mersenneforum.org Google Cloud Compute 31.4 Trillion Digits of Pi
 Register FAQ Search Today's Posts Mark Forums Read

 2019-03-14, 16:58 #1 Mysticial     Sep 2016 34410 Posts Google Cloud Compute 31.4 Trillion Digits of Pi Repost a gazillion times. But I had to drop it here. Blogs: Stats:Decimal Digits: 31,415,926,535,897 Hexadecimal Digits: 26,090,362,246,629 Wall Time: 121 days (September 22, 2018 to January 21, 2019) Program: y-cruncher 0.7.6.9486 (17-SKX ~ Kotori AVX512-DQ) Hardware: Google Cloud PlatformPrimary Node: 1 x n1-megamem-96 (96 vCPU, 1.4TB) with 30TB of SSD Storage Nodes: 24 x n1-standard-16 (16 vCPU, 60GB) with 10TB of SSD
2019-03-14, 20:15   #2
retina
Undefined

"The unspeakable one"
Jun 2006
My evil lair

2·23·137 Posts

Quote:
 Originally Posted by Mysticial Google Cloud Blog
What would be the estimated cost of doing this "in the cloud" for someone as an ordinary user?

Last fiddled with by retina on 2019-03-14 at 20:15

2019-03-14, 20:36   #3
Mysticial

Sep 2016

23·43 Posts

Quote:
 Originally Posted by retina What would be the estimated cost of doing this "in the cloud" for someone as an ordinary user?
Emma told me what that number was, but I'm not sure if i can disclose it.
OTOH, some people on HN calculated it to be $170k USD. I'll just say that both figures are very high. And there's a lot of room to optimize it down should someone decide to replicate or beat it in the future via a cloud. It can be done in under$20k, if you custom-built something specifically for such a computation.

2019-03-14, 21:16   #4
retina
Undefined

"The unspeakable one"
Jun 2006
My evil lair

142368 Posts

Quote:
 Originally Posted by Mysticial Emma told me what that number was, but I'm not sure if i can disclose it. OTOH, some people on HN calculated it to be $170k USD. I'll just say that both figures are very high. And there's a lot of room to optimize it down should someone decide to replicate or beat it in the future via a cloud. It can be done in under$20k, if you custom-built something specifically for such a computation.
Yeah, that is pretty much the sort of difference I would have expected. Using the "cloud" is an expensive way to do things.

2019-03-15, 00:03   #5
GP2

Sep 2003

1010000110012 Posts

Quote:
 Originally Posted by retina Yeah, that is pretty much the sort of difference I would have expected. Using the "cloud" is an expensive way to do things.
The whole point of it was advertising. Google Cloud is in third place and this is a cheap (for them) and viral way to remind the world that they exist.

News media are already picking up the story (USA Today, BBC, Washington Post, etc).

Not unlike what happens when we discover a new Mersenne prime and issue a press release. Except the public cares more about pi.

Maybe it's just me, but it seems like kind of a ho-hum result. No new "thing" was discovered.

2019-03-20, 14:15   #6
LaurV
Romulan Interpreter

"name field"
Jun 2011
Thailand

24·613 Posts

Quote:
 Originally Posted by GP2 The whole point
+1.

 2019-03-20, 19:55 #7 CRGreathouse     Aug 2006 3×1,993 Posts Let me say, when the corporations of the world decide that the best way to get advertising is to lend a bit of their big iron toward pure mathematics to get some word-of-mouth for their latest thing, I'm all for it. They could have just as easily put the money into a teenage model, and where's the fun in that? Edit: but next time, let's convince them to chase down zeta zeros, we could use an update. Last fiddled with by CRGreathouse on 2019-03-20 at 19:55
2019-03-20, 21:35   #8
Uncwilly
6809 > 6502

"""""""""""""""""""
Aug 2003
101×103 Posts

100111100010012 Posts

Quote:
 Originally Posted by CRGreathouse Let me say, when the corporations of the world decide that the best way to get advertising is to lend a bit of their big iron toward pure mathematics to get some word-of-mouth for their latest thing, I'm all for it.
How much TF work would that done or LL on 100M digit numbers?

