Pentium 4 Overclocked to 7.1GHz, Sets World Record

Netmonger writes "This Japanese guy overclocked a Pentium 4 to 7.132GHz!! The system managed to calculate pi to 1 million decimal places in 18.516 seconds, setting the world's record." The article notes that a Pentium 4 had been overclocked faster earlier this year, but at that speed it was not possible for the machine to function beyond BIOS. Of course, they'd yet to try diverting power from the dilthium crystal reactor to the deflector array.

Re:World record?

[Anonymous Coward]

Yes.

He overclocked the Intel Pentium 4 670 processor with stock speed of 3.80GHz.

The processor system bus was overclocked to 1520MHz.

processor's voltage was pumped up to 1.70V, significantly higher than default setting; memory latency settings were CL4 3-3-4, memory voltage was set to 2.3V.

Still no word on what his 3dmark2005 score was! (CPU and Total, of course)

actual link

[Anonymous Coward]

Link to the actual forum posting, complete with pics.

http://www.xtremesystems.org/forums/showthread.php ?t=70225&page=5&pp=25 [xtremesystems.org]

Re:World record?

[magarity]

Calculating pi is a series of mathematical operations where you can't do the next one without the prior because you need the remainders. Supercomputers are super due to a heck of a lot of CPUs all working on different parts of a problem that can be broken into chunks. How exactly do you break a series of operations that depend on the priors into chunks for a supercomputer to rip through? So anyway, it looks like this calculating pi is a record in general, not for just a PC. It's a speed job for a single CPU.

Comparison

[vorm]

I tried the same test on my 2Ghz P4 Northwood with 768MB RAM. It took 1min 34sec to calculate pi to 1 million digits.

Re:More info and a pic or two

[The Hobo]

Here you go Clicky [xtremesystems.org]

Re:World record?

[keesh]

Not true. You can get any digit in base 16 of pi without needing to know any other digit. This means you can get groups of digits in any base by only calculating a relatively small (maximum 16 * base) number of other digits in base 16 and converting.

Re:World record?

[pooly7]

That's not exactly true, you can compute the Nth binary digit of PI without the need of the previous one. Here is the guy who discover it : http://fabrice.bellard.free.fr/pi/ [bellard.free.fr]

Re:World record?

[wfberg]

Calculating pi is a series of mathematical operations where you can't do the next one without the prior because you need the remainders. Supercomputers are super due to a heck of a lot of CPUs all working on different parts of a problem that can be broken into chunks. How exactly do you break a series of operations that depend on the priors into chunks for a supercomputer to rip through?

Use the BBP Formula [wolfram.com]. Pifast is just a benchark, like all benchmarks it's rather silly. The record is for PCs, the top 500 supercomputers [top500.org] are benchmarked using another silly benchmark (LINPACK).

Now at 7285.1 MHz

[Ixalon]

Someone has already clocked up an extra 150MHz!

Calculating 1m decimal places of Pi now down to 18.093s...

http://www.xtremesystems.org/forums/showthread.php ?p=1001108#post1001108 [xtremesystems.org]

Re:athlon 2400+ using FASTPI 1M places in 4.4 secs

[bhtooefr]

FASTPI != SuperPi...

RTFA, please,

[hummassa]

it's just two paragraphs for $DEITY sake.
Ok, I'll tell you, lazy boy: besides cooling with liquid N2, they tweaked the processor and the memory voltages.

Re:wow...

[bhtooefr]

On a *SERIOUS* note, it's about as fast as a 3.5 or 4GHz Pentium M by my estimates...

Re:athlon 2400+ using FASTPI 1M places in 4.4 secs

[TheLink]

He was using super pi.

Some stats: http://www.pcstats.com/articleview.cfm?articleid=1 815&page=3

http://www.planetamd64.com/lofiversion/index.php/t 5459.html

I'm not sure about fastpi. But pifast appears to allow 2-4GHz computers to do 10 million digits of pi in about 30 seconds.

