The latest edition of the Top500 supercomputer rankings showed that HLRS’s Hawk is one of the fastest non-accelerated supercomputers in the world, according to the organization’s November 2020 HPCG (High-Performance Conjugate Gradient) benchmark. With a speed of 334.65 teraflops per second (more than 334 trillion computing operations per second), Hawk placed 18th in the worldwide HPCG list, and is the fastest European supercomputer on the list that is constructed on a purely CPU-based architecture.
The results were announced today in conjunction with the International Conference for High Performance Computing, Networking, Storage, and Analysis (SC20), the world's largest meeting of the high-performance computing (HPC) community.
The HPCG benchmark provides a consistent metric for assessing and comparing supercomputer performance. The score is based on results of running a standardized set of tasks that are typical of what HPC systems perform.
"HLRS's focus has always been on providing HPC systems that are optimized for our users' needs," emphasized HLRS Director Dr.-Ing. Michael Resch,. "The HPCG benchmark is important for us because it best reflects the kinds of calculations that our users most often require, and so is an accurate indicator of the kind of system performance they can expect. CPUs are the workhorses for simulation in computational engineering and the applied sciences, and have long formed the core of our systems. We are pleased that the new HPCG results confirm that we continue to maintain a leadership position in the field."
In addition to the newest HPCG rankings, SC20 also saw the announcement of the November 2020 Top500 List, which is based on the LINPACK test, a different benchmark that assesses the speed at which a supercomputer solves a complex system of linear equations. Hawk achieved a maximum LINPACK performance of 19.3 Petaflops, debuting at place number 16 on the overall Top500 List.
Hawk, a Hewlett Packard Enterprise Apollo 9000 system, was installed earlier this year at HLRS. Its theoretical peak performance — a mathematical calculation based on the number of processors and speed of each processor — is approximately 26 Petaflops.
Hawk is one of several HPC systems currently in operation at HLRS, including complementary systems that integrate GPU cores, also called accelerators. In addition to improving certain kinds of simulations, these other systems can support HPC applications for machine learning, high-performance data analysis, and artificial intelligence.