News: 1654763530

  ARM Give a man a fire and he's warm for a day, but set fire to him and he's warm for the rest of his life (Terry Pratchett, Jingo)

Google calculates Pi to 100 trillion digits

(2022/06/09)


Google has put its cloud to work calculating the value of Pi all the way out to 100 trillion digits, and claimed that's a world record for Pi-crunching.

The ad giant and cloud contender has [1]detailed the feat, revealing that the job ran for 157 days, 23 hours, 31 minutes and 7.651 seconds.

A program called y-cruncher by Alexander J. Yee did the heavy lifting, running on a n2-highmem-128 instance running Debian Linux and employing 128 vCPUs, 864GB of memory, and accessing 100Gbit/sec egress bandwidth. Google created a networked storage cluster, because the n2-highmem-128 maxes out at 257TB of attached storage for a single VM and the job needed at least 554TB of temporary storage.

[2]

32 storage nodes, using the n2-highCPU-16 instance, and a single compute node were assembled into a cluster that offered 64 iSCSI block storage targets.

[3]

[4]

H2 instances run Intel Ice Lake and Cascade Lake CPUs, but Google hasn't said which was used for this job.

Google admitted it did this to show off its cloud, and how fast it's become since last breaking the record for Pi-munching when it went 31.4 trillion digits deep in 2019. The [5]2019 post explaining that effort says it consumed 111.8 days of computing, but used a very different rig.

[6]Wordle recreated in Pascal for the Multics operating system

[7]Pi calculated to '62.8 trillion digits' with a pair of 32-core AMD Epyc chips, 1TB RAM, 510TB disk space

[8]Yahoo! boffin scores pi's two quadrillionth bit

The 2022 Pi run clearly worked faster – which won't be a surprise – although the rigs used in 2019 and 2021 were very different, so an apples-to-apples comparison is tricky.

The Register has estimated the cost of this effort: n2-highmem-128 instances cost $7.706976 an hour at list price, so the cost of the compute server was about $29,000. The n2-highCPU-16 instances cost $0.57 an hour, or about $70,000 for the whole job. Data movement would have added plenty to the task.

[9]

And all to produce a trillion digits of Pi.

Sadly, Google's post doesn't whether or not all those numbers contained any interesting patterns that might hint at the nature of the cosmos, or whether any recreations of Shakespeare appeared.

Figuring that out probably needs more than a single Debian box. ®

Get our [10]Tech Resources



[1] https://cloud.google.com/blog/products/compute/calculating-100-trillion-digits-of-pi-on-google-cloud

[2] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_offprem/front&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2YqHEwiIZ5buB1g-Stjs1OwAAANc&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0

[3] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_offprem/front&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44YqHEwiIZ5buB1g-Stjs1OwAAANc&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0

[4] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_offprem/front&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33YqHEwiIZ5buB1g-Stjs1OwAAANc&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0

[5] https://cloud.google.com/blog/products/compute/calculating-31-4-trillion-digits-of-archimedes-constant-on-google-cloud

[6] https://www.theregister.com/2022/06/07/wordle_multics/

[7] https://www.theregister.com/2021/08/17/pi_world_record_challenged/

[8] https://www.theregister.com/2010/09/16/pi_record_at_yahoo/

[9] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_offprem/front&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44YqHEwiIZ5buB1g-Stjs1OwAAANc&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0

[10] https://whitepapers.theregister.com/



Contrex

Finding any 'pattern' might take a bit longer than the calculation.

errr

charlieboywoof

"Sadly, Google's post doesn't whether or not"???

They'd get a shock...

bofh1961

If they found that the last few trillion digits were all zeroes!

Re: They'd get a shock...

John Robson

I wonder what the longest set of zeros was in those trillion digits...

Re: They'd get a shock...

KittenHuffer

First thought would be that 100 Trillion is 10 15 , so the longest sequence of any particular digit should be around 15 digits in length.

Please correct me if I'm making a dumb error, but I did say it was a first think.

Re: They'd get a shock...

MiguelC

The digits of PI seem, to the naked eye, perfectly random.

If they were effectively random, they might be all the same, all 100 trillion of them (extremely improbable, but possible - just might be one of those random things)

devin3782

Does crunching that much pi help other than making your storage fat?

spot the difference

Pete 2

> Pi-munching when it went 31.4 trillion digits

Has anyone checked that the first 31.4 trillion digits calculated by this latest feat actually matches the 31.4 trillion digits from the previous run?

Re: spot the difference

Tom 7

Pi could easily introduce rounding errors!

Re: spot the difference

Charlie Clark

I'm sure they have but more important is verification of the algorithm used.

They should try something more impressive...

Neurons for Kryton

Why doesn't Google try something more impressive like that tetration monster 'Grahams Number' - wikipedia.org/wiki/Graham's_number

Ron Graham appeared on one of the old BBC Horizon programs on the subject of infinity - highly interesting.

apples-to-apples comparison is tricky

Anonymous Coward

That depends on your definition of Apples. I'd say for this publicity stunt, it's "the fastest setup of cloud VMs Google offer." Which makes the speed comparison perfectly valid.

If they were proper engineers...

Andy 73

..they'd know it was 4, and saved a lot of time and effort.

Re: If they were proper engineers...

Charlie Clark

Bloody Stupid…

Wellyboot

That's an interesting maths question.

Has anyone ever looked at Pi calculations and determined the maximum* single digit repetition length possible?

* not as the final digits such as.333 recurring

Charlie Clark

It's an irrational number so, in theory, there could be series that are infintely long… Actually, more important mathematically than repeating digits are repeating sequences.

Against stupidity the very gods Themselves contend in vain.
-- Friedrich von Schiller, "The Maid of Orleans", III, 6