Homemade computer calculates pi to 10 trillion digits(npr.org)
npr.org
Homemade computer calculates pi to 10 trillion digits
http://www.npr.org/2011/10/23/141629745/homemade-computer-sets-records-in-the-trillions
2 comments
More information about the setup: http://www.numberworld.org/misc_runs/pi-5t/details.html
Could anyone jump in and provide some information on how calculating pi to a gazillion digits is worth the effort?
It's entirely for supercomputer bragging rights. Putting up a new record for pi gets attention by virtue of how much computational time it would take.
To think, if he had used this for bitcoin mining instead, he could be a hundredaire.
To think, if he had used this for bitcoin mining instead, he could be a hundredaire.
It is about pushing limits with a home computer. The processor itself was a 12 core Xeon processor, but all of the storage required - both RAM and disk is quite impressive.
I am working on a similar brute-force analysis, but I have access to a datacenter with many machines. Not so impressive.
I am working on a similar brute-force analysis, but I have access to a datacenter with many machines. Not so impressive.
I never considered the details of Pi digit calculation, but I was surprised to see how it's very dependent on IO:
"For large computations, the bottleneck is either memory bandwidth or disk bandwidth depending on the number of hard drives and their configuration."
"For large computations, the bottleneck is either memory bandwidth or disk bandwidth depending on the number of hard drives and their configuration."
You could say same about any benchmark. The benchmark itself is obviously not the end-goal.
Indeed. Given that you with pi to 39 decimal places you can calculate the circumference of the universe the the accuracy of the size of a hydrogen atom [1] one does have to wonder.
Calculating pi is a fairly reasonable way of measuring CPU performance though. I've seen Anandtech and the like use SuperPi for this purpose.
In my first year of university we had to do (in Pascal) a library of arbitrary-precision functions for the arithmetic operations. It was really interesting at the time. If you measure something to a trillion digits you need a pretty good library for this. You also may not have the entire number of memory at a time (although a trillion digits is 0.5TB BCD-encoded so that's not inconceivable) so the problem itself has a lot of interesting aspects to it.
[1]: http://www.physicsforums.com/showthread.php?t=350023
Calculating pi is a fairly reasonable way of measuring CPU performance though. I've seen Anandtech and the like use SuperPi for this purpose.
In my first year of university we had to do (in Pascal) a library of arbitrary-precision functions for the arithmetic operations. It was really interesting at the time. If you measure something to a trillion digits you need a pretty good library for this. You also may not have the entire number of memory at a time (although a trillion digits is 0.5TB BCD-encoded so that's not inconceivable) so the problem itself has a lot of interesting aspects to it.
[1]: http://www.physicsforums.com/showthread.php?t=350023
There's a theory that Pi digits contain every possible finite sequence of digits.
Just think about it. If that's true, the whole universe is described inside Pi with any precision you want, with any time precision you want. Somewhere in Pi your whole life is stored.
Just think about it. If that's true, the whole universe is described inside Pi with any precision you want, with any time precision you want. Somewhere in Pi your whole life is stored.
Also known as a Normal Number, it is thought that most numbers are of this type. (http://en.wikipedia.org/wiki/Normal_number)
You seem to have misread the link you have given. It is, in fact, known that almost all numbers are normal.