IBM, SUN announcing new petaflop supercomputers

IBM, Sun announcing petaflop class supercomputers

IBM and Sun Microsystems are providing details of their separate supercomputer offerings at the 2007 International Supercomputer Conference in Dresden.

Sun microsystems petaflop machine will be called “the Constellation” From development information in 2005, the new Sun architecture system uses Sun Fire x86, 64-bit (now called x64 within the industry) servers with 10,480 AMD Opteron processor cores, totaling more than 50 trillion floating point operations per second (teraFLOPS). The computer also includes Sun and NEC storage technologies and NEC’s integration expertise as well as ClearSpeed’s Advance accelerator boards.

Sun’s Constellation promises to deliver nearly 2 petaflops of performance.

It features 82 SunFire blade servers, two Sun Magnum ultra-dense switches, an Infiniband host interface (with 288 ports), next-generation Mellanox HCA (high-contrast addressing) and a Sun Fire X4500 storage cluster with 480TB per rack.

The core switch supports up to 3,456 nodes, and each custom rack supports 48 server modules, chief architect Andy Bechtolsheim said.

The Constellation also features Solaris, Linux, OpenMPI, Open InfiniBand interfaces and management, x64 Computing Architecture, and InfiniBand DDR interconnect. Its compute speed is estimated at 1.7 petaflops, and it will store up to 10 petabytes of data, Bechtolsheim said.

The heart of the Sun constellation is the switch

Code-named Magnum, the switch comes with 3,456 ports, a larger-than-normal number that frees up data pathways inside these powerful computers. “We are looking at a factor-of-three improvement over the current best system at an equal number of nodes,” said Andy Bechtolsheim, chief architect and senior vice president of the systems group at Sun.

The Texas Advanced Computing Center (TACC) at the University of Texas is currently preparing a Constellation system. If TACC can get enough Barcelona chips from Advanced Microsystems by October 15, its system will land near the top of the next Top 500 Supercomputers list, Sun says. The TACC system will provide a peak performance of around 500 teraflops, or 500 trillion operations a second.

The Sun system will cost about $59 million, while the IBM supercomputer runs between $1.3 million and $1.7 million for each server in the system cluster.

IBM will introduce the Blue Gene/P system—that will eventually replace the L system, said Herb Shultz, a product marketing manager for IBM’s Deep Computing division.

The Blue Gene/P system looks to offer three times the computing power of IBM’s previous Blue Gene supercomputer. The system now offers a scale ranging from 1 petaflop to 3.5 petaflops when fully configured with 256 server racks.

IBM will use its own Power Architecture with the Blue Gene/P system. Each Blue Gene chip will use four PowerPC 450 processing cores. The chip offers top clock speed of 850MHz and can perform 13.6 billion calculations per second. The current crop of Blue Gene chips are dual-core chips with a clock speed of 700MHz.

The older and new Blue Gene chips use the same thermal envelope, and the newer supercomputer offers greater performance while using about 20 percent more power.

The new Blue Gene chips also offer more memory and SMP (symmetric multiprocessing), which is designed to support multithreaded software applications. The new supercomputer also offers a new interface, which will make writing applications for the system easier for developers. (The supercomputer’s operating system is based on Linux.)

A typical Blue Gene/P system board will hold 32 microprocessors, and the average 6-foot rack server will hold 32 of these boards, which gives the system more than 4,000 processing cores per server rack.

A 72-rack Blue Gene/P system with 294,912 processing cores will achieve the 1 petaflop of computing performance, Shultz said. A 216-rack cluster offers 3 petaflops of performance.

Cnet has some more info on the IBM Blue Gene/p

The most expensive part of a supercomputer is the memory

FURTHER READING:
Nvidia’s deskside supercomputer for teraflops of power for a few thousand dollars Nvidia will be scaling up this offering and significantly improving it for later this year and next year with double precision floating point.

More on the Nvidia teraflop offerings and potential impact

What it would take for zettaflop computing

China is building three petaflop computers by 2010

Japan is building a 10 petaflop machine by 2011

Other petaflop projects including the completed MDGrape3

Flash memory improving faster than Moore’s law will accelerate larger database searches

Other things going faster than Moore’s Law
which includes gene sequencing costs

system integration