Nvidia CEO Announces New H100 Datacenter Chips, AI Platform Updates, Grace CPU Superchip And More At 2022 GTC

Nvidia Corporation's NVDA CEO Jensen Huang announced a slew of new product and initiatives at the keynote address of the GTC 2022 virtual conference, which goes with the tagline "Hopped Up."

The event will run through Thursday, March 24 and feature a slew of speakers drawn from academia and research.

Here are the announcements from Huang's keynote address that you need to know. 

H100 Data Center GPU Based On Hopper Architecture: Nvidia announced its next-gen accelerated computing platform with Hopper architecture, which would replace the Ampere architecture launched two years ago.

The Hopper architecture is equipped with Transformer Engine, which can significantly speed AI performance and capabilities and help train large models within days or hours.

The company also unveiled its first Hopper-based graphic processing unit (GPU) named Nvidia H100. The H100 GPU is built with 80 billion transistors using Taiwan Semiconductor Manufacturing Company Limited's TSM 4nm processor node technology. It boasts of about 5 terabytes per second of external connectivity and is the first GPU to support PCle Gen5 and use HBM3, enabling 3TB/s of memory.

"Twenty H100 GPUs can sustain the equivalent of the entire world's internet traffic, making it possible for customers to deliver advanced recommender systems and large language models running inference on data in real time," Huang said.

The new GPU can be deployed in every data center type, including on-premise, cloud, hybrid-cloud and edge. It is expected to be available worldwide starting in the third quarter, directly from Nvidia as well as from cloud service providers and computer OEMs.

The company's listing of its H100 GPU clients included all leading cloud service providers and system manufacturers.

Related Link: Nvidia Analyst Raises Price Target By 78%: 'Chipmaker Has Largest TAM Expansion Opportunity In All Of Tech'

New DGX System: The chipmaker unveiled the fourth-generation Nvidia DGX system built with new H100 tensor core GPUs.

"Packing eight NVIDIA H100 GPUs per system, connected as one by NVIDIA NVLink, each DGX H100 provides 32 petaflops of AI performance at new FP8 precision — 6x more than the prior generation," the company said.

Nvidia also unveiled the Eos, the world's fastest AI supercomputer.

AI Platform Updates: Nvidia announced major updates to its AI platform, a suite of software for advancing workloads such as speech, recommender system, hyperscale inference and more.

Nvidia AI will include Riva for speech AI and Merlin for smart recommendations. The company also noted that updates have been made across the software suite, including tools such as Triton, NeMo, Maxine and TAO Toolkit.

Grace CPU Superchip: Nvidia announced its first Arm Neoverse-based discrete data center central processing unit (CPU), called Grace, designed for AI infrastructure and high-performance computing. The chip comprises two CPU chips connected coherently over NVLink-C2C, a high-speed, low-latency, chip-to-chip interconnect.

Nvidia Drive Hyperion 9: A next-gen platform for software-defined autonomous vehicle fleets, the Nvidia Drive Hyperion 9, was announced at the event.

Nvidia OVX: The company announced Nvidia OVX, which is purpose-built to operate complex digital twin simulations that will run within Nvidia Omniverse, a real-time physically accurate world simulation and 3D design collaboration platform.

At last check, Nvidia shares were down 0.64% at $265.63.

Photo courtesy of Nvidia.