Just wondering about the ‘worlds fastest computer’. Exactly who owns one or two and what are they used for? Can anyone own one? What are the price tags to it.
Me I am saving up for one. Imagine the opponents reaction when I wheel that sweet baby in a Quake Lan Tourney
Hehehe
The NSA and NRO own a ton of Cray supercomputers. I just saw a segment on CBS’ 60 Minutes that had a tour through the NRO, there were a ton of Cray faceplates visible. The only Cray I ever saw firsthand was in use by a Hollywood effects house that did graphics for movies like “The Last Starfighter.”
Older Crays are coming onto the collectors market, I’ve spoken to a few collectors online. Many of them are just nonfunctioning trophy pieces, but surely some people are getting them working. The problem is that the old Crays are a bitch to keep running, due primarily to their unique cooling systems.
Well at this point IIRC Cray is not the “Worlds fastest computer”. I am sure someone will be along to correct me before too long but I believe that one of the IBM ASCI series (ASCI White?) which is (I think) essentially a tweaked, multi-node parallel network cluster (like the one that beat Kasparov but a lot more powerful) currently holds the WFC distinction by a pretty good margin.
Here is Cray’s website.
IBM’s super computers -
http://www-1.ibm.com/servers/eserver/pseries/news/pressreleases/2001/aug/asci.html
http://www-1.ibm.com/servers/eserver/pseries/
“LIVERMORE, CA, August 15, 2001—The U.S. Government today dedicated the world’s fastest supercomputer, an IBM system at the Lawrence Livermore National Laboratory that will bring the nation an important step closer to the goal of simulating a nuclear detonation inside a computer. The IBM machine, known as ASCI White, is capable of 12.3 trillion calculations per second, more than the combined speed of the next three most powerful supercomputers on earth.”
etc etc etc
This is the monitor I want!
“The simulations are expected to be viewed on IBM’s T220, the world’s highest-resolution flat-panel monitor, with 200 pixels per inch and more than 9 million pixels in total on its 22.2-inch screen. The T220’s screen depicts 12 times more detail than current monitors, displaying images with a degree of accuracy not previously possible.”
Crays are, unfortunately, a dying breed. New supercomputers are made almost exclusively by Intel and IBM.
Here’s the Top 500 List for June 2001. IBM and Intel have the first four places. I was surprised to see Hitachi in there as number 5.
Ah, but Crays just look soooo sweet, and you can sit on their power supplies!
I used to work at the national lab in Los Alamos. In the summer of 92, before I got my clearance, I was taken on a special tour of some of the computer rooms, in one of which I saw four Cray YMPs. I laughed years later when, reading the novel Jurassic Park, author Crichton expected me to be impressed by the power of just three Cray XMPs.
The Cray-1 (serial number: 1) was, when last I saw it, still in the lobby of one of the Lab’s buildings, being used as a bench. The old girl just drew too much power for the computations it performed, so there was no point in keeping it plugged in anymore.
One final note: my dad, who also worked at the Lab, used to tell us stuff about Seymour Cray. As I recall, he said that every time the Lab bought a new Cray, Seymour would come down for the delivery and have a little party, bringing along some godawful skunky beer from his hometown that he just loved, expecting everyone to partake. hehe.
Can just anyone buy a Cray? I don’t know about any restrictions on purchasing the old C-shaped towers, but be warned: you’ll need LOTS of electrical power and a reinforced floor, wherever you put it.
You should’ve been in the market a year ago.
I know it’s apples and oranges to a degree but just how does a YMP’s number crunching ability conpare with say a Dell 2 GHz P4 CPU?
It’s totally apples and oranges. Crays were primarily vector processing units, the olden days’ version of massively parallel processing. No comparison is valid.
Well, Cray’s history page tells us that a Cray YMP performs at a sustained rate of 2.3 gigaflops. I can’t find hard numbers on a Pentium 4 on anything but a single Spanish-language page, which lists the P4-1.4 GHZ as 362.4 megaflops. The PowerPC G4 processor is listed elsewhere as having a theoretical maximum output of 5.3 gigaflops. So, if I were you, I’d stick with the modern stuff…
Chas.E is correct…it is apples and oranges. Still, comparisons have been made.
Also, note that ASCI White is just one supercomputer in a chain that have been asked for by the US government. By 2005 they are hoping to have a computer doing 100 trillion calculations per second blowing away ASCI White’s ‘mere’ 12 trillion per second.
As PC hardware geek I have little if any true familarity with Crays or vector processing architectures. I’m not disagreeing with you and I understand that they are different hardware and OS architectures etc, I’m just curious as to why no comparison at all can be made with respect to say the ability to perform some basic yet repetitive massive calculation matrix or some similar operation.
If we have a Cray YMP in one corner and a 2.0 GHz P4 DELL in the other and you input a huge matrix calculation into both machines and say “Go” why is this a completely meaningless metric in measuring number crunching horsepower for the machines at hand? Just curious.
The “godawful skunky beer” that you’re referring to is probably Leinenkugels, which was also made in Chippewa Falls, WI. It’s damn good, too.
If your looking for a super computer on a budget, try doing a google search on Beowulf clusters. You can network a bunch of old pentiums together han havve yourself a supercomputer. Won’t help you with Quake, though.
Because Crays usually have massive front ends just to feed the data, you’d have to include the data loading in the measurements too. And array processors just don’t process data the same way regular CPUs do. I used to work on an old Data General array processor, you could load up a huge matrix, like a greyscale image and do a matrix multiply in one CPU cycle. A standard CPU would take many cycles to process each bit of the matrix.