New top 500 list released.

[drag]

These are the top 500 most powerfull computers in the world.


Out of the top 10, 7 are clusters.

And one funny thing is that Apple OS X has beat out Linux using a 1100 Dual 2.0 GHz Apple G5's.

That Apple cluster is called "X" and can peak at 17600 Gflops of proccessing power.

#1 is of course the gigantic Earth Simulator from NEC of Japan. Used for enviromental modeling it has been the most powerfull computer in the world for a LONG time. Nobody can touch it. Peaks cpu power 40960 Gflops. 35860 rmax Gflops.

System Configuration
The ES is a highly parallel vector supercomputer system of the distributed-memory type, and consisted of 640 processor nodes (PNs) connected by 640x640 single-stage crossbar switches. Each PN is a system with a shared memory, consisting of 8 vector-type arithmetic processors (APs), a 16-GB main memory system (MS), a remote access control unit (RCU), and an I/O processor. The peak performance of each AP is 8Gflops. The ES as a whole thus consists of 5120 APs with 10 TB of main memory and the theoretical performance of 40Tflops

#2 is also a traditional supercomputer like the earth simulator, it's Ascii Q, and is a HP alpha server. 13880 rmax Gflops.

#3 is that G5 cluster. 10280 rmax Gflops

#4 is a Linux cluster using 1750, P4 Xeon 3.06 GHz's, 9819 rmax Gflops.

#5 is also a Linux cluster, called "Mpp2" and uses 1960 Itanium 2 ("Madison") 1.5 GHz's, with 8633 Gflops.

#6 is a Linux cluster, called "Lightning" produced by Linux Networx for Los Alamos research center and uses 2,816 AMD 2.0 GHz Opteron processors to produces some 8051Gflops

#7 is a Linux cluster, called "MCR Linux Cluster" again made by Linux Networx. MCR has 1,152 nodes, each with two 2.4-GHz Pentium 4 Xeon processors and 4 GB of memory, and was the first Linux cluster in the top5 and was #3 in last year's list. 7634 rms Gflops perfomance.

#8 is ASCI white and use 8192 375mhz Power3 proccessors in the form of 16-way SMP nodes. 7304 rmax performance.
All nodes are of IBM's RS/6000 POWER3 symmetric multiprocessor 64-bit architecture running AIX unix.

#9 is another traditional supercomputer and uses 6080 IBM power3 proccessors divided up in 380 16-way SMP nodes, much like ASCI white. Runs IBM's AIX propriatory Unix operating system and produces 7304 rmax Gflops. Ascii white has a much higher "peak" performance.

#10 is another IBM supercomputer, but is a Linux cluster. 960 nodes each with 4Gig of ram for a total 1920 2.4ghz Xeon proccessors. 6586 rmax Gflops.

Good stuff.

FYI

Flops means floating-point operations per second. I believe that they use the Linpack benchmark for measuring performance, so this is more of a relative rating system rather then actual performance potential.
It uses double percision 64bit floating points for it's calculations.

I found a decent linpack that has been ported to C (originally it was written in fortran) for gcc users. here: ftp://ftp.nosc.mil/pub/aburto

I think that link to that linpack is for single cpu stuff, there is a threaded version for clusters aviable thru that top500 site somewere, but I can't find it because right now there servers are maxed out.

My 2400+ (1994.363mhz) AMD box scored (gcc version 3.2.3 20030422 (Gentoo Linux 1.4 3.2.3-r2, propolice))
:
Rolled Double Precision 599.96 Mflops
Unrolled Double Precision 798.89 Mflops
Rolled Single Precision 449.33 Mflops
Unrolled Single Precision 530.64 Mflops

That compares to the earth simulator which has 35860000 or Mflops, or around 40000000 therotetical Mflops.

Roughly 59766 times as powerfull as my PC.

 

[Goosemaster]

And to think that all this crap will just serve as cheezy furniture in the future just as the cray's are doing.

 

[drag]

Originally posted by: Goosemaster
And to think that all this crap will just serve as cheezy furniture in the future just as the cray's are doing.

Crays are still putting out respectable supercomputers. Take the Cray X1 supercomputer.

Phoenix, the Cray X1 at ORNL, currently has 256 multi-streaming vector processors (MSPs). Each MSP has 2 MB of cache, and four MSPs form a node with 16 GB of shared memory. Memory bandwidth is very high, roughly half the cache bandwidth. The interconnect functions as an extension of the memory system, offering each node direct access to memory on other nodes at high bandwidth and low latency.

2932.9 rmax Gflops. or 488 times as powerfull as my fastest computer.

They can be just as powerfull as any other computer, including the Earth Simulator. If anybody would have enough money to go out and by one big enough.

However Linux on PC hardware can do 10x as much at a fraction of the price, for most aplications, however in somethings supercomputers are the only solution. The amount of information that can be proccessed at once using there tightly integrated cpu archatectures with the massive memory bandwidth can't be equalled with no amount of PC's strung together using myranet switches.

As far as Historical Cray's go, The first one was developed in 1976. Called the Cray-1

from here

The Cray-1A weighed 5.5 tons including the Freon refrigeration system. The computer had a 'horse-shoe' cross-section in order to reduce wire lengths within the casing, no wire in the system was more than four feet long. It used vector processors and contained 200,000 specialized ECL circuits. The CRAY-1A had a 12.5-nanosecond clock (80 MHz), 64 vector registers, and 1 million 64-bit words of high-speed memory (8MB of RAM). It could execute over 80 million floating-point operations per second (MFLOPS), later Cray-1s increased this to a world-record speed of 133 MFLOPS.

I suppose it's nice to know that my PC is seven or so times as powerfull as a computer first sold in 1976. The crays-2 sold in the mid-eighties still beat my computer by a factor of 4 or so.

Then again, the Cray-2 WAS liquid-cooled (chilled flourite) and probably cost 30 million dollars or so brand new. with some hardcore chilled cooling I bet a smp Athlon-mp could do the same.

edit: I overclocked my proccessor to 2174mhz and scored a 870 Mflops on the double unrolled fpu test.