India Enters Supercomputer Race, Builds 220 TeraFLOPS Machine

The Indian Space Research Organization (ISRO) has just announced that it has started operating the country's fastest supercomputer yet, the SAGA-220 which uses Intel Xeon CPUs and Nvidia Tesla GPGPU add-on cards in order to deliver a peak performance of 220 teraflops.

India's SAGA-220 supercomputer is housed by the Satish Dhawan Supercomputing Facility located at the Vikram Sarabhai Space Centre (VSSC) in Thiruvananthapuram.

The SAGA-220 is fully designed and built by VSSC using commercially available hardware, open source software components and other in house developments.

Its computational power is provided by no less than 400 Nvidia Tesla 2070 general purpose GPUs which are paired with 400 Intel quad-core Xeon CPUs and linked together via a high-speed interconnect.

According to the Indian Space Research Organization, each GPU provides 500 gigaflops of performance while each quad-core Xeon chip adds another 50 gigaflops to the total crunching power of the supercomputer. The system can theoretically scale to about 1000 teraflops.

Aside from touting the processing power of the SAGA-220 system, ISRO also states that its new creation is energy efficient as its power consumption is estimated at about 150 kilowatt.

The new system will be used for running computational fluid dynamics (CFD) applications which should enable scientists to build more complex space launch platforms.

Before the SAGA-220 went live, the fastest Indian supercomputer was the Hewlett-Packard powered Eka at the Computational Research Laboratories in Pune.

This is able to provide a maximum performance of 132 teraflops and is ranked at number 47 on the Top500 supercomputer list.

Right now, China holds the record for the world's fastest supercomputer with the Tianhe-1A that uses a proprietary Chinese interconnect, 7,168 Nvidia Tesla M2050 GPUs as well as 14,336 Intel Xeon X5670 CPUs and has a consistent speed of 2.5 petaflops and a theoretical peak of 4.7 petaflops. (via HPC Wire)