SoB: Possible P4 speedup

[Ken g6]

Private messages from me to Memnoch, testing a way to possibly get the speed of SoB back on the P4:

Some time ago, I discovered that a program called PRP, the parent program of SoB (get it for windows or Linux) could read the files generated by SoB. Recently I checked and they are completely compatible - they both consider a known prime to be prime, even when their files are interchanged midway.

So the idea is to get PRP, and use it to calculate most of your test. The problem is, while the format of PRP files is identical to SoB, the names are slightly different. So first, you need to get PRP started on your exponent. To do this, create a text file with the following:

47:0:2:257
[your k] [your n]

Run PRP on this and it will create a z*** file, but *** won't be your n. Stop PRP, delete this file, rename your SoB file to replace it, and restart PRP.

Then you just need to see if PRP is faster. It doesn't exactly report cEMs, but it does report its bit position. So you may just have to time it with a stopwatch or something.

As for the speed measurements, I believe you need to compare the bit completion level, over time, to the n.high level produced by SoB.

Good luck, and I hope this works. Please let me know, and let me know which OS(es) you tried (Linux may be faster than Windows, since the client was built later.)

Memnoch says he thinks it sped up processing, but he wasn't absolutely sure.

P.S. One thing I forgot to tell Memnoch: be sure to stop the PRP test before it finishes and switch back to SoB. Otherwise the z*** file is deleted, and your work won't count!!!

 

[Ken g6]

Lunchtime bump...

 

[Allio]

Interesting... are you sure that there aren't inconsistencies? It might be worth asking a project admin first, as they might consider using a separate client to be a deadly sin. If it could compromise the project in any way, it's really not worth it, so it would definitely be wise to get that issue out of the way.

However, it's promising - at the very least maybe the admins can fix their client based on the code in PRP, as it doesn't seem to have a problem with SSE2

 

[amcdonald]

bump. As soon as its fixed you'll have my computers back.

 

[Ken g6]

Originally posted by: Allio
Interesting... are you sure that there aren't inconsistencies? It might be worth asking a project admin first, as they might consider using a separate client to be a deadly sin. If it could compromise the project in any way, it's really not worth it, so it would definitely be wise to get that issue out of the way.

However, it's promising - at the very least maybe the admins can fix their client based on the code in PRP, as it doesn't seem to have a problem with SSE2

I discussed this with Louie a few months back. He said he wanted me to prove that it correctly found a number prime. I tested 6151*2^40960+1 with each program alone, and with the test switched between them. The results were identical, and correct, so I don't think there's a problem.

 

[Allio]

Ah. Well, with the tests getting larger and larger, it could definitely be worthwhile. Anyone with a P4 available to try this out?

 

[MemnochtheDevil]

I'm still using PRP on one client for my P4 machine. Its a bit of a pain monitoring the client to make sure it won't finish while I'm asleep like last time, but its definitely faster than the current client. I use this machine as my main computer (gaming) so it's hard to get solid numbers but I'd guess its 20-25% faster.

 

[Tarca]

true neffer at work here...

18 togo

sorry about your toes