PFGW 3.8.3 (with gwnum v28.7) Released

[rogue]

The Mac version has been posted and hopefully the Linux can be posted in a few hours. This has been hampered by an upgrade to sourceforge the requires me and Steven Harvey (my Linux builder) to check everything out again.

[Honza]

Originally Posted by rogue

I have posted pfgw 3.7.0 at sourceforge. You will have to access it via the "Files" menu option as I don't have it set up as the default d/l for Windows users.

Mark, isn't the ZIP file for Windows corrupted?
Can't unpack using TotalCMD 8, now 7-zip, nor WinRAR 4.20.

EDIT: What switch would you reccomend if I would like to test low ns Genefer with high b values (<100M range)? We have been doing on this back in 2009 but it may have changed since then...

[rogue]

Originally Posted by Honza

Mark, isn't the ZIP file for Windows corrupted?
Can't unpack using TotalCMD 8, now 7-zip, nor WinRAR 4.20.

EDIT: What switch would you reccomend if I would like to test low ns Genefer with high b values (<100M range)? We have been doing on this back in 2009 but it may have changed since then...

I'll take a look at the zip file tomorrow.

I believe that -gx is what your looking for.

[WMHalsdorf]

In the Windows zip file the 64 bit version is corrupt. The 32 bit version seems to work just fine.

[pepi37]

Also on Boinc Confederation PFGW is not 3.7.0 as reported, but old one , since it is in byte same as version of pfgw in prp522 package
And I also try several times to download from sourceforge , but archive is invalid :(

[pepi37]

Originally Posted by henryzz

Did I read you saying on the mailing group that it was faster on 32-bit because of asm?

I try and tested every available version of PFGW (32 or 64 bit) and never find that PFGW was faster then LLR.
64 bit LLR was always much faster then 64 bit PFGW
32 bit LLR was +/-small fraction of time same as 32 bit PFGW
Both are tested on 64 bit Windows 7

So if I understand correctly on 32 bit host PFGW is faster from what program ( llr )?
Thanks for reply!
P.S PFGW32 bit 3.7.0 can be extracted from archive in sourceforge, and as stated above, works fine.
PFGW64 bit 3.7.0 is broken :( ( please fix it)

[rogue]

Originally Posted by pepi37

I try and tested every available version of PFGW (32 or 64 bit) and never find that PFGW was faster then LLR.
64 bit LLR was always much faster then 64 bit PFGW
32 bit LLR was +/-small fraction of time same as 32 bit PFGW
Both are tested on 64 bit Windows 7

So if I understand correctly on 32 bit host PFGW is faster from what program ( llr )?
Thanks for reply!
P.S PFGW32 bit 3.7.0 can be extracted from archive in sourceforge, and as stated above, works fine.
PFGW64 bit 3.7.0 is broken :( ( please fix it)

The changes for 3.7.0 affect trial factoring, which is only done for single tests or odd forms that don't have a sieving program.

If you are referring to base 2, then llr is much faster. I have heard similar claims for other bases, but I haven't compared them myself.

Even if llr is faster for other bases, I would prefer to leave pfgw alone. The simple reason is that pfgw and llr perform a different test for primality. This allows one to use the both verify primality. (Changing the primality proving code of pfgw would be very hard as the code is fairly generic, i.e. it is not only use for k*b^n+/-1 but other forms as well.)

[Honza]

Originally Posted by pepi37

I try and tested every available version of PFGW (32 or 64 bit) and never find that PFGW was faster then LLR.

Generally, yes. But for example Genefer numbers, LLR is not so good. Genefer is faster when running several instanaces (but giving only PRP). PFGW is ~1% slower than LLR, that's very close.
It can differ more depending on CPU/cache/AVX/64-bit version and tested numbers/formats.
____

I've conducted a small test with GFN32768 and Genefer80, LLR64, PFGW64/AVX on i5-3570. All 4 apps where run at once.

>genefer80.exe -q "4108672^32768+1"
genefer80 2.3.0-0 (Windows x86 80-bit x87)
...
4108672^32768+1 is a probable prime. (216718 digits) (err = 0.0022) (time = 0:17:46)
[Honza] That's 1065 sec.

>CLLR64.exe -q "4108672^32768+1"
4108672^32768+1 is prime! Time : 1158.581 sec.

>pfgw64.exe -q"4108672^32768+1"
PFGW Version 3.6.3.64BIT.20120316.Win_Dev [GWNUM 27.5]
4108672^32768+1 is 3-PRP! (1164.8072s+0.0041s)

>pfgw64.exe -q"4108672^32768+1"
PFGW Version 3.6.7.64BIT.20121129.Win_Dev [GWNUM 27.8]
4108672^32768+1 is 3-PRP! (1166.5173s+0.0045s)

Running only a single instance

>pfgw64.exe -q"4108672^32768+1"
PFGW Version 3.6.7.64BIT.20121129.Win_Dev [GWNUM 27.8]
4108672^32768+1 is 3-PRP! (875.9579s+0.0053s)

>CLLR64.exe -q "4108672^32768+1"
4108672^32768+1 is prime! Time : 870.493 sec.

>genefer80.exe -q "4108672^32768+1"
...
4108672^32768+1 is a probable prime. (216718 digits) (err = 0.0022) (time = 0:15:41)
[Honza] That's 941 sec

When running a single test, PFGW64 and LLR are fastest, not Genefer80

[pepi37]

But I only use PFGW for LLR searching, not for Genefer, and in that case PFGW is about 10% slower , and even more, and when you calculate hundreds of results , and every result is calculating about hour, difference is noticeable :)

LLR

C:\Users\Phenom>C:\Users\Phenom\Desktop\llr64.exe -d -q"7061*2^293569+1"
Starting Proth prime test of 7061*2^293569+1
Using all-complex AVX FFT length 24K, Pass1=128, Pass2=192, a = 3
7061*2^293569+1 is prime! Time : 27.093 sec.

C:\Users\Phenom>C:\Users\Phenom\Desktop\llr64.exe -d -q"6679*2^449430+1"
Starting Proth prime test of 6679*2^449430+1
Using all-complex AVX FFT length 32K, Pass1=128, Pass2=256, a = 3
6679*2^449430+1 is prime! Time : 55.232 sec.

PFGW

C:\Users\Phenom>C:\Users\Phenom\Desktop\pfgw64.exe -d -q"7061*2^293569+1"
PFGW Version 3.6.7.64BIT.20121129.Win_Dev [GWNUM 27.8]

7061*2^293569+1 is 3-PRP! (30.3008s+0.0002s)


C:\Users\Phenom>C:\Users\Phenom\Desktop\pfgw64.exe -d -q"6679*2^449430+1"
PFGW Version 3.6.7.64BIT.20121129.Win_Dev [GWNUM 27.8]

6679*2^449430+1 is 3-PRP! (63.3292s+0.0002s)

---------------------------------------------------------------
So 27 seconds compared to 30 seconds, and 55 seconds compared to 63 seconds are far away from only 1% of difference :)

[Honza]

I'm glad you found what suites you and your hardware.

As I stated before, it depends on several factors like CPU, tests done, single or on all cores etc.
Intel i7-3820, 32GB RAM, LLR64 and PFGW gives same runtime when only single instance.

D:\temp\PG>llr64.exe -d -q"7061*2^293569+1"
7061*2^293569+1 is prime! Time : 30.263 sec.

D:\temp\PG>llr64.exe -d -q"7061*2^293569+1"
7061*2^293569+1 is prime! Time : 30.306 sec.

[pepi37]

Yes, it looks like it is CPU dependent. Thanks for point this thing up Honza.