Top 500 Supercomputers Ranked

Shadow Wrought writes "The Register is reporting on (alternate ZDNet article) the latest list of the top 500 supercomputers in the world. Top of the list is the Earth Simulator Center in Yokohama, Japan, with a benchmark performance of 35.86 Tflop/s. HP and IBM claim 159 and 158 of the systems respectively. I wonder how many teraflops Deep Thought could have done?"

Comments

The Top 10

[Anonymous Coward]

The site may rank Supercomputers, but obviously doesn't run on one :) It's already chugging. Here's a straight, unformatted, copy and paste of the top 10:

1 NEC
Earth-Simulator/ 5120 35860.00
40960.00 Earth Simulator Center
Japan/2002
2 Hewlett-Packard
ASCI Q - AlphaServer SC ES45/1.25 GHz/ 8192 13880.00
20480.00 Los Alamos National Laboratory
USA/2002
3 Linux Networx
MCR Linux Cluster Xeon 2.4 GHz - Quadrics/ 2304 7634.00
11060.00 Lawrence Livermore National Laboratory
USA/2002
4 IBM
ASCI White, SP Power3 375 MHz/ 8192 7304.00
12288.00 Lawrence Livermore National Laboratory
USA/2000
5 IBM
SP Power3 375 MHz 16 way/ 6656 7304.00
9984.00 NERSC/LBNL
USA/2002
6 IBM
xSeries Cluster Xeon 2.4 GHz - Quadrics/ 1920 6586.00
9216.00 Lawrence Livermore National Laboratory
USA/2003
7 Fujitsu
PRIMEPOWER HPC2500 (1.3 GHz)/ 2304 5406.00
11980.00 National Aerospace Laboratory of Japan
Japan/2002
8 Hewlett-Packard
rx2600 Itanium2 1 GHz Cluster - Quadrics/ 1540 4881.00
6160.00 Pacific Northwest National Laboratory
USA/2003
9 Hewlett-Packard
AlphaServer SC ES45/1 GHz/ 3016 4463.00
6032.00 Pittsburgh Supercomputing Center
USA/2001
10 Hewlett-Packard
AlphaServer SC ES45/1 GHz/ 2560 3980.00
5120.00 Commissariat a l'Energie Atomique (CEA)
France/2001

Re:The Top 10

[Aardpig]

From this list, would I be correct in thinking that no Microsoft products (i.e., operating systems) run on these ultra high-end machines? Or, to paraphrase what I really mean (and in the interests of honesty), does Windows suck like a Dyson when it comes to High-Performance Computing?

Re:The Top 10

[TheViffer]

Microsoft PR

Because of their ground-breaking work with Velocity, CTC was recognized in Washington by ComputerWorld and the Smithsonian American History Museum and was made part of the Smithsonianâ(TM)s permanent research collection. Velocity was also named to the list of the top 500 most powerful computers in the world. This was a watershed event since it was the first Windows 2000-based system to obtain this ranking and one of only two Windows-based systems to place on the list.

So in the list thier is another.

Re:Deepthought

[Anonymous Coward]

WTF are you talking about? Deep Thought wasn't a chess computer, idiot. That was Deep Blue.

Re:Here's a thought

[Anonymous Coward]

With the lowest one on the list still weighin in at 245 GFlops, assuming each cycle could process 1 floating point operation (usually not a safe assumption) puts it at a minimum of 123 of those suckers. Start saving those pennies.

Re:We're on the edge

[msgmonkey]

Well it depends, when you're talking about your average desktop computer the quoted FLOPS or MIPS are usually useless because the supporting architecure does n't have bandwidth to supply the processor the data at that speed, they are normally based on data that is in cache memory.

Supercomputers are designed with high bandwidth in mind and thats why in general their FLOPS are taken with less of a pinch of salt.

Re:We're on the edge

[smitty45]

If you don't have any idea of why FLOPs aren't meaningless, then you haven't ever run a program/problem/simulation on these machines able to put the numbers into context for you.

I spent 4 years running dynamic finite element analysis simulations on alot of the kinds of these parallel monsters, and when FLOPs indicate numbers that reflect quite well the length of time it would take for a run to finish, you realize that benchmarks ARE useful, in the right context.

Re:Here's a thought

[coult]

Actually the G5 has dual floating point units, which means it can do 2 double-precision floating point ops per cycle (maximum)...and technically speaking the altivec can do 4 single-precision floating point ops per second, which would allow for a theoretical peak of 16 GFlops per machine. Of course the actual performance would be lower, especially in a cluster. I don't think you'd need 123 machines though.

Re:Deepthought

[lostchicken]

For those who don't get it, chess computers don't really deal with floating point math, but are dependent upon integer calculations. So their FLOPS scores are very low while their MIPS scores are outstanding.