Leapfrog time!

Joe O · 2010-09-23, 21:51

http://www.mersenneforum.org/showthr...137#post231137
LLR V3.8.2 using gwnum 26.2 is available!
OK PFGW, it's your turn!

rogue · 2010-09-23, 22:19

Quote:

Originally Posted by Joe O

http://www.mersenneforum.org/showthr...137#post231137
LLR V3.8.2 using gwnum 26.2 is available!
OK PFGW, it's your turn!

It's close and I have a 64-bit version pending, but it is not ready for prime time yet. George told me about problems with some FFT sizes that will be fixed in v26.3, so it will probably be released with that version of gwnum.

Joe O · 2010-09-24, 23:41

Round off problem fixed on Windows binaries!
http://www.mersenneforum.org/showthr...307#post231307

I recommend we wait for PFGW. If you have downloaded LLR 3.8.2 for Windows, you must download it again. The round off problem mentioned above, applies to our tests.

Joe O · 2010-09-25, 21:25

http://www.mersenneforum.org/showthread.php?t=13969

I would like to urge some caution in adopting this new release since there are so many changes. Perhaps we could start out with timing benchmarks?
32 bit PFGW 3.4.0 vs 32 bit pfgw_win_3.3.6_20100907 timings and 32 bit PFGW 3.4.0 vs 64 bit PFGW 3.4.0 timings comparing the residuals as well. Let's use 71492*5^378851+1 as our test test <G>.

ltd · 2010-09-26, 16:55

Here some results from an i7-920 and windows7-64.

V336 used a zero padded FFT 96K.
V34 used a all-complex Core2 type-1 FFT length 80K in 32 and 64 bit.

All residues were identical.

Runtimes:

V336: 2293 seconds
V34 64Bit: 2078 seconds
V34 32Bit: 1863 seconds

So at least on my system the 64bit version is significant slower then the 32bit version of pfgw 3.4.

Joe O · 2010-10-22, 19:40

You can download the latest release for Windows, MacIntel, and Linux from here: http://sourceforge.net/projects/openpfgw/
I'd like to see some timings for this release vs prior releases using 71492*5^378851+1 as our test test. I would also like to see some people test it in their normal work and report their findings. So far, I am pleased with it and have found no problems.

Ok LLR, it's your turn.

mdettweiler · 2010-10-22, 20:09

Quote:

Originally Posted by Joe O

You can download the latest release for Windows, MacIntel, and Linux from here: http://sourceforge.net/projects/openpfgw/
I'd like to see some timings for this release vs prior releases using 71492*5^378851+1 as our test test. I would also like to see some people test it in their normal work and report their findings. So far, I am pleased with it and have found no problems.

Ok LLR, it's your turn.

FYI, as I understand it PFGW 3.4.2 should be the exact same speed as 3.4.1 for most people. By that same token, LLR 3.8.2 should be no slower than the (assumedly) forthcoming 3.8.3 based on the same gwnum version as PFGW 3.4.2.

The only speedup I've observed/heard of going from PFGW 3.4.1>3.4.2 is on my own Core 2 Duo, which had a mysterious unexplained sluggishness on 3.4.0 and 3.4.1. It was totally unreproducible by anyone else; it would appear to be a very, very isolated issue and it doesn't even affect LLR or Prime95 at all. I suspect it is merely one of the many crazy issues I've run into on this particular computer since its Windows installation is kind of messed up.

Joe O · 2010-10-26, 14:38

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

Jean Penne made the announcement here.

Joe O · 2011-01-22, 16:44

http://www.mersenneforum.org/showthr...382#post248382

Jean Penne has made a change to LLR V3.8.4. Please refetch your binaries, and or source.

2010-09-23, 21:51	#1
Joe O Aug 2002 3×5²×7 Posts	Leapfrog time! http://www.mersenneforum.org/showthr...137#post231137 LLR V3.8.2 using gwnum 26.2 is available! OK PFGW, it's your turn!

2010-09-25, 21:25	#4
Joe O Aug 2002 3·5²·7 Posts	PFGW 3.4.0 Released http://www.mersenneforum.org/showthread.php?t=13969 I would like to urge some caution in adopting this new release since there are so many changes. Perhaps we could start out with timing benchmarks? 32 bit PFGW 3.4.0 vs 32 bit pfgw_win_3.3.6_20100907 timings and 32 bit PFGW 3.4.0 vs 64 bit PFGW 3.4.0 timings comparing the residuals as well. Let's use 714925^378851+1 as our test test <G>. Last fiddled with by Joe O on 2010-09-25 at 21:29*

2010-10-22, 19:40	#6
Joe O Aug 2002 20D₁₆ Posts	PFGW 3.4.2 Released You can download the latest release for Windows, MacIntel, and Linux from here: http://sourceforge.net/projects/openpfgw/ I'd like to see some timings for this release vs prior releases using 71492*5^378851+1 as our test test. I would also like to see some people test it in their normal work and report their findings. So far, I am pleased with it and have found no problems. Ok LLR, it's your turn.

2010-10-26, 14:38	#8
Joe O Aug 2002 1000001101₂ Posts	LLR 3.8.4 dev. version is available for testing! Available at http://jpenne.free.fr/Development/ Jean Penne made the announcement here.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Time to End	davar55	Lounge	4	2013-02-23 02:40
How much time	Unregistered	Information & Answers	4	2008-12-20 21:00
New .dat time?	benjackson	Prime Sierpinski Project	16	2008-07-29 07:26
Time	Xyzzy	Science & Technology	26	2008-01-19 03:28
P3 TF time	PrimeCruncher	Software	30	2003-12-21 05:26

2010-09-24, 23:41	#3
Joe O Aug 2002 20D₁₆ Posts	Round off problem fixed on Windows binaries! http://www.mersenneforum.org/showthr...307#post231307 I recommend we wait for PFGW. If you have downloaded LLR 3.8.2 for Windows, you must download it again. The round off problem mentioned above, applies to our tests.

2010-09-26, 16:55	#5
ltd Apr 2003 1404₈ Posts	Here some results from an i7-920 and windows7-64. V336 used a zero padded FFT 96K. V34 used a all-complex Core2 type-1 FFT length 80K in 32 and 64 bit. All residues were identical. Runtimes: V336: 2293 seconds V34 64Bit: 2078 seconds V34 32Bit: 1863 seconds So at least on my system the 64bit version is significant slower then the 32bit version of pfgw 3.4.

2011-01-22, 16:44	#9
Joe O Aug 2002 1015₈ Posts	http://www.mersenneforum.org/showthr...382#post248382 Jean Penne has made a change to LLR V3.8.4. Please refetch your binaries, and or source.