What is meant by a “supercomputer”? When I first heard this term in 1986ish. Way back then, it could have meant today’s PCs!!! So, is there a definition, and is the definition based on standards from a certain time frame? Also, who builds these “supercomputers”?
Thanks,
Here is the Department of Commerce view of supercomputers:
http://www.fas.org/spp/starwars/offdocs/940224.htm
I am not aware of any PCs that can individually match a Cray II or a Cray III. (A few years ago a group did set up a series of 6 processors, using one to direct the operations and the other five to perfom simultaneous operations in a true multi-parallel environment and they were able to perform a number of calculations with faster performance than a Cray II. I have not kept track of those efforts, so I don’t know whether anyone has followed up on that.)
One thing to remember about measuring computer speed is that the number of operations is not always a true indicator of performance (hence the acknowledgement that MIPS (intended to mean Millions of Instructions Per Second) actually stands for Meaningless Indicator of Performance Standards).
Beside raw processor speed, another measure of performance is throughput. It is all very well to say that a processor is “faster” than another, but if there are not the appropriate facilities for I/O and temporary storage of large amounts of data, the machine will be constrained by its inability to get out of its own way. (I saw this demonstrated effectively when a manager who hated IBM and loved HP talked a company into exchanging their 2.6 MIPS mainframe for an 8.6 MIPS mini-computer because it was “more than three times as fast” only to see their Friday evening inventory reconciliation job go from a four-hour run to a 78 hour run because the HP (a nice machine where appropriate) simply could not handle the throughput of the larger “slower” machine.)
You might be interested in this list of “top 500” supercomputers:
http://www.netlib.org/benchmark/top500.html
See their section on “Introduction and Objectives”. Their ranking is arrived at by using the “linpack” benchmark, which measures the system’s performance at solving large systems of linear equations. Of course, in any real world scenario, tomndebb’s remarks concerning throughput are well heeded. Most of us don’t use our computers for operations which are strictly compute bound.
At any rate, if you look at “current list” link on that page, you will find out who is making them (IBM is probably the highest number of entries on the list).
You have to go way back before 1986, but my Radio Shack, programmable, pocket calculator had more computational power than the main computer at the place I started working in 1950. What it lacked in comparison was memory space and flexible I/O devices.
Another thing besided I/O functions that makes processor chip speed academic in many cases is bus speed. If all functions can be done on the processor chip, fine, but if the program does a lot of transferring over the bus then things slow down to bus speed. At one time that was [sob]30 mhz[/sob], but I guess it is now up around 100 or a little more.
David, we don’t have to go back all that far. There are people with more horsepower on their desks, today, than an IBM 360/45 could put out a little before 1970. The Cray II will be beaten soon, I suspect–I just don’t think it has happened, yet.
You’re right in general, of course. There was a 1980 ad that claimed that if the auto industry had advanced at the rate of the computer industry, we could by a Rolls Royce for $49 and drive 1,000,000 miles on a gallon of gas. (That ad has been updated in a joke about GM and Microsoft, but the original point is still valid.)
As far as raw CPU processing power only, PCs passed the Cray II a couple years ago. It was a really sorry machine by supercomputer standards. Used Cray II’s hit the used market in no time, and the recycling pile shortly thereafter. PCs will pass the Cray III in about a year.
Of course, Cray’s came with a lot of extra goodies that PCs don’t have. (E.g., many an original Cray I had a classic Mac for a control console!)
One way of defining a supercomputer is bang per buck. If you are willing to spend top dollar for a small gain in performance, that’s a supercomputer. This makes those mesh systems you see now (lots of old Pentuims, even 486’s) non-supercomputers even though they have equivalent power. Mainframes are less costly for the power, minis less so, etc. So even though some minis are more powerful than some mainframes; if you paid more it’s a mainframe.