2019-03-21, 01:12   #9
Mysticial

Sep 2016

23·43 Posts

Quote:
 Originally Posted by CRGreathouse Let me say, when the corporations of the world decide that the best way to get advertising is to lend a bit of their big iron toward pure mathematics to get some word-of-mouth for their latest thing, I'm all for it. They could have just as easily put the money into a teenage model, and where's the fun in that? Edit: but next time, let's convince them to chase down zeta zeros, we could use an update.
This isn't the first time that Google has pulled a stunt like this. They did the same thing with SHAttered.

Obviously, they don't attempt something like this unless there's a reasonable probability of success to be worth the resource investment. It's not just about the hardware time, more importantly, it's the human resources of coordinating the whole thing.

Google may have spent 6 figures of GCP time on this Pi computation, but I'm pretty sure they've spent much more than that on employee time - considering the number of levels of management and VPs this had to go through on their side.

So it's not as simple as to just throw a dude's code on the cloud and let it run for half a year. Emma spent a considerable amount of time working out the right configuration to do this. She also had to deal with a lot of people in the company to make it happen. And since Google's reputation is at stake as well, I was told afterwards that they did independent cross-checking with other sources before they could even trust me and the program. All of this carries a high implicit cost.

-----

Pi and SHAttered were easy to see as a high probability of success. y-cruncher has done this 5 times already so it's tried and proven. SHAttered would've been easy to see once they had their algorithm and could draw a probability graph of hitting the collision given X amount of computing resources.

For something like finding the next Mersenne prime, it's a lot more uncertain. Let's put aside the lower amount of publicity generated by something like a new Mersenne prime compared to Pi or SHAttered and look at the practicality of it.

We seem to be hitting a lot more Mersenne primes than we should. And there are notoriously large gaps at the smaller sizes. So it's really uncertain if you would be able to find a prime by throwing a ton of resources at it. Likewise, Google probably doesn't want the public to know about the failed projects. So if they threw a ton of resources into GIMPS, they'd either have to do it in secret (maintaining their own version of the database), or they'd collaborate with the GIMPS database and it would be apparent what's going on including if they fail to find a prime.

One thing that would top both Pi and SHAttered is actually finding a non-trivial Zeta function zero off the critical line. But not a lot of people believe that's likely to happen.

Last fiddled with by Mysticial on 2019-03-21 at 01:14

2019-03-21, 05:04   #10
CRGreathouse

Aug 2006

597910 Posts

Quote:
 Originally Posted by Mysticial This isn't the first time that Google has pulled a stunt like this. They did the same thing with SHAttered.
I wouldn't class SHAttered as a stunt, that's a real security issue near and dear to their heart. Doing the public demo got their point across in a way they couldn't have done with pure marketing.

2019-03-21, 05:06   #11
CRGreathouse

Aug 2006

135338 Posts

Quote:
 Originally Posted by Mysticial We seem to be hitting a lot more Mersenne primes than we should. And there are notoriously large gaps at the smaller sizes. So it's really uncertain if you would be able to find a prime by throwing a ton of resources at it. Likewise, Google probably doesn't want the public to know about the failed projects. So if they threw a ton of resources into GIMPS, they'd either have to do it in secret (maintaining their own version of the database), or they'd collaborate with the GIMPS database and it would be apparent what's going on including if they fail to find a prime.
Of course Amazon did it, so it's not at all unreasonable to suggest Google might (though there are a lot of fish in the sea, I don't think we'll get that lucky).

 Similar Threads Thread Thread Starter Forum Replies Last Post GP2 Cloud Computing 4 2020-08-03 11:21 GP2 Cloud Computing 0 2019-01-12 12:52 pepi37 Software 14 2018-09-08 01:26 kergy47 Cloud Computing 0 2018-05-31 11:35 GP2 Cloud Computing 32 2018-01-23 02:16

All times are UTC. The time now is 12:14.

Mon Dec 6 12:14:47 UTC 2021 up 136 days, 6:43, 0 users, load averages: 0.81, 1.08, 1.27