Four Key Highlights From NVIDIA GTC 2023
AI inference platforms, DGX Cloud, NVIDIA cuLitho and more from NVIDIA GTC 2023.
NVIDIA kicked off this year’s GPU Technology Conference, or GTC, on Monday, March 20. Like previous years, NVIDIA continues its charge to capture the AI hardware market. At NVIDIA GTC 2023, the hardware maker has announced multiple offerings besides its trademark computing and processing units.
This year, the unofficial theme seems to be accessibility. The overwhelming success of ChatGPT and its rapid adoption and implementation across several existing products and services led NVIDIA CEO Jensen Huang to proclaim, “We are at the iPhone moment for AI,” during his keynote address.
As such, the company is ensuring its AI products are accessible for what seems to have blown open a new vertical, i.e., generative AI applications and enterprise-grade cloud-delivered AI infrastructure. Let us see what these are.
NVIDIA GTC 2023 Highlights (So Far)
AI inference platforms
NVIDIA released three computing platforms for AI inferencing, the process of leveraging trained neural networks to assess and make a prediction. These include DGX H100, NVIDIA L4 and H100 NVL.
According to NVIDIA, DGX H100 delivers 9x performance, 2x faster networking, high-speed scalability, and more.
The NVIDIA L4 is a standard accelerator designed for efficient video, AI and graphics, applicable universally across a multitude of servers and capable of delivering 120x faster than the fastest CPU platform and offers greater energy efficiency (99% lesser energy consumption).
H100 NVL is designed with an accelerated Transformer Engine and 94 GB of memory for large language model inferencing in real-time. Based on the Hopper architecture, introduced at NVIDIA GTC 2022 and now generally available, H100 NVL delivers 12x faster performance than the A100 in GPT3 inference. Additionally, NVIDIA L40 is for 2D and 3D image generation.
Early adopters include Microsoft (for DGX H100) and Google Cloud (for NVIDIA L4).
See More: Monetizing the Math: AI Strategies that Boost ROI for Enterprise Investments
DGX Cloud
NVIDIA has taken its DGX AI supercomputing service to the cloud. Cloud-based data acquisition, model training and inferencing are widely perceived as the next step in delivering AI supercomputing services to a broader customer base.
DGX Cloud eliminates the excessive costs associated with on-premise AI systems. It simplifies access to AI supercomputing infrastructure and related software to a web browser. Of course, the $36,999 price tag may seem a tad steep for one DGX Server box that comes with 640GB of memory and eight Nvidia H100 (or A100) GPUs.
However, it is far lesser than the hundreds of thousands of dollars that would otherwise be required for on-premise AI infra. Also, it does simplify advanced model training at scale without a client needing internal resources or infrastructure. It comes with pre-trained models, frameworks, accelerated data science software libraries, and NVIDIA AI Enterprise 3.1.
DGX Cloud is currently available on Oracle Cloud and will come to Microsoft Azure by next quarter and later to Google Cloud and others.
NVIDIA cuLitho
Beyond AI-focused product offerings, NVIDIA is looking to accelerate the design and manufacture of a new generation of chips. NVIDIA cuLitho is a software library intended for computational lithography developed in collaboration with foundry leaders TSMC, ASML, and electronics design company Synopsys. All three have also adopted cuLitho.
Huang said, “The chip industry is the foundation of nearly every other industry in the world. With lithography at the limits of physics, NVIDIA’s introduction of cuLitho and collaboration with our partners TSMC, ASML and Synopsys allows fabs to increase throughput, reduce their carbon footprint and set the foundation for 2nm and beyond.”
NVIDIA claims cuLitho will perform 40x more than existing lithography techniques. Fabrication units using cuLitho could produce 3-5x more photomasks daily with 9x less power than current configurations. As such, the company said 500 NVIDIA DGX H100 systems could do the work that 40,000 CPU systems achieve.
“The cuLitho team has made admirable progress on speeding up computational lithography by moving expensive operations to GPU,” said Dr. C.C. Wei, CEO of TSMC. “This development opens up new possibilities for TSMC to deploy lithography solutions like inverse lithography technology and deep learning more broadly in chip manufacturing, making important contributions to the continuation of semiconductor scaling.”
ASML CEO Peter Wennink added that the company plans to integrate support for GPUs in all of its computational lithography software products.
See More: NVIDIA and Microsoft Join Forces To Build a Scalable Generative AI Supercomputer
Omniverse Cloud collaboration with Microsoft
Modeled after the metaverse, NVIDIA Omniverse is a world simulation and 3D design collaboration platform. The tool was released to make up for the loss in revenue after Ethereum’s shift to Ethereum Merge.
Omniverse Cloud is a suite of cloud services that enables content creators, artists, and developers to access the Nvidia Omniverse platform over the cloud. Powered by GeForce Now, Omniverse Cloud allows creators to design 3D workflows and collaborate further without needing on-premise computing power.
Besides NVIDIA DGX Cloud, Microsoft Azure has also signed up with NVIDIA as the cloud service provider for Omniverse Cloud, which will be available as a platform-as-a-service, including access to a full-stack environment for designing, developing, deploying and managing industrial metaverse applications.
As part of the collaboration, the Redmond-based IT giant’s Microsoft 365 applications, including Teams, OneDrive and SharePoint, will be available on Omniverse Cloud.
Three companies, viz., BMW Group, Geely Lotus and Jaguar Land Rover, have already adopted Omniverse Cloud. Jaguar Land Rover has integrated Omniverse with its vehicle dynamics models, virtual electronic control units, virtual automotive networks and cloud infrastructure. The company is using it to train AI models and validate perception and control algorithms through real-world driving scenarios.
“NVIDIA Omniverse has given us an unprecedented ability to design, build and test complex manufacturing systems, which means we can plan and optimize a next-generation factory completely virtually before we build it in the physical world,” said Milan Nedeljković, board member for production at BMW AG. “This will save us time and resources, increase our sustainability efforts and improve operational efficiencies.”
Watch the NVIDIA GTC 2023 keynote below:
Let us know if you enjoyed reading this news on LinkedIn, Twitter, or Facebook. We would love to hear from you!
Image source: Shutterstock
MORE ON ARTIFICIAL INTELLIGENCE
