HPC at FSU

Photo by P Van Der Mark

• General access and Owner-based components
• Yearly upgrades since inception
• Started August 2007

The FSU HPC consists of multiple login nodes, compute nodes, high-performance storage, and InfiniBand / IP communication fabrics. For more information on the current HPC system click here.

Click here to download a brief description of FSU's shared-HPC and sustainability model. This paper came out of a workshop sponsored by the National Science Foundation on the subject of "Sustainable Funding and Business Models for High Performance Computing Centers".

Eclipse at FSU

• 512 Power4 Processors
• 297 Gigabytes of Memory
• 2,253 GigaFLOPS Peak Performance
• Combined cost of Teragold and Eclipse, $8 million dollars
• Installed Summer 2002, Decommissioned Summer 2007

In 2002, "Eclipse", an IBM eServer p690 was installed and integrated with the existing "Teragold" supercomputer. This combined system was ranked 34th on the TOP500 list of supercomputers in June, 2002.

Teragold at FSU

• 168 Power3 Processors
• 375 Mhz, 1/2 Gbyte of Memory each
• 2 interactive nodes, each with four processors
• 3 Tbytes of disk space
• 40 Tbytes of archival storage
• Installed July 2000, Decommissioned Summer 2007

In 2000, an IBM RS/6000 SP system, "Teragold", was installed at FSU. This system was later combined with the "Eclipse" system in 2002.

Origin 2000 at FSU

• 18 Processors (250MHz MIPS R10000 CPUs)
• 4.5 Gigabytes Memory
• 9 GigaFLOPS Peak Speed
• Installed December 1998, Decommissioned July 2000

A Silicon Graphics Inc., Origin 2000 with 18 processors (250MHz MIPS R10000 CPUs), was installed in December 1998. Interconnecting the Origin 2000’s processors were a hypercube of Cray Links, each of which is a bidirectional 1.6 GB/second pathway, enabling our 18 processor/4 cray link Origin 2000 to move data to the tune of 6.4 Gigabytes per second.

QCDSP at Columbia University

• 1,024 Processors
• Installed 1997, Decommissioned 1998

In 1997, SCRI Scientists Robert Edwards and Tony Kennedy, in collaboration with physicists from Columbia University, Ohio State University, Fermilab, and Trinity College, Dublin, designed an inexpensive Teraflop scale massively parallel supercomputer - QCDSP (Quantum Chromodynamics on Digital Signal Processors) - suitable for work in Lattice QCD. Three computers were constructed, one of which, a 1024 node 50 gigaflop version, was installed at FSU.

The cost of the this 50 gigaflop computer was $175,000, making it affordable to small research groups. The machine had 3 times the sustained speed of the other supercomputer at FSU (the CM-2), and about a factor of 20 beyond anything else. The processors were widely used Texas Instruments Digital Signal Processors running at 50 MHz.

Challenge XLS at FSU

• 2 SGI Challenge XLS Systems Combined
• 18 Processors (10 in 1st system, 8 in 2nd system)
• 4,096 Megabytes of Central Memory (2,048 Megabytes in each system)
• 6.5 GigaFLOPS Peak Speed (3,600 MegaFLOPS on 1st system, 2,900 MegaFLOPS on 2nd system)
• Installed July 1995, Decommissioned December 1998

FSU purchased two Silicon Graphics Power Challenge XLSs in 1995. The first of the machines had 10 MIPS R1000 RISC processors, 2,048 Megabytes central memory with a peak speed of 3,600 Megaflops. The second machine had 8 processors, with 2,048 Megabytes central memory, and a peak speed of 2,900 MegaFLOPS. The Challenge was a Scalable Shared-Memory Processor (SSMP) system with the advantages of both shared-memory systems and distributed computing systems.

