Superchips For Accelerated Generative AI From NVIDIA Have Reached Full Production

The NVIDIA GH200 Grace Hopper Superchip, which will power systems going online globally to execute challenging AI and HPC workloads, was just revealed by NVIDIA as being in full production.

The GH200-powered systems join the more than 400 system configurations that use various NVIDIA CPU, GPU, and DPU architectures, such as NVIDIA GraceTM, NVIDIA HopperTM, NVIDIA Ada LovelaceTM, and NVIDIA BlueField®, to help fulfil the growing need for generative AI.

NVIDIA founder and CEO Jensen Huang unveiled new systems, partners, and additional information on the GH200 Grace Hopper Superchip at COMPUTEX. This chip uses NVIDIA NVLink®-C2C interconnect technology to combine the Arm-based NVIDIA Grace CPU and Hopper GPU architectures. This provides amazing computational capabilities to handle the most demanding generative AI and HPC workloads. It can give up to 900 GB/s total bandwidth, which is 7x higher bandwidth than the regular PCIe Gen5 lanes used in conventionally accelerated systems.

Ian Buck, vice president of accelerated computing at NVIDIA, stated that “generative AI is rapidly transforming businesses, unlocking new opportunities, and accelerating discovery in healthcare, finance, business services, and many more industries.” “Manufacturers around the world will soon offer the accelerated infrastructure enterprises need to build and deploy generative AI applications that leverage their specific proprietary data,” states the statement. “Grace Hopper Superchips are in full production.”

Customers who will have access to GH200-powered systems include global hyperscalers and supercomputing facilities in Europe and the United States.

Numerous cloud instances and accelerated systems
Several system manufacturers, including those from Taiwan, offer a wide range of systems on the market that are powered by various NVIDIA accelerators and CPUs. The following were mentioned in Huang’s keynote speech at COMPUTEX today as important partners: AAEON, Advantech, Aetina, ASRock Rack, ASUS, GIGABYTE, Ingrasys, Inventec, Pegatron, QCT, Tyan, Wistron, and Wiwynn.

Additionally, a wide range of NVIDIA-accelerated systems are offered by major server manufacturers such as Cisco, Dell Technologies, Hewlett Packard Enterprise, Lenovo, Supermicro, and Eviden, an Atos subsidiary.

Amazon Web Services (AWS), Cirrascale, CoreWeave, Google Cloud, Lambda, Microsoft Azure, Oracle Cloud Infrastructure, Paperspace, and Vultr are some of the cloud providers that work with the NVIDIA H100.

On Google Cloud, NVIDIA L4 GPUs are typically available.

Accelerated Systems for Full-Stack Computing
The NVIDIA software stack, which comprises NVIDIA AI, the NVIDIA OmniverseTM platform, and NVIDIA RTXTM technology, is supported widely by the next portfolio of systems accelerated by the NVIDIA Grace, Hopper, and Ada Lovelace architectures.

With over 100 frameworks, pretrained models, and development tools available, NVIDIA AI Enterprise, the software component of the NVIDIA AI platform, makes it easier to develop and deploy production AI, such as generative AI, computer vision, and speech AI.

Individuals and teams can work across several software suites and interact in real time in a shared environment thanks to the NVIDIA Omniverse development platform for creating and running metaverse apps. The Universal Scene Description framework, an open, extensible 3D language for virtual worlds, serves as the platform’s foundation.

With support for market-leading tools and APIs, the NVIDIA RTX platform blends ray tracing, deep learning, and rasterization to radically revolutionise the creative process for content producers and developers. Applications built on the RTX platform give millions of designers and artists the power of real-time photorealistic rendering and AI-enhanced graphics, video, and image processing to produce their finest work.