Large companies are always competing to see who has the world ‘s most powerful supercomputer, and the trajectory of IBM was impossible to stay behind. A few weeks ago, China announced its new supercomputer Sunway TaihuLight, a team that achieved a computing power of 93 petaflops, a fact that seems to have caused unrest in the United States and have been quick to confirm that you are financing IBM in order to build another supercomputer reaches 200 petaflops of computing power. Almost nothing.
According to initial reports published, the new IBM supercomputer will be called Summit and will replace the previous Watson. Summit will not only have more than twice the computing power of the largest Chinese supercomputer, but also be much more efficient and small, because to start the Chinese Sunway employs a whopping 40,960 nodes to achieve those 93 petaflops, while Summit will need as only 3,400 nodes to reach 200 petaflops. Not to take many accounts to see the difference.
How they will achieve such computing power?
IBM will use the Summit 3,400 nodes, each equipped with PowerPC processors IBM and 9 GPUs NVIDIA Volta. In addition, each node will have 512 GB of RAM DDR4 and HBM memory, accompanied by 800 GB of non-volatile RAM that will act as the “extended buffer”, dramatically improving the speed and access times system (basically use nonvolatile RAM as hard disk). In addition, the faculty has considerably reduced the number of nodes needed to build this machine is parallelized, the capacity of all nodes are able to work in the same simultaneously.
However, the US and IBM are now completing the development phase of Summit, and still have ahead the work of building the machine, when it announced that it will occur in 2018, so for now we have to wait to find out what You will be able this new IBM supercomputer.