Cluster Computing at FSU

FSU played a large and important role in the creation of what is now known as “cluster computing”; the term in fact originated at FSU as part of the language used to describe the developing computing environment in the early 1990s. For example, in 1990-1991 GigaFLOP levels of sustained performance were almost routinely achieved in many areas of computational science on both vector supercomputers and massively parallel machines. The long anticipated ‘Cray on a desktop’ was really available, in the latest generation of high performance RISC workstations.

A similar technological breakthrough was on the horizon as new higher speed network interconnects were developed and installed. Distributed computing moved beyond the confines of local area networks into a truly global scale. The Department of Scientific Computing currently has over 1000 available clustered processors, including desktop workstations, for well over 750 GigaFLOPS of throughput.

Stock CM-2 Photo

• 65,536 one-bit Processors
• 2 Gigabytes of Memory (256 Kbits per processor)
• 5 GigaFLOPS Peak Speed
• 10 Gigabyte DataVault
• Installed February 1990

In 1990, SCRI also installed the CM-2 Connection Machine from Thinking Machines Corporation. This massively parallel computer had 65,536 processors and 2 gigabytes of memory. Problems, which require massively parallel architecture, finally achieved the gigaflop computational rates that are realistically often a prerequisite for success. The classic example is lattice guage theory, which ran almost continuously on the CM-2 for a period of approximately 9 years at 5 gigaflops sustained speed.

Cray Y-MP at FSU

Photo by R. Holden

• 4 Processors
• 256 Megabytes (32 Megawords) Central Memory
• 1300 MegaFLOPS (1.3 GigaFLOPS) Peak Speed
• Installed April 1990, Decommissioned October 1995

In 1990, FSU took delivery of a Cray Y-MP/432 supercomputer. In 1988, Cray Research introduced the Cray Y-MPR, the world's first supercomputer to sustain over 1 Gigaflop on many applications.

ETA10-G at FSU

Photo by R. Holden

• 4 Processors
• 32 Megabytes (4 Megawords) Local Memory per Processor
• 5000 MegaFLOPS Peak Speed
• Cooled by liquid nitrogen
• The only ETA10-G ever in operation; at the time, the fastest computer in the world
• Installed April 1989, Decommissioned April 1990

The last of four ETA Systems was delivered to FSU in April of 1989. It was the only known ETA10-G ever in operation. At the time (1989), it was the fastest supercomputer in the world, outperforming the Cray-2 at Lawrence Livermore National Laboratory (record-holder since 1985).

ETA10-Q at FSU

Photo by R. Holden

• 2 Processors
• 32 Megabytes (4 Megawords) Local Memory per Processor
• 800 MegaFLOPS Peak Speed
• Air-cooled
• The biggest "piper" ever built
• Installed February 1989, Decommissioned November 1991

The second of four ETA Systems Class VII supercomputers was delivered to FSU in February of 1989. This system was air-cooled and the biggest "Piper" system ever built.

ETA10-E at FSU

Photo by J. Vagi

• 4 Processors
• 32 Megabytes (4 Megawords) Local Memory per Processor
• 1500 MegaFLOPS Peak Speed
• Cooled by liquid nitrogen
• The first ETA10 in operation
• Installed January 1987, Decommissioned April 1989

In 1987, the world’s first ETA Systems Class VII supercomputer was delivered to FSU as the first of four ETA 10 machines. It came without an operating system, since this was about a year before a reliable operating system was available and the computer could be used on a network. In March 1987, a major milestone was achieved at FSU, when a single user was able to log in to the ETA-10, interactively and successfully load, compile and execute a number of Fortran jobs. Some CFD benchmarks rated the 10.5 nsec machine at 11 times the performance of a single-processor CRAY X-MP when 32-bit working was used on the ETA-10.

Cyber 205 at FSU

Photo by J. Vagi

• 1 Processor
• 2 Vector Pipelines
• 32 Megabytes Central Memory
• 0.2 GigaFLOPS Peak Speed (200 MegaFLOPS)
• Installed March 1985, Decommissioned October 1989

The CYBER 205 was FSU’s first supercomputer. It was manufactured by Control Data Corporation, and served the university from March 1985 until October 1989.