Starting a new Distributed Math Project - Started, Need x64 cpus

[Jjoshua2]

I am working on an Honor's project and have started a distributed computing project. I am attempting to find a counter-example to the beal-conjecture. See www.bealconjecture.com. It can run any number of threads efficiently, but needs 64-bit, and uses about 1 gig of ram. Coordinating a small project will guarantee an A.

All that you would have to do is run my program with the range you reserved, and zip the small txt file outputted and attach it or email it.

There has not been much computer search done on this, but two people have written programs before me, a NASA research scientist and a Google engineer, and my program hundreds of times faster than the first guy's who wrote it about ten years ago, and about 4x as fast as the second guy's who wrote it a couple summers ago.

Thanks!

EDIT: The project is really just for this quarter, so a short term like just for one month. Even if you don't want to run the project long, I would be interested in timings from different CPU. I have q6600. Interested in say q9450/9550 with more cache and Penryn upgrades as well as I7 chips w/ DDR3.

Here is a website for the project that I update.

 

[Jjoshua2]

I've started a thread on a math site on which someone volunteered already. http://mersenneforum.org/showthread.php?t=11756 I am merging the results from both groups on the website.

 

[Philippart]

I'd like to help out! I have a Q9450@3.4Ghz running a 64bit OS with 4GB RAM. I have a Phenom 9500 with 6GB Ram on 64bit and a Phenom 8750 with 4GB Ram on 64bit too that I can manage easily.

Edit: does the program checkpoint? if not you need to tell me what range is good for approx 15-16h uptime per day. The Phenom 8750 could run 24h/7

 

[Alyx]

Sadly I don't have a 64-bit windows OS. I could run it if you got a java version going. Interesting project, I read the thread you linked.

 

[videogames101]

I'm running it from 100000-150000, but with crashes whenever i use less than 4 cores, and how long will that take?

 

[Jjoshua2]

Originally posted by: videogames101
I'm running it from 100000-150000, but with crashes whenever i use less than 4 cores, and how long will that take?

I'll release a version that works besides 4 threads soon. One could time a chunk of 100, say 125k-125.1k, and multiply by 500 and it should give an accurate time for your cpu. I put your reservation on the website. It should take about 4 days I think 24/7.

 

[Jjoshua2]

Originally posted by: Alyx
Sadly I don't have a 64-bit windows OS. I could run it if you got a java version going. Interesting project, I read the thread you linked.

I'm thinking about building a 32 bit or mono version, in which case it should work for you.

 

[Jjoshua2]

Originally posted by: Philippart
I'd like to help out! I have a Q9450@3.4Ghz running a 64bit OS with 4GB RAM. I have a Phenom 9500 with 6GB Ram on 64bit and a Phenom 8750 with 4GB Ram on 64bit too that I can manage easily.

Edit: does the program checkpoint? if not you need to tell me what range is good for approx 15-16h uptime per day. The Phenom 8750 could run 24h/7

It has the basics for checkpointing, it would have to be done manually now. The program outputs files with candidates that have passed many filters which can be checked brute force. One can see how far the range got by looking in these files. I plan to automate this in the future.

You could grab 150k-160k and that should take about 15-16 hours. Try timing a chunk of 100 like I suggested in the previous message. The search time is non linear so time around 20k is slower than around 200k.

 

[videogames101]

I'm lost, my files go from 66666 to 150000, very sporadically, i don't know where to resume.

 

[Jjoshua2]

If you upload (or email to my profile addy) the files I can finish them for you, and you can start the next range, until I get my program to resume automagically. Are you sure it didn't finish? Did it crash or you had to reset?

 

[Jjoshua2]

I got a new version out which includes a readme and changelog! Download it from my beal homepage or directly here. It adds checkpointing so videogames101 can finish his range automatically.

In order for it to resume, it must find its files untouched (its *.gp files) and you must have used the same number of threads and start and stop range as the time before. It will then see if the proper gp files exist and read the last line and resume using that data.

 

[Jjoshua2]

I updated to v 1.2, which is a minor update that fixed a bug that probably no one would have had. However, I also updated the website to explain what the equations output represent.

 

[Philippart]

i'd like to do 150k to 250k

 

[Jjoshua2]

Thanks! I added your reservation, as well as an explanation (involves math ) of the output file that someone requested.

 

[Silverthorne]

Seems to crash for me on start up. Win XP Pro x64 Q9550

 

[Jjoshua2]

Does it have a message? Try latest 1.2 entering "bealconjecture.exe 4 1 500". Did you try something very close to that? My guess is it doesn't like the arguments you passed in.

 

[Silverthorne]

It crashes right after opening the program.

 

[Jjoshua2]

It must have output some error message into the cmd or else there isn't anything I can do. If you look underneath the windows error message maybe? Anyway, we're almost done here, so no biggie, maybe your computer isn't stable or for some crazy reason win xp doesn't like it.

How is it going for you other guys, almost done w/ your ranges?

 

[PCTC2]

So what are you going to do with your $100k if you find a counterexample?

 

[Jjoshua2]

Whoops, accidentally sent him a pm instead of a post . I think I must have done that before, because I've had posts not show up before and I wonder what happened. Maybe now I know

I was wondering when that would come up. 20k for faculty at my school, 20k for programming, 20k for me and any other people who help administrate the project, 20k for finder, and 20k for charity. But in all honesty, all the top theoreticians are very sure it is true (no counterexample) and our search is not likely to find one even if there is one. It's not too much expanded over previous searches (maybe 10-100x but that's peanuts). Think about how many numbers there are bigger than 1 million (our max base now) . And the quarter's half over too.

 

[PCTC2]

NVRMND. I IZ MUCHO TIRED.

 

[filibusterman]

Have you gotten the code updated yet to not crash? The only 64-bit rig I have is running windows 7 and it only has a dual core on it.

 

[Jjoshua2]

It doesn't crash on mine, and it didn't crash on the other peoples (expect for one person). I would be interested to know if it works on yours. We finished our first run here, except for the ranges videogamers, and phillipart have. But they seem to be gone.
I suppose it might need .net framework. I could try publishing a setup and see if that works for whoever. I think I recall that fixing a similar problem once.

 

[Philippart]

hey I'm still there, I didn't get a PM???

I had a strange checkpoint issue I think (actually even twice!) because it started from the beginning, I'm still running...

 

[Jjoshua2]

It looks like from the resume file that you resumed twice. It did not log any duplicate work, so I doubt that it did any. (the file has "print(250000);" when it resumed, and it continued right where it left off, not restarting)

I must have just pm'ed videogames, sorry. It looks like all 4 cores finished succesfully.

I am almost done with my last range (3 threads finished) and then we just have videogames range. I'm undecided if we will continue after that with a broader search. I guess its up to you guys if you want to. We have not found a counterexample. I can recompile another exe that will work with a larger space.

Edit: videogames finished, so it looks like we'll be finished tonight!