# Boffin calculates pi to 2.7 trillion digits

Computer scientist Fabrice Bellard says he's calculated pi to a whopping 2.7 trillion digits - a Herculean task which took a humble desktop PC 131 days. According to the BBC, the previous record of 2.6 trillion digits was held by Daisuke Takahashi of Japan's University of Tsukuba. That number-crunching exercise took just 29 …

## COMMENTS

This topic is closed for new posts.

#### Yawn

surely there are more fun things to do with an i7 and 6 GB of RAM along with all that disk space?

#### Humble Desktop?

The spec was...

Core i7 CPU at 2.93 GHz

6 GB of RAM

7.5 TB of disk storage using five 1.5 TB hard disks (Seagate Barracuda 7200.11 model)

That's not a humble desktop PC is it now! Thats a pretty impressive desktop PC.

A humble one would be the crappy PIII with 256MB of ram and a 40GB hard drive (like the one i am typing this on).

#### I disagree

A number that requires over a Terabyte just to store is completely useless, I mean what would you use it for ? How much paper would be required to print it out ? Can kids use it in their homework? assuming the national curriculum has room for something esoteric like geometry, now that we spend so much time testing and indoctrinating....

How can we check he got it right ? I mean right to the last digit, He might have just done the first few thousand and then stuck any old random stuff on the end ....

Ohhh, and raid 0, just some disks then...Doesn't red-hat 'n' fedupra use LVM anyway ?

#### Of course the numbers ar ecorrect

>How can we check he got it right ? I mean right to the last digit, He might have just done the first few thousand and then stuck any old random stuff on the end ....

Simply by checking the code he's going to release, and by analysing it and seeing it is indeed equivalent to one of the many ways of computing Pi mathematically.

And of course, you can also match it against the 2.7 trillions decimals of the previous record-holder, and since the record will always be pushed further, it would be simply stupid to lie, since then your reputation would be destroyed once the next record comes in and the discrepancies force a check to see who was wrong, the previous or the new record holder.

So, to sum it up, no, he CAN'T make up the decimals without being spotted as a fool and cheater by scientists very quickly.

#### re Of course the numbers ar ecorrect

Greg 10, lighten up, get laid, or do something less uninteresting instead

#### Wow...

...must get me this software....

Oh hold on a life is calling me......

#### Oh well that'll come in handy

I don't understand why people do stuff like this.

It's not interesting or useful at all. Who the frig is going to use that number for _anything_ in reality? Seriously if anyone can explain that then tell me, because I'd be really interested in understanding the mentality of people like this.

#### Normal

It's about getting closer to proving Pi is a normal number, and at least confirming it is as far as "we've" looked. See: http://en.wikipedia.org/wiki/Normal_number

#### Nah...

Don't worry, go back to your little daily activities, peon. This is done by people with so much more brain power than you that, even if it were explained to you, it would still make no sense.

#### Normal proof?

No chance. 2.7x10^12 is nowhere near enough to even being to prove pi isn't normal, let alone get closer to proving it is. Suppose pi (in decimal) starts recursing 4242424242 from, oh, a googleplex onwards? There's not enough storage anywhere to hold pi to that many digits and even if there was supposing the recursion stops at a googleplexplex and then continues being random after that: those "few" digits would just be a little curiosity.

Infinity is a big number.

And to all those other peope, pi comes up in so many more places than circles you wouldn't believe it.

#### Of course there is a purpose...

Rounder circles. I hate those half-baked approximations they keep churing out. But the question is, can I get an iPhone app for that?

#### Testing

"if your computer isn't operating flawlessly some of those digits will be wrong"

So after 181 days you will be told either there is something wrong or there is nothing wrong. I'm sure we can devise better tests than that. Actually it would probably take less than 181 days to write the tests.

Probably the most likely failure will be out of virtual memory.

#### Talking of testing.

If a use of a very precise pi is to validate that your humungous number-cruncher and it's algorithms are working correctly and nobody's calculated pi to that precision before, it does beg the question:

How the fuck did he check the answer he got?

There can't be much point using his 1.7 trillion digit pi to validate your shiny supercomputer cluster against if it's not been validated now, is there?

#### I'm impressed

This is really, really small beer compared to the supercomputers that are usually used.

Dudes got crazy programming/ algorith skills!!

#### Yes yes yes....

But what we really need to know is does it play Crysis?

#### On the more useful side of things...

Fabrice also founded the FFMPEG and QEMU projects, both of which I use on a regular basis. Thanks, man.

#### hmmmm

Now there's something to use up my spare processor cyles on my VMWare host.

#### RAID 0???

What?

Imagine the 'red faces' if it failed whilst writing the last digit - all gone!

