The Indian Institute of Science, Karnataka, has claimed to have launched India’s most powerful supercomputer, Param Pravega. As of 2020, India has two supercomputers, PARAM Siddhi-AI and Mihir, that have made it to the Top 500 global supercomputer lists, ranking 63rd and 146th, respectively. Param Pravega, with its claims, could possibly beat these computers in being the most powerful Indian supercomputer. But is it true?
Analytics India Magazine explores how Param Pravega stands in comparison to top Indian and global supercomputers.
About the supercomputer
Param Pravega is designed by the Centre for Development of Advanced Computing, with most of the components being made in India. It is set up under the National Supercomputing Mission that aims to empower Indian academic institutions with supercomputing grids in over 70 HPC facilities. The machine is a resource for executing high-performance computing applications as well as hosting various tools and libraries. Given this characteristic, Param Parvega has been leveraged by all; faculty members, students and researchers at the facility for studies on COVID-19, modelling viral entry, protein study, green energy technologies, climate change, and more.
Currently, ten of these supercomputers have been deployed at IISc, IISER, IIT, JNCASR, C-DAC, and NABI-Mohali, offering about 3,100,000 computational jobs and enabling 2,600 cross country researches. The supercomputer falls under the National Knowledge Network program of the government that connects academia, research and development labs over a high-speed network.
The software stack
Param Pravega has a total supercomputing capacity of 3.3 petaflops, with the ten supercomputers adding up to 17 petaflops. The software runs on Linux OS based CentOS 7.x distribution with two master nodes, 11 login nodes, two firewall nodes, four management nodes, one NIS slave, and 624 (CPU + GPU) compute nodes. These are further divided into regular CPU nodes, high-memory CPU nodes, and GPU nodes. Intel Xeon Cascade Lake processors are used for the CPU nodes, and NVIDIA Tesla V100 cards are used on the GPU nodes. A FAT-tree topology with a 1:1 subscription ratio and 4-petabyte parallel storage connects the nodes and provides parallel file system access.
Param Pravega compared to the top global supercomputers
In comparison, India’s PARAM Siddhi, 63rd on the top 500 list, has met several performance benchmarks, including 4.6 petaflops sustained double precision, 6.5 petaflops of peak double-precision, and overall 210 AI petaflops. It was built on NVIDIA DGX SuperPOD reference architecture networking and C-DAC’s indigenously developed HPC-AI engine. The DGX SuperPOD reference architecture is known for providing blueprints for assembling world-class infrastructures and powerful supercomputers.
Weather forecasting supercomputer Pratyush at the Indian Institute of Tropical Meteorology ranked 78th on the November edition of the list. Along with Pratyush, Mihir, another supercomputer noted on the list, is a 2.8 Petaflop machine. Together, Pratyush and Mihir consist of several computers that can deliver a peak power of 6.8 PetaFlops, making them the first multi-PetaFlops supercomputer ever built in India. Pratyush and Mihir are employed for weather forecasting and climate monitoring, accurately forecasting monsoon, fishing, air quality and catastrophic events such as tsunamis, cyclones, earthquakes, lightning, floods, droughts, and other natural calamities. India is the fourth company after Japan, the United States, and the United Kingdom to have a high-performance computing facility dedicated to weather and climate research.
NVIDIA Cambridge-1, the top supercomputer in the UK, also runs on the NVIDIA DGX SuperPOD cluster. Additionally, the stack consists of NVIDIA A100 GPUs, BlueField-2 DPUs and NVIDIA HDR InfiniBand networking, which add up to deliver 8 petaflops of performance and 400 petaflops of AI performance.
The world’s fastest supercomputer, Fugaku, reported an initial performance of Rmax of 416 petaFLOPS in the FP64 high-performance LINPACK benchmark. However, after last year’s upgrade, Fugaku’s performance increased to an Rmax of 442 petaFLOPS.
Pangea III – the world’s most powerful commercial supercomputer built by IBM, has a computing power of 25 petaFLOPS, with a storage capacity of 50 petabytes. It is built on IBM POWER9 AI, the company’s high-performance architecture with an industry-exclusive technology to handle massive data loads.
Recently, Meta introduced their AI Research SuperCluster (RSC) supercomputer in collaboration with NVIDIA. It uses 760 NVIDIA DGX A100 systems and consists of 6,080 NVIDIA A100 GPUs linked on an NVIDIA Quantum 200Gb/s InfiniBand network giving 1,895 petaflops of TF32 performance.
IISc’s track record with supercomputers is impressive. Their SahasraT was the fastest supercomputer in India in 2015 and has been used by the faculty and students for impactful research studies since then. Param Pravega, with its claims, ensures a similar productivity promise.
