PRPclient 2.4.7 with LLR 3.8.0 now available!

[Jean Penné]

Originally Posted by Sloth

so in english that means we are fine but other projects not using +/- 1 have problems?


S.

They have problem with "standard" released V 3.8.0, but :

Users who are testing k*b^n+c numbers with |c| !=1 can now download the
fixed development version from my development directory :

http://jpenne.free.fr/Development/

Please, let me know if any problem is remaining...

Best Regards,
Jean

[opyrt]

Originally Posted by Jean Penné

Perhaps you did not see my last post about the |c| !=1 bug in LLR 3.8.0 :

http://www.mersenneforum.org/showthread.php?t=13072

You're quite correct, I didn't see it, thanks!

I will update our forums and webpages to point to the newest package as soon as I find the time.

[VJS]

Are you guys sure its faster???

It's looking like my times are now in the low 11ms as opposed to the high 10's with 3.7...

n's are near 5.8M

phenom XII at 3.8Ghz

WinXP-64

[opyrt]

Hmmm...

That's very interresting! I've only tested on Intel Quads and Intel Xeons (of the same generation). I have however tested both Win32 and Win64 and saw an increase in speed on both.

Any chance you could check one specific candidate running on all cores with 3.7.1c and the same candidate running on all cores with 3.8.0?

Have you checked if FFT coincidentally increased while you were doing your testing?

[VJS]

Well thats very interesintg...

I'd suggest that people hard boot there computer after install.

My speeds were roughly 65500 seconds each with 3.7.1

After upgrading I saw no difference in those running on core1 and those running on core0 increased to above 67000sec. Only completed 3.

My wife was complaining about the speed of the computer so I rebooted...

Looks like the speed is back to 3.7.1 levels, hard to say if its faster yet. Certainly not slower.

just odd hopefully is not something on the machine.

[opyrt]

Originally Posted by VJS

Looks like the speed is back to 3.7.1 levels, hard to say if its faster yet. Certainly not slower.

It's good that it's not slower, but I find it strange that a reboot should affect anything. I have some team mates with phenoms, I'll see if I can get them to test to see if they get the same results.

Does anyone else have any experiences to share with regards to 3.7.1 vs 3.8.0 on AMDs?

[Cybertronic]

I have no important speed up's between 3.7 and 3.8
(machine is a Phenom II 3,4 GHz , without oc)

LLR 3.7: 148,834 s for 65516468355*2^333333-1
LLR 3.8: 145,195 s for 65516468355*2^333333-1

2,4% faster or 35min faster per day.

[opyrt]

Originally Posted by Cybertronic

I have no important speed up's between 3.7 and 3.8
(machine is a Phenom II 3,4 GHz , without oc)

LLR 3.7: 148,834 s for 65516468355*2^333333-1
LLR 3.8: 145,195 s for 65516468355*2^333333-1

2,4% faster or 35min faster per day.

That's within the 2-5% speed-up that Jean predicted.
I've only seen a higher speed increase on quad cores or more where LLR is running on all cores. It's a known fact that one instance of LLR alone is faster than one of four instances on a quad. It seems to me that "slowdown on multiple cores" issue is somewhat smaller on LLR 3.8.0.

Thanks for posting your numbers!