China and the United States compete at the forefront of the supercomputing race. China currently leads the way with the announced Sunway TaihuLight supercomputer placing first on the Top 500 List. This processor reached a Linpack benchmark of 93 petaflops, and claimed 124.5 petaflops at its peak, making it the fastest computer in present day.
The US seeks to not only reach a competitive speed, but to surpass it entirely. The United Sates Department of Energy responded to China with claims of a new supercomputer under development by IBM at the Oak Ridge National Laboratory. This new processor dubbed Summit will have a 200 petaflops capability at its peak. The revelation of the new computer gave a shock to many, especially as the claims are double the TaihuLight’s current capabilities. Until the Summit is released, the TaihuLight will remain to be on the first place.
The Department of Energy is confident in the goal set, despite how far-reaching it is believed to be by others. Since 1993, supercomputing abilities have exponentially increased by a factor of 300,000. Cray’s Titan, and IBM’s Sequoia are two US supercomputers trailing Chinese processors at #3 and #4 on the Top 500 list. Though there’s a large gap of 75.42 petaflops between the TaihuLight and the Titan, by 2018, US supercomputers have been predicted to reach 150 petaflops. That forecast, and the recent breakthrough from China, make IMB’s goal appear more attainable.
Though specifics have not been released, it is known that the processor will include thousands of Power8 Central Processing Units designed by IBM and Nvidia’s Volta GPUs, both manufactured for the highest-end supercomputers. IBM plans on ensuring the processor is extra efficient, especially when considering the overall power in these supercomputers. More assumptions are made about the type of capabilities Summit will really have. If IBM expects to surpass the bar, it has to take a data centric computing approach in the development of the architecture of the processor.