Supercomputers
Introduction to Supercomputing
Definition
Supercomputers are the physical embodiment of High-Performance Computing (HPC). They are highly advanced machines designed to solve incredibly complex and massive computational problems that would be impossible for standard, general-purpose computers to handle.
Core Mechanism
Unlike standard computers that process tasks sequentially, supercomputers utilize multiple Central Processing Units (CPUs). They break down a massive required task into multiple smaller parts and process them simultaneously, a method known as parallel processing, resulting in ultra-high-speed computation.
Measurement of Performance
- The performance of a supercomputer is not measured in standard clock speed (like GHz used for regular laptops) but in FLOPS (Floating-Point Operations Per Second).
- Modern supercomputers are typically evaluated in PetaFLOPS (quadrillions of operations per second) and are currently transitioning into the ExaFLOP (quintillions of operations) era.
Key Features and Characteristics
Supercomputers possess several distinctive features that set them apart from conventional machines:
- Multiple CPUs: They incorporate thousands of processors, each containing circuits dedicated to interpreting program instructions and executing complex arithmetic and logic operations simultaneously.
- Massive Memory and Speed: Speed and memory are the two absolute prerequisites of a supercomputer. Their memory capacity averages around 250,000 times that of a normal personal computer.
- High Storage Capacity: They require immense storage systems to allow for the rapid retrieval of stored data and instructions, which is crucial to support their extreme computational speeds.
- Vector Arithmetic: Instead of operating on mere pairs of single numbers (scalar arithmetic), supercomputers use Vector Arithmetic to operate on pairs of lists of numbers simultaneously, drastically speeding up scientific calculations.
- Multi-User Capability: A supercomputer is not a single-user desktop; it allows multiple users (such as hundreds of scientists) to access its computing power simultaneously.
- Specialized Infrastructure: Due to their massive power draw and the heat they generate, supercomputers are highly expensive and must be housed in large, clean rooms equipped with specialized high-air-flow cooling systems.
Applications of Supercomputers
Because of their high cost, they are used by governments and large research organizations for special areas requiring complicated calculations:
- General Purpose: Weather forecasting, climate modelling, quantum mechanics, and simulating nuclear detonations.
- Special Purpose: Historically, supercomputers have been built for highly specific tasks, such as playing and mastering complex games to test AI capabilities. Notable examples include Belle, Hydra, and IBM’s Deep Blue (famous for defeating world chess champion Garry Kasparov).
Evolution of Computing Speed (The FLOPS Scale)
To understand the historical progression of supercomputing power, it is important to know the scale of FLOPS and the decade they were achieved.
Unit | Mathematical Value | Historical Example | Decade Achieved |
Hundred FLOPS | 10 to the power of 2 | ENIAC | 1940s |
KFLOPS (KiloFLOPS) | 10 to the power of 3 | IBM 704 | 1950s |
MFLOPS (MegaFLOPS) | 10 to the power of 6 | CDC 6600 | 1960s |
GFLOPS (GigaFLOPS) | 10 to the power of 9 | Cray-2 | 1980s |
TFLOPS (TeraFLOPS) | 10 to the power of 12 | ASCI Red | 1990s |
PFLOPS (PetaFLOPS) | 10 to the power of 15 | Jaguar | 2010s |
EFLOPS (ExaFLOPS) | 10 to the power of 18 | Frontier (Under rapid global development) | 2020s |
Evolution of Supercomputing
The development of supercomputing began in the late 1950s, heavily funded by the U.S. government for military applications. For competitive exams, the following technological milestones are important:
- Cray-1 (1976): Recognized as the first successful implementation of vector processing.
- Cray-2 (1985): A four-processor system that became the first computer to exceed one billion FLOPS (GigaFLOPS).
- Roadrunner: Built by IBM, it was the first supercomputer to break the PetaFLOPS barrier (exceeding 1,000 TeraFLOPS).
Note on Contemporary Systems: The United States and China historically dominate the global TOP-500 supercomputer list with massive systems like the IBM Summit and Sierra (USA), and Sunway TaihuLight and Tianhe-2 (China).
The Genesis of India's Supercomputing Programme
India’s journey into supercomputing was born out of technological denial. In 1987, the USA denied India’s request to purchase the Cray X-MP supercomputer, which India needed for academic research and weather forecasting. This denial acted as a catalyst, compelling the Indian government to launch a completely indigenous supercomputing development programme.
Key Agencies and Indigenous Supercomputers
Following the denial, several national research agencies took the lead in developing high-performance computing systems:
- C-DOT (Centre for Development of Telematics): Developed CHIPPS, a high-performance Parallel Processing System.
- BARC (Bhabha Atomic Research Centre): Created the Anupam series of supercomputers specifically for nuclear research.
- ANURAG (Advanced Numerical Research and Analysis Group): A DRDO laboratory that created the PACE series for defense applications.
- C-DAC (Centre for Development of Advanced Computing): Headquartered in Pune, it is the primary agency responsible for the famous PARAM series.
The PARAM Series
- The word PARAM means “supreme” in Sanskrit and also serves as an acronym for “PARAllel Machine”.
- PARAM 8000 (1991): This was C-DAC’s first major success, delivering a parallel computer in the GigaFLOPS range, placing India on the global supercomputing map.
- Subsequent Upgrades: It was followed by the PARAM 9000 series in 1993 (5 GFLOPS peak) and the PARAM 10000 series in 1998 (38 GFLOPS sustained).
Current Status: Indian Supercomputers in the Global TOP 500
India continues to build world-class systems. As of the June 2023 TOP-500 list, four Indian supercomputers hold global rankings. AIRAWAT and PARAM Siddhi – AI are currently the fastest supercomputers in India.
Global Rank | Supercomputer Name | Host Institution | Processing Speed (Rmax) |
75 | AIRAWAT – PSAI | Centre for Development of Advanced Computing (C-DAC), Pune | 8.5 PetaFLOPS |
131 | PARAM Siddhi – AI | Centre for Development of Advanced Computing (C-DAC), Pune | 4.62 PetaFLOPS |
169 | Pratyush (Cray XC40) | Indian Institute of Tropical Meteorology (IITM), Pune | 3.76 PetaFLOPS |
316 | Mihir (Cray XC40) | National Centre for Medium-Range Weather Forecasting (NCMRWF), Noida | 2.57 PetaFLOPS |