Not only is it a top-spec desktop PC, it's capable of storing 6TB on a RAID5, which is how it should have been configured in my view.

ttfn

#### eh, what?

Hmm... well spotted. Smart guy like him surely couldn't have raid 0'ed it could he... then again...

#### Safety in numbers

"No, no officer, that's not 6TB of encrypted porn, it's Pi to 2.7 trillion places"

"Yes, I'm sure it is sir, now if you'll just come along with me"

#### Useful? Yes, actually...

The impressive thing isn't that he has calculated such a massive number, just that he has written an algorithm that can do it on a high end desktop rather than a supercomputer, which is nothing to be sneezed at.

Commen-tarding on the article though, it contains less info than the beeb article and not even a cursory examination of what kind of algorithm expertise it would need - I expected better from El Reg.

#### I bet

that there will be or already is a CSI episode based on the Pi's power to break the most secure encryption mechanisms.

#### @Supertim - Humble Desktop?

An i7 with 6gb of RAM and a few extra disks thrown in is *not* something special, or top flight - it's a standard build with some extra disk. It is a humble desktop. It's barely any more sophisticated than my £500 Q6600/8gb/HD4850 build, just with a bit more performance and more disk.

You can buy the components for an i7 machine of this ilk for £500 if you look around - give it a power supply, cheap GPU and case and you'll be looking at £600-650, then add your hard drives.

So it's a humble desktop in my book. It's not like he used a dual socket mobo and fibre-channel for the storage!

Steven R

[really needs to get his eSATA DAS RAID 10 box built]

#### Cool!

Never thought about that before, but yeah, this PI generation as computer test is a great idea! Level 5 diagnostic Mr. Spock: have the CPU cores calculate out to like 100 places. Level 1 diagnostic Mr. Spock: have the CPU cores calculate out to like a trillion places. After all, even in the future a proper in-depth diagnostic should retard and shut down the systems to make the ship more vulnerable to attack--after all, what is life, and plot, without risk.

#### @Cool!

It's pretty common to use something like pifast to check stability of an overclock. Or it was when I did it. Or at least I did.

#### Now he needs a super computer

To calculate "why"

#### If he'd done it on a Mainframe

He could have had 24/7/365 availability, scalability and grahamcrackerabiltity!

Mind you, he have

InstanceX: 3.14156 (crash)

InstanceY: 3.142 (process killed)

Instance Z: 3.13 (process killed)

At 10000 numbers per second, he'd get the result in only 8.5 years, or something like the result from many crashing parallel processes. It would be made up of the same 0 to 9 digits at least.

#### That's nothing

A bloke in my local reckons he can lift 3kg with his nob!

#### How to calculate pi?

So I take it that the last 100 billion digits that broke the record are a result of actually recording the output rather than doing extra calculations? Does it take twice as long to 'calculate' 2 trillion places compared to 1 trillion places?

Anyway when will they release it, I want to find my phone number since it's not available within the first 200 million places! :(

http://www.angio.net/pi/bigpi.cgi

#### But...

... if you convert it to letters, does it spell out somewhere "Haha! I'm God, I existed all along!"?

#### Re :But ..

Would you rather have "Haha! I'm God, I never existed !" ?

#### Depends..

..if you are using ASCII or EBCDIC

#### Deep Thought....

Rounded up, it comes to 42

#### Sir

If only he'd reported the output in base 2.

Bi-Pi

No need to get my coat, I'm already wearing it. It's foccin freezing in here.

#### Big number arithmetic has quite a few uses.

Mostly they seem to be connected with cryptography.

Limbering up to crack the Sky digital 2048 PK crypto?

#### @ LuMan

proof of concept please.

#### Pfffffffftttttt....

In school I could spell 'shell oil' on my calculator *

*if you turned it upside down..

#### Ah, the calculator jokes...

I miss 'em, as I've been using dot-matrix graphical calculators such as the TI-83 and TI-89 since high school.

Then again, I can actually type stuff on those calcs as well... :)

#### Pi is for computing lightweights...

In my mind, the ultimate prize in pointless arithmetic is computing Graham's number.

#### I bet Archamedes never thought of that.

Still for most practical day to day calculations using pencil and paper, abacus and in the mind, 3.1412 is generally close enough.

#### Bignums or ReallyBigNums

The article says that bignum systems are "... limited only by the computer system's available memory". Assuming that means virtual memory, that's true of most bignum systems. This guy went beyond that - it would take roughly 1TB of memory to hold the resulting number, and he had "only" 6GB. (And that's without taking into account that calculations at this precision require more than one number of this size.)

Don't tell me he had all that hard disk space as virtual memory, because I will find it hard to believe. To get that sort of performance, he'd almost _have_ to do it himself, rather than relying on the OS to page out the right parts of the number.

Impressive feat. Now let's see it put to some other use, because I'm sure he had one in mind before he started!

This topic is closed for new posts.