Links to Precompiled GMP-ECM versions

[ATH]

Following GMP-ECM 6.1 binaries were compiled on Windows XP with Msys and MinGW (not the latest versions, does it matter?) and with GMP 4.2.1.
They were compiled with:
./configure --with-gmp=/usr/local
make
make check
make ecm-params; make
make install

Pentium4 Prescott 3.4 Ghz: ecm61-p4p.zip
Pentium4 Northwood 2.4 Ghz: ecm61-p4n.zip
Pentium4 Willamette 1.50 Ghz: ecm61-p4w.zip
Athlon XP 2200+ 1.80Ghz: ecm61-athlon.zip

[jasong]

For those of you who won't, or can't, compile on their own:

I've discovered in my own experience that downloading an ecm.exe file made by someone else, as long as it's compatible, has improved by speed from 10.5 hours to 4.5 hours when comparing the ElevenSmooth client to one I downloaded in this thread.

[ATH]

GMP-ECM 6.1.1 with GMP 4.2.1 compiled on Windows XP with Msys+MinGW with:
./configure --with-gmp=/usr/local
make
make check
make ecm-params; make
make install

Pentium4 Prescott 3.4 Ghz: ecm611-p4p.zip
Pentium4 Northwood 2.4 Ghz: ecm611-p4n.zip (coming august 2nd)
Pentium4 Willamette 1.50 Ghz: ecm611-p4w.zip (coming august 7th-ish)
Pentium3 Celeron 1.10 Ghz (P6C Coppermine128): ecm611-p3c.zip
Athlon XP 2200+ 1.80Ghz: ecm611-athlon.zip

[Mystwalker]

Thanks, ATH!

Did you already test on which architectures "--enable-asm-redc" gave you advantages?

[ATH]

No. I can try, is it enabled or disabled by default?

[ATH]

GMP-ECM 6.1.1 with GMP 4.2.1 compiled on Windows XP with Msys+MinGW. The asmredc files is compiled with --enable-asm-redc, check txt-files for speedtest: asm-redc seems slightly faster up to 190-200 digit numbers. No asm-redc on Pentium3 since no SSE2.

Pentium4 Prescott 3.4 Ghz: ecm611-p4p.zip ecm611asmredc-p4p.zip ecm611-p4p.txt
Pentium4 Northwood 2.4 Ghz: ecm611-p4n.zip ecm611asmredc-p4n.zip ecm611-p4n.txt (coming august 2nd)
Pentium4 Willamette 1.50 Ghz: ecm611-p4w.zip ecm611asmredc-p4w.zip ecm611-p4w.txt (coming august 7th-ish)
Pentium3 Celeron 1.10 Ghz (P6C Coppermine128): ecm611-p3c.zip ecm611-p3c.txt
Athlon XP 2200+ 1.80Ghz: ecm611-athlon.zip ecm611asmredc-athlon.zip ecm611-athlon.txt

[Yamato]

Have you got a precompiled version for Linux, too? I'm using a P4 and I'm not keen on installing GMP-ECM (and the GMP libraries)...

[ATH]

No sorry, I don't have Linux installed.

[akruppa]

I have Linux on a couple of architectures, but sadly, no P4 among them. Only P3, Athlon, Opteron.

Alex

[Yamato]

Following the instructions in the INSTALL and the README file I compiled a linux version for Athlon64, you can get it here.

System: Linux (64bit), Athlon64 (gmp 4.1.3, ecm 6.1.1)

I took the c301 from Kevin Hare for a quick test on an Athlon64 3400+ (@2.4GHz), program output:

ecm 3e6 < in.txt >> out.txt
GMP-ECM 6.1.1 [powered by GMP 4.1.3] [ECM]
Input number is 3831565799519436303487742350308454794716675157894098584352121252263510024611805907320592374654433186020517//1086654671434719340358393954962433533212457600196112076644876654207767427267797808629935905445969160204965//10980740067901995154639576852120198067468078357247366647828551141390739467161074462608561 (301 digits)
Using B1=3000000, B2=5706890290, polynomial Dickson(6), sigma=4115968196
Step 1 took 91619ms
Step 2 took 24242ms

In comparison to my P4 it's really a nice speed (forget all those P4 processors...)

[xilman]

Originally Posted by Yamato

Following the instructions in the INSTALL and the README file I compiled a linux version for Athlon64, you can get it here.

System: Linux (64bit), Athlon64 (gmp 4.1.3, ecm 6.1.1)

I took the c301 from Kevin Hare for a quick test on an Athlon64 3400+ (@2.4GHz), program output:

ecm 3e6 < in.txt >> out.txt
GMP-ECM 6.1.1 [powered by GMP 4.1.3] [ECM]
Input number is 3831565799519436303487742350308454794716675157894098584352121252263510024611805907320592374654433186020517//1086654671434719340358393954962433533212457600196112076644876654207767427267797808629935905445969160204965//10980740067901995154639576852120198067468078357247366647828551141390739467161074462608561 (301 digits)
Using B1=3000000, B2=5706890290, polynomial Dickson(6), sigma=4115968196
Step 1 took 91619ms
Step 2 took 24242ms

In comparison to my P4 it's really a nice speed (forget all those P4 processors...)

Interesting.

I ran your binary on my 3500+ with this result:

pcl@imhotep ~/ecm-6.1.1 $ ./ecm 3000000
GMP-ECM 6.1.1 [powered by GMP 4.1.4] [ECM]
3831565799519436303487742350308454794716675157894098584352121252263510024611805907320592374654433186020517108665467143471934035839395496243353321245760019611207664487665420776742726779780862993590544596916020496510980740067901995154639576852120198067468078357247366647828551141390739467161074462608561
Input number is 3831565799519436303487742350308454794716675157894098584352121252263510024611805907320592374654433186020517108665467143471934035839395496243353321245760019611207664487665420776742726779780862993590544596916020496510980740067901995154639576852120198067468078357247366647828551141390739467161074462608561 (301 digits)
Using B1=3000000, B2=5706890290, polynomial Dickson(6), sigma=2206158585
Step 1 took 48445ms
Step 2 took 18764ms

This is approximately twice the speed of your run for stage 1 and noticeably faster in stage 2. The major difference, as I see it, is that I've GMP 4.1.4 installed. The clock speed of my system is slower than yours (2188.896 MHz) according to /proc/cpuinfo.

Performing the same test on the ecm 6.0.1 I have installed gives this result:

pcl@imhotep ~/ecm-6.1.1 $ /usr/local/bin/ecm 3000000
GMP-ECM 6.0.1 [powered by GMP 4.1.4] [ECM]
3831565799519436303487742350308454794716675157894098584352121252263510024611805907320592374654433186020517108665467143471934035839395496243353321245760019611207664487665420776742726779780862993590544596916020496510980740067901995154639576852120198067468078357247366647828551141390739467161074462608561
Input number is 3831565799519436303487742350308454794716675157894098584352121252263510024611805907320592374654433186020517108665467143471934035839395496243353321245760019611207664487665420776742726779780862993590544596916020496510980740067901995154639576852120198067468078357247366647828551141390739467161074462608561 (301 digits)
Using B1=3000000, B2=4016636513, polynomial Dickson(6), sigma=1196414020
Step 1 took 50620ms
Step 2 took 26185ms

So 6.1.1 is marginally faster in stage 1 and substantially faster in stage 2 despite having a markedly higher B2 value. Both programs run much faster on my machine, despite it having a lower clock speed.

Quite remarkable. I'm wondering whether your GMP has the x64 assembler code incorporated.


Paul