[ad_1]
In August 2018, Dell EMC and Intel announced plans to jointly develop Frontera, a $ 60 million grant-funded university-based supercomputer from the National Science Foundation, to replace Stampede2 at the University of Texas at the Texas Advanced Computing Center. TACC). These plans came to fruition in June when both companies deployed Frontera, which was officially unveiled this morning.
Intel claims that Frontera can achieve maximum performance of 38.7 quadrillion floating-point operations per second, making it the fastest computer in the world designed for academic workloads such as modeling and simulation. , big data and machine learning. (This is compared to the 18 petaflops at Stampede2.) Earlier this year, Frontera won the fifth place on the Top500 list twice a year with 23.5 petaflops at the LINPACK benchmark, which ranks the non-distributed computer systems. the most powerful in the world.
"The Frontera system will provide researchers with artificial intelligence and artificial intelligence capabilities that did not exist before for academic research," said Intel vice president and general manager Trish Damkroger, an extreme computing company. . "With Intel technology, this new supercomputer opens up new opportunities in science and engineering to advance research, including cosmic understanding, medical treatments and energy needs."
Hundreds of 28-cores (Cascade Lake) Xeon second-generation scalable processors in Dell EMC PowerEdge servers support Frontera's heavy computing workload, as well as Nvidia nodes for single-precision computing. The chip architecture is based on Intel's Advanced Vector Extensions 512 (AVX-512), a set of instructions that provides twice as much FLOPS per clock as the previous generation.
Frontera uses liquid cooling technology for the majority of its nodes. Dell EMC provides cooling and water cooling solutions from CoolIT and Green Revolution Cooling, systems integration companies, and leverages Mellanox HDR and HDR-100 interconnects to transmit data at speeds up to 200 Gbps per second. link. between the switches that connect its 8,008 nodes. Each node is expected to consume about 65 kilowatts, about one third of which comes from wind power credits and its production, as well as solar energy.
In terms of storage, Frontera has four different environments designed by DataDirect Networks, which total more than 50 petabytes coupled with 3 petabytes of NAND Flash capacity. (This equates to about 480 GB of SSD storage per node.) Three are general purpose, while the fourth offers "very fast" connectivity of up to 1.5 terabytes per second.
Finally, Frontera takes advantage of Intel Optane DC's persistent memory, a non-volatile memory technology developed by Intel and Micron technology, PIN-compatible with DDR4, and combining large caches with a smaller DRAM pool (192 GB per node). for better performance. Coupled with the latest generation of Xeon scalable processors, Intel sets Optane DC PM performance at 287,000 operations per second (compared to 3,116 operations per second for traditional DRAMs and storage), with a restart time of only 17 seconds .
Manuela Campanelli, professor of astrophysics at the Rochester Institute of Technology and director of the center for computational relativity and gravitation, is already using Frontera to develop a simulation that could explain the origin of the energy bursts emitted during the merger. a neutron star. Professor George Biros of Austin Frontera University drew on this initiative to develop biophysical models of brain tumor development to more effectively diagnose and treat gliomas, a type of brain tumor. And Olexandr Isayev, an assistant professor at the University of North Carolina at Chapel Hill, uses this system to form an AI model that describes the force fields and the potential energy of the molecules as a function of their 3D structure.
Frontera joins more than a dozen state-of-the-art computer systems currently deployed by TACC, including Lonestar and Maverick, and is expected to operate for five years. Its next phase will involve application-specific accelerators, including quantum simulators and tensor core systems, which will provide an overall computing factor 10 times faster.
"Frontera will provide scientists across the country with access to unprecedented computing modeling, simulation and analysis capabilities," said NSF's Assistant Director for Computing, Computing and Engineering. . "Frontera is the next step in NSF's more than three decades of support for advanced computing capabilities that allow the United States to maintain its global leadership in the field of research."
According to TACC, up to 80% of the hours available on Frontera will be accessible via the NSF Petascale Computer Resource Allocation Program.
[ad_2]
Source link