Nvidia reportedly planning Arm-Based Chips
Nvidia reportedly planning Arm-Based Chips

NVIDIA unveils DGX GH200: A Supercomputer revolutionizing Generative AI

What To Know

  • One of the highlights was the unveiling of the company’s forthcoming supercomputer, the DGX GH200, which is intended to help businesses construct generative AI models.
  • NVIDIA claims that this integration allows the DGX GH200 to deliver a remarkable performance of 1 exaflop and possess a shared memory capacity of 144 terabytes.
  • To put things into perspective, the latest Top500 supercomputer rankings only list Frontier at Oak Ridge National Laboratory in Tennessee as an exascale system, achieving close to 1.
  • According to NVIDIA, the architecture of the DGX GH200 offers ten times the bandwidth of the previous generation, providing the power of an extensive AI supercomputer while maintaining the programming simplicity of a single GPU.

Jensen Huang, CEO of NVIDIA, made numerous big announcements during his keynote speech at Computex. One of the highlights was the unveiling of the company’s forthcoming supercomputer, the DGX GH200, which is intended to help businesses construct generative AI models. It is hardly surprising given the industry’s current trajectory.

The DGX GH200 supercomputer incorporates a novel NVLink Switch System, enabling 256 GH200 Grace Hopper Superchips to function as a unified GPU. Each chip comprises an Arm-based Grace CPU and an H100 Tensor Core GPU.

NVIDIA claims that this integration allows the DGX GH200 to deliver a remarkable performance of 1 exaflop and possess a shared memory capacity of 144 terabytes. In comparison, this is nearly 500 times the memory found in a single DGX A100 system.

To put things into perspective, the latest Top500 supercomputer rankings only list Frontier at Oak Ridge National Laboratory in Tennessee as an exascale system, achieving close to 1.2 exaflops on the Linmark benchmark. This performance surpasses the second-ranked system, Japan’s Fugaku, by over double.

Essentially, NVIDIA asserts that it has developed a supercomputer capable of rivaling the world’s most powerful known system. (Meta is also constructing a supercomputer that claims to be the fastest AI supercomputer upon completion.)

According to NVIDIA, the architecture of the DGX GH200 offers ten times the bandwidth of the previous generation, providing the power of an extensive AI supercomputer while maintaining the programming simplicity of a single GPU.

Numerous prominent companies, including Google Cloud, Meta, and Microsoft, have shown keen interest in the DGX GH200. They will be among the first to access the supercomputer to test its ability to handle generative AI workloads. The DGX GH200 supercomputers are expected to be ready by the end of 2023, according to NVIDIA.

Furthermore, NVIDIA is building its supercomputer, Helios, which will be made up of four DGX GH200 systems. Helios is expected to be operating by the end of this year.

During his keynote, Huang also discussed other advancements in generative AI, particularly in the gaming sector. NVIDIA Avatar Cloud Engine (ACE) for Games is a service that allows developers to construct personalized AI models for voice, dialogue, and animation.

ACE for Games gives non-playable characters conversational abilities, allowing them to answer to questions with increasingly lifelike personalities.

Comments

No comments yet. Why don’t you start the discussion?

Leave a Reply

Your email address will not be published. Required fields are marked *