommon and dear quantum {hardware}.
Sooner or later, Nvidia plans to supply entry to third-party quantum computer systems by way of the platform.
As Nvidia continues to push the boundaries of synthetic intelligence and robotics, the corporate’s newest improvements showcase its dedication to main the business and shaping the way forward for these transformative applied sciences.
