HPCG Performance List announced at SC20

The new HPCG Performance List was announced at the SC20 conference, a virtual event. The session entitled TOP500, Green500 & HPCG Awards took place on Monday, November 23, 2020.

The HPCG results are now integrated into the TOP500 list. The main HPCG list of results contains the complete list of submissions some of which might have not made the TOP500 list.

The bi-annual High Performance Conjugate Gradients (HPCG) Benchmark list was announced on November 23, 2020 at the ISC 2020 conference.

This is the fourteenth list produced for the benchmark designed to complement the traditional High Performance LINPACK (HPL) benchmark used as the official metric for ranking the TOP500 systems. The first HPCG list was announced at ISC 2014 over seven years ago, containing only 15 entries. Since then, HPCG list was released bi-annually and the number of entries has steadily increased as the new results became available. The current list contains nearly 187 entries as HPCG continues to gain traction in the HPC community.

It is now customary that the new list contains entries from many of the TOP500's highest ranked, mainly top-50 systems — 75 system cross-list machines in total. The current list is no exception. However, the HPCG ranking continues to provide a significantly different ordering of the machines when compared with the TOP500 ranking. A major change in TOP500 ranking is now reflected in the HPCG results. The upgraded Fugaku system from Japan leads the list with over 16 Pflop/s of achieved performance. Summit and Sierra supercomputers follow on the second and third positions, respectively. Their combined performance is currently 4 times lower than the result from the upgraded Fugaku supercomputer.

Summary highlights from the 14th HPCG list include:

• The #1 entry is the upgraded Fugaku Supercomputer housed at RIKEN, that is over 5.5x faster than the 2nd-ranked Summit hosted at ORNL.
• Fugaku is nearly as fast as all the other entries on the list combined which indicates large performance skew at the top of the list.
• Entering the top-10 part of the list now requires HPCG performance of over 0.5 Pflop/s.
• JUWELS Booster Module added another generation of GPU accelerators from NVIDIA outside of NVIDIA's vendor sites.
• Large number of systems were added featuring AMD EPYC CPUs to increase the dominance of x86 systems on the list..