There are rankings for everything. Also supercomputers, and the one we present on these pages boasts of being the most powerful of those built to date. They have named it El Capitan in honor of the 1,742 exaflops of power it handles. The machine works with a 100% fanless direct liquid cooling system that places it among the top 20 on the list of most sustainable supercomputers -yes, there is that other classification too. And sustainability is not a trivial topic. Even more so if we take into account that El Capitan will work for the Lawrence Livermore National Laboratory (LLNL) of the United States Department of Energy. Among its missions, we can mention support for research on climate change, energy security, modernization of the electrical grid, drug discovery and other strategic fields.
Designed by Hewlett Packard Enterprise (HPE), that 1,742 exaflops power will deliver a performance of 58.89 gigaflops per watt. “El Capitan marks another major milestone in exascale supercomputing, combining exceptional performance, energy efficiency, and breakthrough capabilities to accelerate AI-powered scientific discovery. Additionally, it will enable breakthroughs in national security and open new opportunities in renewable energy.” “said Trish Damkroger, senior vice president and general manager of HPC & AI Infrastructure Solutions at HPE. He adds that, for the configuration and design of this machine, HHPE engineers have worked over the last five years with experts from the United States Department of Energy, the National Nuclear Security Administration, Lawrence Livermore National Laboratory and the manufacturer of AMD processors.
From the company responsible for El Capitan, they trust that “it will allow the United States to maintain its leadership in national security and will strengthen the capacity of the Tri-Labs of the National Nuclear Security Administration, formed by the Lawrence Livermore National Laboratory, the National Laboratories of Sandia and the Los Alamos National Laboratory. They add that the laboratory in which it has been installed “plans to use artificial intelligence models to address both classified and unclassified workloads.”
“This long-awaited asset will allow us to conduct high-fidelity 3D modeling and simulations essential to effectively fulfill our national security mission,” explains Rob Neely, director of LLNL’s Weapons Computing and Simulation Program. “In addition, we are committed to maximizing the use of artificial intelligence on El Capitan, leveraging it to train and run large-scale AI models, which will make our calculations faster, more efficient and, potentially, more accurate. Having a single system that “Combining modeling, simulation and AI workloads so efficiently is a truly exciting development.”
From the United States Department of Energy they explain that El Capitan will also address secondary missions related to national security such as nuclear non-proliferation and the fight against terrorism. Additionally, it will be used in advanced research in materials discovery, nuclear data and high-energy-density science, such as inertial confinement fusion research conducted at the National Ignition Facility. And they add that the advances achieved thanks to El Capitan will also have applications in public domain projects such as the discovery of new drugs and other strategic fields.
Regarding the more technical composition of El Capitan, HPE explains that its platform integrates the HPE Cray EX supercomputing solution, which combines the AMD Instinct MI300A APUs with CPU and GPU cores along with high-bandwidth memory in a single package. Add to that the HPE Slingshot high-speed interconnect and a custom storage solution. HPE Slingshot acts as the backbone of the system, enabling massive computations on more than 11,000 nodes. As a result of the public-private collaboration, LLNL and HPE have also developed a custom local storage solution close to the nodes, designed to reduce latency. This dynamically configurable system organizes data into tiers within a Luster-based global file system, efficiently shared among all compute nodes.
#worlds #fastest #supercomputer #starts #fight #climate #change
