NVIDIA’s Jensen Huang is at Computex 2023 in Taiwan this week, delivering a keynote filled with new product announcements with an emphasis on enabling the development of next-gen models for generative AI applications, data analytics, and recommender systems. The NVIDIA DGX supercomputer powered by NVIDIA GH200 Grace Hopper Superchips and the NVIDIA NVLink switch system took center stage.
NVIDIA’s Jensen Huang is at Computex 2023 in Taiwan this week, delivering a keynote filled with new product announcements with an emphasis on enabling the development of next-gen models for generative AI applications, data analytics, and recommender systems. The NVIDIA DGX supercomputer powered by NVIDIA GH200 Grace Hopper Superchips and the NVIDIA NVLink switch system took center stage.
The NVIDIA DGX GH200 uses NVLink interconnect technology with the NVLink Switch System to combine 256 GH200 Superchips to perform as a single GPU, providing 1 exaflop of performance and 144 terabytes of shared memory. That’s almost 500x more memory than a single NVIDIA DGX A100 system!
NVLink Technology Expands AI at Scale
The GH200 Superchips combine the Arm-based NVIDIA Grace CPU with the NVIDIA H100 Tensor Core GPU using NVLink-C2C chip interconnects, eliminating the need for traditional CPU-to-GPU PCIe connection. The bandwidth between GPU and CPU is increased by 7x compared to the latest PCIe technology, reducing interconnect power consumption by over 5x and providing a 600GB Hopper architecture GPU building block for DGX GH200 supercomputers.
This is the first supercomputer to pair Grace Hopper Superchips with the NVLink Switch System. This new interconnect enables all GPUs in a DGX GH200 system to work as one, compared to the eight GPU limit combined with NVLink as a single GPU without affecting performance. The DGX GH200 architecture delivers 10x more bandwidth than the previous generation, delivering the power of a massive AI supercomputer with the simplicity of programming a single GPU.
AI Pioneers Get New Research Tools
The first to access the new supercomputer is expected to be Google Cloud, Meta, and Microsoft, allowing them to explore its capabilities for generative AI workloads. NVIDIA intends to provide the DGX GH200 design as a blueprint to cloud service providers and other hyperscalers so they can customize it for their infrastructure.
NVIDIA researchers and development teams will get access to the new NVIDIA Helios supercomputer featuring four DGX GH200 systems. Helios will include 1,024 Grace Hopper Superchips and is expected to be online by the end of the year. Each system in the Helios supercomputer will be interconnected with NVIDIA Quantum-2 InfiniBand networking with up to 400Gb/s bandwidth data throughput for training large AI models.
Integrated and Purpose-Built
The DGX GH200 supercomputers will include NVIDIA software providing a turnkey, full-stack solution, supporting the largest AI and data analytics workloads. NVIDIA Base Command software provides AI workflow management, enterprise-grade cluster management, libraries to accelerate compute, storage, and network infrastructure, and system software optimized for running AI workloads. NVIDIA AI Enterprise software will be included providing more than 100 frameworks, retrained models, and development tools to streamline the development and deployment of production AI, including generative AI, computer vision, speech AI, and others.
Availability
NVIDIA DGX GH200 supercomputers are expected to be available by the end of the year.
Engage with StorageReview
Newsletter | YouTube | Podcast iTunes/Spotify | Instagram | Twitter | TikTok | RSS Feed
