evolution of supercomputers
Evolution of Supercomputers
Supercomputers are not your average computers. They are designed to work at very fast speeds, handling huge amounts of data and performing multiple complex tasks at the same time. In the following article, take an in-depth look at the evolution process of these computing giants.
- Had 2 Central Processing Units (CPU's) and one I/O (input/output) processor.
- Had a core memory power of 8 banks, which stored 20,000 words.
- Accessing memory speed was 8 microseconds and cycle time was 4 microseconds.
- 1 CPU for arithmetic and logical operations, different simpler processors (I/O processors or peripheral processors) for other tasks.
- Introduced Reduced Instruction Set Computer (RISC) concept, where instruction set of the main CPU was smaller, different processors could work in parallel and clock speed was very fast (10 MHz).
- Introduced logical address translation.
1969 | CDC 7600 was released. It surpassed the 6600 with a clock speed of 36.4 MHz and used a pipelined scalar architecture. It surpassed the 6600 by 10x times, with its performance figure of 10 MFLOPS. |
1972 | Seymour Cray left CDC to form his own computing firm, Cray Research. |
1974 | CDC released the STAR-100, a supercomputer with a vector processor. It had a performance speed of 100 MFLOPS. |
1976 | The Cray-1 was unveiled, a machine with a vector processor and had a clock speed of 80 MHz and a performance figure of 160 MFLOPS. This system was a 64-bit system and had its own OS, assembler and used a FORTRAN compiler. |
1982 | The Cray X-MP was unveiled. This machine was designed by Steve Chen and used a shared-memory parallel vector. Its clock speed was 105 MHz or 9.5 nanoseconds. This was the first multi processor supercomputer. |
1985 | Cray-2 was born. This machine exceeded the MFLOPS factor and touched GFLOPS (1000 MFLOPS) with a performance figure of 1.9 GFLOPS. It had 4-8 processors in a completely new design and structure, with pipelining and a high memory latency. |
1990 | The Fujitsu Numerical Wind Tunnel was created. It had a vector parallel architecture and its sustained performance factor was 100 GFLOPS, with a clock cycle time of 9.5 nanoseconds. It had 166 vector processors, each with a speed of 1.7 GFLOPS. |
1996 | HITACHI SR2201 used a distributed memory parallel system to attain a performance of 600 GFLOPS from 2048 processors. |
1997 | Intel and Sandia Labs jointly created the ASCI RED. This mesh-based machine was designed for extremely large parallel processing and had 9298 Pentium II processors. Its performance touched 1.34 TFLOPS, making it the first supercomputer to do so and it remained the king of its kind, till the year 2000. It was also a very scalable supercomputer, with its processors found in most home computer systems. |
2004 | The Earth Simulator was designed to simulate the world's climatic conditions, on both land and sea, as well as atmospheric. It was built by NEC and had 8 vector processors. Its performance factor was 131 TFLOPS. |
2005 | The first machine from the IBM Blue Gene supercomputer series, was the Blue Gene/L. This machine started out with a peak performance of 280 TFLOPS. There are 4 main Blue Gene projects and 27 supercomputers using the architecture, which uses approx 60,000 processors. |
2008 | The IBM Roadrunner is a hybrid supercomputer, with 2 different processor architectures working in tandem. It uses Red Hat Enterprise Linux and Fedora as its OS and its performance is 1.456 petaflops at peak. |
2010 | Tianhe-I was a record breaker in so many ways. It was China's first supercomputer to enter the Top500 list of supercomputers. It has a performance factor of 2.566 PFLOPS, which made it the fastest supercomputer till 2011. |
2011 | The reigning champion amongst supercomputers is the K computer, a Japanese supercomputer, which touches performance rates of 8.162 PFLOPS. It uses 68,544 8-core processors and its construction is still being completed. |