NVIDIA’s 2024 GTC event runs until March 21, and will see the usual slew of announcements from major technology conferences. The most notable was founder and CEO Jensen Huang’s keynote: the next generation Blackwell GPU architecture, enabling organizations to build and run real-time generative artificial intelligence on large-scale language models with trillions of parameters.
“The future is generative…that’s why this is a completely new industry,” Huang told attendees. “Our approach to computing is fundamentally different. We have created a processor for the age of generative artificial intelligence.”
However, that wasn’t the only “next generation” announcement made at the San Jose rally.
NVIDIA announced a blueprint for building a next-generation data center, promising to create an “efficient artificial intelligence infrastructure” with the support of partners such as Schneider Electric, data center infrastructure company Vertiv, and simulation software provider Ansys.
The data center is said to be fully operational and was demonstrated on the GTC show floor as a digital twin of NVIDIA Omniverse, a platform for building 3D work from tools to applications and services. Another announcement is the introduction of cloud APIs to help developers easily integrate core Omniverse technology directly into Digital Twins’ existing design and automation software applications.
The latest NVIDIA AI supercomputer is based on the NVIDIA GB200 NVL72 liquid cooling system. It has two racks, both containing 18 NVIDIA Grace CPUs and 36 NVIDIA Blackwell GPUs, connected via fourth-generation NVIDIA NVLink switches.
Cadence, another partner mentioned in the announcement, plays a special role with its Cadence Reality digital twin platform, which was also announced yesterday as “the industry’s first comprehensive AI-driven digital twin solution to facilitate sustainable data center design.” and modernization.” The result was a 30% improvement in data center energy efficiency.
The platform is used in this demonstration for a variety of purposes. Engineers unified and visualized multiple CAD (computer-aided design) data sets with “enhanced accuracy and realism” and used Cadence’s Reality Digital Twin solver to simulate airflow and the performance of the new liquid cooling system. Ansys’ software helps bring simulation data into the digital twin.
“This demonstration shows how digital twins allow users to fully test, optimize and validate data center designs before producing physical systems,” NVIDIA noted. “By visualizing data center performance in digital twins, teams can better optimize their designs. and plan for what-if scenarios.”
For all the promise of the Blackwell GPU platform, it needs to run somewhere – and the largest cloud vendors are actively involved in delivering NVIDIA Grace Blackwell. As Huang said, “The entire industry is preparing for Blackwell.”
NVIDIA Blackwell on AWS will “help customers across industries unlock new generative AI capabilities faster,” the companies said in a statement. As early as re:Invent 2010, AWS already had NVIDIA GPU execution entities. Huang Jensen appeared at last year’s re:Invent conference together with AWS CEO Adam Selipsky, which is worth paying attention to.
This stack includes AWS’s Elastic Fabric Adapter Networking, Amazon EC2 UltraClusters, and virtualization infrastructure AWS Nitro. AWS exclusively launched Project Ceiba, an AI supercomputer collaboration project that will also use the Blackwell platform for use by NVIDIA’s internal R&D team.
Microsoft and NVIDIA have expanded their long-term collaboration to also bring the GB200 Grace Blackwell processor to Azure. The Redmond company claims Azure is integrating with the Omniverse Cloud API for the first time. The demo at GTC shows how plant operators can use the interactive 3D viewer in Power BI to view live plant data overlaid on a 3D digital twin of their facility.
Healthcare and life sciences are considered key industries for AWS and Microsoft. The former is working with NVIDIA to “expand computer-assisted drug discovery with new artificial intelligence models,” while the latter promises that numerous healthcare stakeholders “will soon be able to rapidly innovate and increase efficiency in clinical research and care delivery.” “
At the same time, Google Cloud has the advantages of Google Kubernetes Engine (GKE). The company is integrating NVIDIA NIM microservices into GKE to help accelerate the deployment of generative AI in the enterprise and make it easier to deploy the NVIDIA NeMo framework on its platform through GKE and the Google Cloud HPC Toolkit.
However, in order to adapt to the “next generation” theme, it is not only the hyperscale enterprises that need to apply. NexGen Cloud is a resilient infrastructure-as-a-service cloud provider with Hyperstack powered by 100% renewable energy and delivered as a self-service, on-demand GPU-as-a-service platform. The NVIDIA H100 GPU is the flagship product, and the company made headlines in September when it promoted a $1 billion European AI supercloud, promising to deliver more than 20,000 H100 Tensor Core GPUs upon completion.
NexGen Cloud announced that computing services supported by the NVIDIA Blackwell platform will become part of the AI super cloud. “With solutions powered by Blackwell, we will be able to provide customers with the most powerful GPU products on the market, enabling them to drive innovation while achieving unprecedented efficiencies,” said Chris Starkey, CEO of NexGen Cloud.
Image source: NVIDIA
