IBM is working with Nvidia to create a new generation of supercomputers

US Department of Energy has allocated 325 million dollars to IBM for the creation of two supercomputers that have already received their own names: Sierra and Summit. Systems will be used technologies IBM, Nvidia processor accelerators and network infrastructure from Mellanox.

One of supercomputers will perform calculations on projects that are needed for power engineers, meteorologists, physicists and representatives of other professions. This work will serve as Sierra, and installed system will be in Oak Ridge National Laboratory (Tennessee). The second supercomputer, Summit, will work for the military in the Lawrence Livermore National Laboratory (California).

systems are identical. Peak performance of each of supercomputers will be 100 petaflops, which automatically puts the system to 1st place in the list of Top500 . According to rough estimates of experts Nvidia, to achieve the performance you need to collect 3 million powerful laptops into a single system.

Additionally, US Department of Energy will provide approximately $ 100 million on the project FastForward2. The aim of the project - the creation of the next generation of supercomputers that will be 20-40 times more productive than the current system.

It is now known that in supercomputers will be installed Power processor family from IBM, and graphics accelerators Volta from Nvidia. Joining Power8 and Volta planned through the interface Nvidia NVLink. This interface, according to experts, allows faster data transfer from 5 to 12 times. According to representatives of IBM, this project is one of progressive ideas - placing data centers as close as possible to the data store, to increase performance of supercomputers.

To improve performance, the system also will be used not the usual hard disks and system with flash-drives. As mentioned above, the third partner in the project is the company Mellanox, which supplies high-performance network hardware products that support InfiniBand.

Source: habrahabr.ru/company/ibm/blog/247153/