I have computed perft(17)

Daniel Shawul · Post by **Daniel Shawul** » Wed Apr 01, 2015 5:23 pm

I have computed perft(17) on a few million core-hours access I had to a supercomputer at Aragonne labs. I used close to 700000 cores for computation so the computational power is really big. Those guys will not allow you to waste their resources without purpose and benefit to the society, so I had to write a proposal and take training to get that done. Anyway I am not sure if everything went well as intended, so I can not be sure if the result will hold when I do the verification later. Without further adieu, here is the result I got for perft(17) that is not verified yet...drum rolls please...
2172314159265358979323846

vittyvirus · Post by **vittyvirus** » Wed Apr 01, 2015 5:46 pm

Daniel Shawul wrote:I have computed perft(17) on a few million core-hours access I had to a supercomputer at Aragonne labs. I used close to 700000 cores for computation so the computational power is really big. Those guys will not allow you to waste their resources without purpose and benefit to the society, so I had to write a proposal and take training to get that done. Anyway I am not sure if everything went well as intended, so I can not be sure if the result will hold when I do the verification later. Without further adieu, here is the result I got for perft(17) that is not verified yet...drum rolls please...
2172314159265358979323846

What program did you use? And by the way, your calculation suggests perft(17) had a branching factor of almost 27, so it seems somewhat correct.

Daniel Shawul · Post by **Daniel Shawul** » Wed Apr 01, 2015 6:06 pm

I used an optimized version of Perft from my engine Nebiyu. Yes it looks right to me but can not be sure until I run it again.

Arpad Rusz · Post by **Arpad Rusz** » Wed Apr 01, 2015 6:21 pm

Daniel Shawul wrote:I have computed perft(17) on a few million core-hours access I had to a supercomputer at Aragonne labs. I used close to 700000 cores for computation so the computational power is really big. Those guys will not allow you to waste their resources without purpose and benefit to the society, so I had to write a proposal and take training to get that done. Anyway I am not sure if everything went well as intended, so I can not be sure if the result will hold when I do the verification later. Without further adieu, here is the result I got for perft(17) that is not verified yet...drum rolls please...
2172314159265358979323846

Pi = 3,14159265358979323846...

Daniel Shawul · Post by **Daniel Shawul** » Wed Apr 01, 2015 6:27 pm

Yes it is like computing pi

Desperado · Post by **Desperado** » Wed Apr 01, 2015 7:25 pm

Cool ! I would repeat the test exactly in one year. Guess that is the best chance for verification

vittyvirus · Post by **vittyvirus** » Wed Apr 01, 2015 7:25 pm

Daniel Shawul wrote:Yes it is like computing pi

I don't think so. Computation of pi is more disk and memory bounded, not CPU-power bounded. Besides, it's difficult to do it in parallel (e.g. with GPU) because of carry propagation.

vittyvirus · Post by **vittyvirus** » Wed Apr 01, 2015 7:27 pm

Arpad Rusz wrote:
Daniel Shawul wrote:I have computed perft(17) on a few million core-hours access I had to a supercomputer at Aragonne labs. I used close to 700000 cores for computation so the computational power is really big. Those guys will not allow you to waste their resources without purpose and benefit to the society, so I had to write a proposal and take training to get that done. Anyway I am not sure if everything went well as intended, so I can not be sure if the result will hold when I do the verification later. Without further adieu, here is the result I got for perft(17) that is not verified yet...drum rolls please...
2172314159265358979323846
Pi = 3,14159265358979323846...

Well, if it's not already getting off topic, I've an good approximation for pi:

Code: Select all

         426880 * Sqrt&#40;10005&#41;
pi ~= -------------------------
                 13591409

Daniel Shawul · Post by **Daniel Shawul** » Wed Apr 01, 2015 8:08 pm

Yes I have seen your efforts to compute pi in chess-thinkers-forum
http://talkchess.com/forum/viewtopic.ph ... pproximate

I think Arpad is just poking fun at computing perft on a supercomputer has
similar worth as computing pi on supercomputer. I don't agree with him.

zullil · Post by **zullil** » Wed Apr 01, 2015 8:42 pm

vittyvirus wrote:
Daniel Shawul wrote:I have computed perft(17) on a few million core-hours access I had to a supercomputer at Aragonne labs. I used close to 700000 cores for computation so the computational power is really big. Those guys will not allow you to waste their resources without purpose and benefit to the society, so I had to write a proposal and take training to get that done. Anyway I am not sure if everything went well as intended, so I can not be sure if the result will hold when I do the verification later. Without further adieu, here is the result I got for perft(17) that is not verified yet...drum rolls please...
2172314159265358979323846
What program did you use? And by the way, your calculation suggests perft(17) had a branching factor of almost 27, so it seems somewhat correct.

http://en.wikipedia.org/wiki/April_Fools%27_Day

I have computed perft(17)

I have computed perft(17)

Re: I have computed perft(17)

Re: I have computed perft(17)

Re: I have computed perft(17)

Re: I have computed perft(17)

Re: I have computed perft(17)

Re: I have computed perft(17)

Re: I have computed perft(17)

Re: I have computed perft(17)

Re: I have computed perft(17)