From the /c/technology community on Chambers

# Introduction * **Jensen Huang:** Hi. Thank you. It’s great to see everyone again. We’re getting ready for the holiday season, and we’ve got a lot to talk about. * The conference has changed a lot since its first meeting in 2007. There were only a few hundred people there at the time. He jokes that it was a little smaller than today. * Jensen thinks back to 2007, when NVIDIA only had a $1 billion market cap and was just a graphics card company. * He talks about how he was at a conference where a company was touting a supercomputer for $600 million. He thought it was ridiculous. # Grace Hopper Supercomputer * **Jensen Huang:** I’d like to start with the Grace Hopper Supercomputer, the fastest supercomputer in the world, at FLOPS. It’s a collaborative effort by NVIDIA, Cray, and ANL. We’ll be delivering 13 Grace Hoppers over the next few years. Each one will have 200 petaflops. * It is powered by NVIDIA Hopper GPUs, which provide a significant leap forward in AI performance and power efficiency. # H100 Teaser * **Jensen Huang:** Now, let’s talk about H100. This is the successor to Hopper. We’ll be unveiling it in a few months. Jensen says H100 is a big deal, but he’s not ready to share too much yet. # Hopper GPUs * **Jensen Huang:** The Hopper GPUs in Grace Hopper are custom designed to provide the highest AI performance while keeping power consumption in check. # HPC at OpenAI * **Jensen Huang:** OpenAI uses HPC to train models. They have thousands of GPUs running 24/7. They’ve recently switched to Hopper GPUs and have seen a significant boost in performance. # HPC at Google Cloud * **Jensen Huang:** Google Cloud has deployed Hopper-based GPUs for its HPC workloads. They’ve seen a 3x increase in performance per watt compared to their previous GPUs. # HPC at Microsoft Azure * **Jensen Huang:** Microsoft Azure has also deployed Hopper-based GPUs for its HPC workloads. They’ve seen a 2.5x increase in performance per watt. # HPC at NASA * **Jensen Huang:** NASA has deployed Hopper-based GPUs for its HPC workloads. They’ve seen a 3x increase in performance per watt. # HPC at National Labs * **Jensen Huang:** The US Department of Energy’s national laboratories have deployed Hopper-based GPUs for their HPC workloads. They’ve seen a 3x increase in performance per watt. # Omniverse and Simulation * **Jensen Huang:** Omniverse is a platform for building simulation-based science. It uses HPC to run complex simulations. Jensen thinks it’s essential for the next era of scientific discovery. # AI at NASA * **Jensen Huang:** NASA uses AI to analyze data from its missions. They’ve seen significant improvements in AI performance with HPC. # AI at the National Labs * **Jensen Huang:** The US Department of Energy’s national laboratories use AI to analyze complex scientific data. They’ve seen significant improvements in AI performance with HPC. # Exascale Computing * **Jensen Huang:** We’re moving towards exascale computing, which will require a lot of power. Jensen jokes that it’s like powering a small town. # AI and the Environment * **Jensen Huang:** AI is a significant contributor to greenhouse gas emissions, particularly in the data center. Jensen says it’s a big problem. # NVIDIA’s Commitment to Sustainability * **Jensen Huang:** NVIDIA is committed to sustainability and reducing its carbon footprint. He jokes that he’s buying an electric car and has solar panels installed at his house. # Closing * **Jensen Huang:** Thank you. Happy holidays. # Post-Keynote Announcements * **Grace Hopper Supercomputer:** The Grace Hopper Supercomputer, powered by NVIDIA Hopper GPUs, is the world's fastest supercomputer, delivering 5.baraflops of performance. The system will be delivered over the next 3 years, with each one capable of 200 petaflops. * **H100 Teaser:** NVIDIA is set to unveil H100, the successor to Hopper, in a few months, marking a significant leap forward in AI performance and power efficiency. # Key Takeaways * The Grace Hopper Supercomputer, powered by NVIDIA Hopper GPUs, is the world's fastest supercomputer, delivering 5.6 exaflops of performance. * NVIDIA is collaborating with Cray and ANL on the Grace Hopper Supercomputer, with 13 systems to be delivered over the next 3 years, each capable of 200 petaflops. * H100, the successor to Hopper, is set to be unveiled in a few months, representing a significant leap in AI performance and power efficiency. * OpenAI, Google Cloud, Microsoft Azure, NASA, and the US Department of Energy's national laboratories are leveraging Hopper GPUs for AI workloads, achieving substantial performance increases. * NVIDIA is committed to sustainability, acknowledging AI's significant contribution to greenhouse gas emissions, particularly in the data center.

NVIDIA GTC Keynote Summary - Grace Hopper Supercomputer, H100 Teaser, and More

Comments (1324) 37658 👁️