Re:World record?

[cperciva]

Calculating pi is a series of mathematical operations where you can't do the next one without the prior because you need the remainders.

Leaving aside the BBP algorithm which several other people have mentioned, you're mostly correct here.

How exactly do you break a series of operations that depend on the priors into chunks for a supercomputer to rip through?

But you're going a bit astray here. Large classical computations of Pi are exercises in performing big Fast Fourier Transforms; and there are very good algorithms for doing those in parallel. Using the AGM or a Borwein iteration, computing a million digits of Pi requires approximately 200 full-length FFTs plus some additional linear-time trivially parallelizable work.

So anyway, it looks like this calculating pi is a record in general, not for just a PC.

Give me a 4 processor 3.8GHz Pentium 4 system, and I can beat the reported time by a factor of two. If you can do parallel FFTs, you can do a parallel classical computation of Pi.

Re:World record?

[jmak]

Certainly not even for a single CPU. Performance of SuperPI (using Gauss-Legendre series) sucks by today's standards in comparison to the programs using Chudnovsky elliptic curve algorithm. QuickPI spits out 1 million digits in about 5 seconds on an Athlon-XP.

Re:World record?

[root_42]

Yes, but you still cannot do it in constant time. To compute the nth digit in base 10, you need at least quadratic time. So adding more CPUs to the calculation of Pi does not necessarily scale well.

Re:World record?

[ikkonoishi]

I know you are joking, but try adobe liposuction [theinquirer.net]

It worked for me. I can open a pdf in less than the time it takes a politician to go from idealistic young upstart to corporate whore.

Re:Apple was right!

[Man in Spandex]

Apple calls it liquid cooling when in fact it's a bunch of heatpipes transfering the heat towards the radiator. Nothing too impressive.

There are no reservoirs or tubes or pumps so its not "actually" liquid-cooled hardware unlike what most of us think liquid cooling is which involves the above.

Re:The "dangers" of overclocking.

[mOdQuArK!]

If your instructions take longer to execute than the number of cycles available to them, it just can't do the work and you get junk in your registers.

Your statement is true on its face, but do you really understand why this guy could get his chip to overclock so high? He's not cooling it in LN2 just to keep it from melting (although that is certainly very important).

At low temperatures, typical silicon transistors operate much faster, and wires have less impedance, thus allowing a properly-designed chip to operate correctly at a much higher frequency than it would normally be able to achieve.

It's certainly not useful for a user who wouldn't have a constant source of LN2 available, but the fact that it can be done makes some interesting engineering scenarios possible.

And some luck.

[btarval]

They also must have gotten lucky with the CPU. Back before Intel started adding their stupid locks to limit overclocking, it was painfully obvious that the production run played a factor. And even within the production run, there were always CPU's within any given batch that were better than others.

The last set of great overclocked CPU's were the Celeron 300's. Many of those went to 450-500 MHz with no problem. A very few could be made to hit 600 MHz, though it is questionable on how reliable they were at that point. Certainly reliable enough to calculate the value of PI quickly; but you wouldn't want one for reliable web server.

Granted, some of the one's which could do 450-500 MHz were made for that speed, and then sold as 300's. But certainly not all of them.

The bottom line is that cherry-picking your CPU's helps lead to a better chance of success with overclocking.

World record? LOL

[acidblood]

That may be a SuperPi world record, but definitely not the overall record: Steve Pagliarulo's QPi can compute 1 million digits of Pi in 6.68 seconds [istar.ca] in a Pentium 4 1.6 GHz box. You read that right.

BTW, the same computer takes 189 seconds [istar.ca] to compute 2^20 (~1 million) digits using SuperPi. Among the community [yahoo.com] of Pi-calculating programmers, it's well known that SuperPi is terribly slow. I don't know why overclockers still hang on to it when most programs out there for calculating Pi are faster than it.