mini-drive for high-n testing on Sierp base 4

[gd_barnes]

To all,

About a month ago, Tcadigan released his Sierp base 4 k=64494 and provided us with a sieved file. He had tested it to n=885253 base 4 (n=1770506 base 2) and the file goes up to n=1M. Just for grins, last night, I put it on one of my 3.2 Ghz P4's. It completed 3 tests overnight although I haven't checked it today yet. I was getting about 9500 secs. per test but those of you with a 2+ Ghz duo should get faster.

I don't want to officially reserve it but will test it here and there as fill-in work on 1 or 2 of my machines. Anyone is free to take it off of my hands at any point.

What I'd like to do is divvy up the file by small n-ranges in a mini and unofficial team drive. Would anyone have interest in helping with that?

FYI, this is technically part of the Sierp base 4 project but we are coordinating it here. If we find a prime for it, I would suggest reporting it under both projects so that both get half credit. Jean or Tcadigan, does that sound reasonable? At some point in the future, we'll probably also provide some sieving help for Jene Penne's three Sierp base 4 k's after he's done LLRing them to n=2M base 2.

Edit: We have now started a mini-drive for this. See the 5th post of this thread below.


Thanks,
Gary

[tcadigan]

Originally Posted by gd_barnes

To all,
What I'd like to do is divvy up the file by small n-ranges in a mini and unofficial team drive. Would anyone have interest in helping with that?

FYI, this is technically part of the Sierp base 4 project but we are coordinating it here. If we find a prime for it, I would suggest reporting it under both projects so that both get half credit. Jean or Tcadigan, does that sound reasonable? At some point in the future, we'll probably also provide some sieving help for Jene Penne's three Sierp base 4 k's after he's done LLRing them to n=2M base 2.

I too tried it out for a couple runs on my new computer just to see the difference, I get ~7000 seconds on 2.8 C2Q

the crediting sounds fine to me. I just really want a prime to be found

as for the breaking up the file...I would lean towards it if it isn't too much effort to create them. since the tests take so long I doubt there will be vast amounts of interest in them (if any) and maintenance shouldn't be all that high, and like you said it fills a good purpose for temp idle comps.

[gd_barnes]

Originally Posted by tcadigan

I too tried it out for a couple runs on my new computer just to see the difference, I get ~7000 seconds on 2.8 C2Q

the crediting sounds fine to me. I just really want a prime to be found

as for the breaking up the file...I would lean towards it if it isn't too much effort to create them. since the tests take so long I doubt there will be vast amounts of interest in them (if any) and maintenance shouldn't be all that high, and like you said it fills a good purpose for temp idle comps.

OK, I may do that. It may just be you and me trudging slowly forward but that's fine.

Quick question...How far was the file sieved? It shows P=1G in the file but I know it's much further than that. Just thought I'd check to make sure we're sieved far enough before creating files for a mini-drive.


Gary

[mdettweiler]

If you did a do a mini-drive like this, then I'd be glad to do some testing for it.

Also, assuming we could find someone with an always-on machine and internet connection to host an LLRNet server, this would be perfect for LLRNet; the LLRnet server shouldn't use hardly any resources, so any always-on cruncher with a broadband connection should be perfect. I'd be glad to administrate an LLRNet server given SSH/VNC/Remote Desktop/etc. access. Even after we've cleaned up this k, then it would still be useful for other drives--LLRnet is sort of like a regular team drive, except that it's done automatically over the Internet, rather than through a forum thread.

Any volunteers for hosting an LLRNet server?

[gd_barnes]

We are starting a mini-drive for one of the final 4 remaining k's for Sierp base 4...k=64494 using Tcadigan's sieved file that he left us. The range is n=885K-1M base 4.

I have divided it up in to appropriate-sized files, n=5K at a time. There are 964 total candidates remaining in 23 files, which means an avg. of 42 candidates/file.

On a 3.2 Ghz P4, testing time is about 9600 secs./test. Tcadigan quoted 7000 secs./test on a 2.8 Ghz Core 2 quad. On the P4, that's an avg. of 4.7 CPU days/file. On the C2Q, avg. would be 3.4 CPU days/file. Feel free to take what you can do in a reasonable time frame.

Please post all reservations, status, and primes (oh, please a prime!) here.

Any prime should be reported under both the Sierp base 4 project and CRUS; half-credit each.

I have reserved and am testing the first piece of the file from n=885K-890K and am at about n=888K now.

Status:

n-range     tested by      Status       
925K-1000K  LLRnet (IB4)   complete
920K-925K   Anonymous      complete
890K-920K   em99010pepe    complete
885K-890K   gd_barnes      complete

Files for LLRing:
The mini-drive is now complete. Thanks to all who contributed!


As for results files, feel free to wait until you're done with this effort before posting them.

Let's take down a MEGAbit prime!!


Gary

[em99010pepe]

Taking 890-910.

Taking 910-915.

Taking 915-920.

[gd_barnes]

Originally Posted by em99010pepe

Taking 890-910.

Taking 910-915.

Taking 915-920.

Ah, nothing like a drive to generate a little interest here!

Good luck, Carlos!

[em99010pepe]

Thanks Gary, my quad-core tests a number in less than 6200 sec. Core 0 to 2 between 6000-6100 sec, core 3 between 6100-6200 sec....With FFT length at 192K the forth core just can't handle the same speed due to cache limitations...

Carlos

[Jean Penné]

Good work, Gary!

About k = 18534, only 16 exponents are remaining on my test file, so I will complete this work probably before the end of this month (and very likely with no prime found...). I don't know if a sieve file exists for exponents larger than my limit (2,000,000 base two) ; if yes, it will be the time to split it as you did for k = 64494, if no, there will be an important sieve work to do...

Regards,
Jean

[gd_barnes]

Originally Posted by em99010pepe

Thanks Gary, my quad-core tests a number in less than 6200 sec. Core 0 to 2 between 6000-6100 sec, core 3 between 6100-6200 sec....With FFT length at 192K the forth core just can't handle the same speed due to cache limitations...

Carlos

Amazing machines Carlos! If I know you well enough, you'll figure out a way to increase the cache so that you get full speed out of all 4 cores!

Originally Posted by Jean Penné

Good work, Gary!

About k = 18534, only 16 exponents are remaining on my test file, so I will complete this work probably before the end of this month (and very likely with no prime found...). I don't know if a sieve file exists for exponents larger than my limit (2,000,000 base two) ; if yes, it will be the time to split it as you did for k = 64494, if no, there will be an important sieve work to do...

Regards,
Jean

Nope, no sieving above n>1M base 4 that I know of. So much sieving on different things is needed! We'll see what we can do.

If you guys haven't seen it, there's some good-sized testing needed now on drive #1 - Sierp base 16. Ranges from n=100K-120K have been posted.


Gary

[em99010pepe]

Originally Posted by gd_barnes

Amazing machines Carlos! If I know you well enough, you'll figure out a way to increase the cache so that you get full speed out of all 4 cores!

Impossible! It's a CPU limitation, the real problem of the false quad-core.

Originally Posted by gd_barnes

If you guys haven't seen it, there's some good-sized testing needed now on drive #1 - Sierp base 16. Ranges from n=100K-120K have been posted.

If you want I can easily clean those ranges in a few days....just let me know...