The Summit computer can perform 200 quadrillion calculations per second or 200 petaflops. This makes it the fastest computer by far. According to Wikipedia the previously fastest super computer owned by the China supercomputer center in Wuxi called Sunway MPP is capable of peak performance of 125.43 Petaflops. It is interesting to mention that China’s fastest super computer uses the country's own Sunway SW26010 260 core many core 64-bit RISC processor.
The US Government Summit with 200 Petaflops clearly outperforms the fastest from China and it is interesting to mention that the third of the supercomputer top ten list belongs to the National Supercomputing Center in Guangzhou China - the model Tianhe 2 using Xeon E5 2692 and Xeon Pi 31S1P combination only scores 54.90 Petaflops. The IBM Power Summit computer is almost four times faster than the third on the list. Summit can run at 1.88 exaops and analyze millions of genomes.
The top speed of this supercomputer can reach 3.3 exaops using mixed precision calculations. This computational power can be used to research on new materials, calculate molecules and bonds and the way universe and super novas work. The applications are limitless as cancer researchers would be able to use such a fast machine too. This computation power is just a prelude to the speeds and power of commercial quantum computers.
Smallest fastest Supercomputer
The world’s fastest is currently the world's smallest supercomputer too. Dr. John E. Kelly, III a Senior Vice President, Cognitive Solutions and IBM Research, confirms that the supercomputer with 4,600 individual nodes could fill two tennis courts.
The summit supercomputer has more than 9,000 22-core IBM Power9 processors and more than 27,000 NVIDIA Tesla V100 GPUs. This is a lot of computational power and since the machine is water-cooled, the system needs 4,000 (15,141.65 liters) gallons of water a minute to cool it down. The Summit uses enough power to run 8,100 homes.
The Summit computer is optimized for AI and a data intense world and the combination of Power 9 processors and Nvidia V100 Volta GPUs are designed to solve robust data problems. IBM calls the solution a whole new heterogeneous architecture that results in unparalleled performance critical for new applications.
The commitment to make this supercomputer with the most advanced technologies and architectures started back in 2014. The US Department of Energy was the organisation which ordered this computational beast. Summit is the first step towards the next goal, the Exaflop (billion billions) calculation computer. The US wants to beat the China to this Exaflop goal and has contracted IBM, Nvidia and Intel to help. Each of the Summit-like machines will cost between $400 and $600 million and the Exascale should be within reach between 2021 and 2023.
IBM has a history of making really fast computers from Deep Blue to Blue Gene. To refresh your memory, Deep Blue was the computer who was able to beat Kasparow in a chess game back in 1996. The best of technology that time was capable of evaluating 200 million positions per second and score 11.38 GigaFlops performance. Deep Blue was the 259th fastest computer in the world in 1997, some 21 years before.
Blue Gene with 104 racks was the fastest computer between November 2004 – June 2008 with 596 TFLOPS peak.
From 0.586 to 200 Petaflops in a decade
If you compare Blue Gene between November 2004 and June 2008 to the currently fastest Summit computer, you can easily see that the new machine ends up 335.57 times faster. This is an incredible performance jump within one decade
IBM launched a very promising Power System AC922 platform powered by IBM Power 9 based servers. Businesses from banking to healthcare to retail and transportation can benefit from the new architecture created with new data sets and artificial intelligence in mind.
Each Summit node contains two IBM POWER9 CPUs and six NVIDIA Tesla GPUs all water cooled. One important thing to mention is that this platform uses NV Link 2 very fast interconnect capable of incredible speeds of 150 Gigabyte per second per card. This is, of course, if the Power 922 platform is using four GPUs and in case of six V100 solution each GPU scores 100 GB/s still incredible speed.
Bear in mind that AI and grasping the next generation data sets will create winners and losers in the new economy. It affects businesses but governments too and obviously the fact that the US now has significantly faster supercomputer than China is a statement on its own.