India turns on a new supercomputer
Pram Pravega's 3.3 petaFLOPS should make it India's fourth entrant in the Top500 list
The Indian Institute of Science (IISC) has installed and commissioned a supercomputer that ranks among the most powerful and largest on the subcontinent.
The machine, called Param Pravega, has an Rpeak – or total supercomputing capacity – of 3.3 petaFLOPS. That means it can process a quadrillion floating point operations per second on its ATOS BullSequana XH2000 series system.
The system is based around 428 CPU nodes that employ Intel Xeon Cascade Lake 8268 processors running at 2.9GHz in a two-socket configuration with 48 cores, 192GB RAM (4GB per core), and 480GB SSD local storage per node. These nodes collectively offer 20,544 cores and up to 1.9 petaFLOPS of power.
A further 156 high memory nodes have similar specs but offer 16GB RAM per core and a collective 0.694 petaFLOPS of peak computing capability.
Param Pravega also consists of 40 GPU nodes, each packing Intel Xeon G-6248 2.5GHz processors in a two-socket configuration for 40 cores per node, plus 192GB RAM and 480GB SSD local storage. GPU nodes also host a pair of Nvidia V100 Tesla 16GB cards. The 40 GPU nodes offer 1600 CPU cores and 80 Nvidia V100 for a total 0.688 petaFLOPS, 0.128 from the host CPUs and 0.560 GPU computational capability.
The project was designed by the Centre for Development of Advanced Computing (C-DAC) as part of the government's National Supercomputing Mission (NSM). CDAC built the software stack, based on CentOS 7.x, in-house to satisfy a the "Make In India" government initiative that encourages domestic manufacturing, which has been tasked with installing over 70 supercomputing facilities across academic and research facilities that will link into the National Supercomputing grid.
So far, ten of the 70 supercomputer systems have been deployed, forming a cumulative 17 petaFLOPS of computing power to serve as a platform for the country's research activities – including those related to COVID-19 and other infectious diseases, genomics, climate modelling, weather predictions, as well as telecom networks and anything else that needs big brains to calculate.
- Google Cloud started running its servers for an extra year, still loses billions
- Sure, the James Webb Space Telescope is cool and all. But try making one out of Lego
- Website fined by German court for leaking visitor's IP address via Google Fonts
- Machine learning the hard way: IBM Watson's fatal misdiagnosis
- Trio of Rust Core Team members take their leave
Around 2,600 researchers in India have already used NSM's existing network on about 3,100,000 computational jobs since the installation of its first supercomputer – SahasraT – in 2015. Coincidentally, the XC40 supercomputer with 33,024 Intel Haswell Xeon cores is also located at the IISC.
That original NSM supercomputer no longer makes it onto the Top500 list – the catalogue of Earth's mightiest computing machines. Two Indian XC40s are on the list – Pratyush and Mihir, ranking 121 and 228 respectively. Pratyush is located at the Indian Institute of Tropical Meteorology in Pune and Mihir is employed at the National Centre for Medium Range Weather Forecasting outside Delhi. One other supercomputer in India also makes the Top500 list – an Nvidia DGX A100-powered beast called PARAM Siddhi-AI, located at none other than CDAC.
Based on Param Pravega's technical specs, and particularly its Rmax, it will probably join Pratyush, Mihir and PARAM Siddhi-AI at around 160th place when the list is next updated. ®