ECM memory usage calculation too low

Alles zum Projekt yoyo@home
Everything about the project yoyo@home
Nachricht
Autor
Benutzeravatar
yoyo
Vereinsvorstand
Vereinsvorstand
Beiträge: 8045
Registriert: 17.12.2002 14:09
Wohnort: Berlin
Kontaktdaten:

Re: ECM memory usage calculation too low

#49 Ungelesener Beitrag von yoyo » 21.01.2011 14:35

Thanks for posting the results.
Would be nice if somebody with a Win64 Intel system can run the tests also.

The 2h runtime of the old version was probably only the runtime until the crash. So you must add something to make it compareable. I think that the long runtime of the new versione is caused by the missing assembler optimizations which you will get with the asmredc extension.

yoyo

btw: I'm running a win32 test with the asmredc version.
HILF mit im Rechenkraft-WiKi, dies gibts zu tun.
Wiki - FAQ - Verein - Chat

Bild Bild

Benutzeravatar
Norman
Klimawolke
Klimawolke
Beiträge: 2188
Registriert: 20.03.2003 14:34
Wohnort: Saarland
Kontaktdaten:

Re: ECM memory usage calculation too low

#50 Ungelesener Beitrag von Norman » 21.01.2011 15:14

tested with ecm63core2-64-asmredc succesfully ! :good:
win7 x64
intel i7 920

Code: Alles auswählen

ecm.exe -v -v -v -v -nn -timestamp -chkpnt checkpnt -inp in 250000 > out

Benutzeravatar
yoyo
Vereinsvorstand
Vereinsvorstand
Beiträge: 8045
Registriert: 17.12.2002 14:09
Wohnort: Berlin
Kontaktdaten:

Re: ECM memory usage calculation too low

#51 Ungelesener Beitrag von yoyo » 21.01.2011 15:15

Super, how long did it run?
yoyo
HILF mit im Rechenkraft-WiKi, dies gibts zu tun.
Wiki - FAQ - Verein - Chat

Bild Bild

Crystal Pellet

Re: ECM memory usage calculation too low

#52 Ungelesener Beitrag von Crystal Pellet » 21.01.2011 15:18

Normally the runtime of 1 curve ~70% low memory and 30% higher and before the crashes it was already running longer with high memory usage.

I'm already running the 64-asmredc version on the same machine. It's running for 2 hours now.
Shall I run the asmredc-version on a 64bit Intel machine too?

I'm running the ecm64rc4-core2-64 too on a faster 64-bit machine Athlon II @2.95GHz and 4GB RAM Win7. Almost 3 hours and should finish soon.
Zuletzt geändert von Crystal Pellet am 21.01.2011 15:20, insgesamt 1-mal geändert.

Benutzeravatar
Norman
Klimawolke
Klimawolke
Beiträge: 2188
Registriert: 20.03.2003 14:34
Wohnort: Saarland
Kontaktdaten:

Re: ECM memory usage calculation too low

#53 Ungelesener Beitrag von Norman » 21.01.2011 15:20

uhmm, i don´t know. i was away for some hours but i think longer then the old one ?!?
should i do the test again ?

Crystal Pellet

Re: ECM memory usage calculation too low

#54 Ungelesener Beitrag von Crystal Pellet » 21.01.2011 15:30

The ecm64rc4-core2-64 version running on my Athlon II Win7 @2.95GHz and 4GB RAM just finished successfull in 2 hours and 48 minutes.
The old version almost ready curve crashed after 1 hour. So this version is also much slower.
Peak working set: 2,082,700K
Max Working Set: 1,866,180K
Peak Virtual Memory: 2,216,164K

The out-file:
GMP-ECM 6.3-rc4 [configured with GMP 5.0.1] [ECM]
Input number is 2486977010...7938509483 (65050 digits)
[Fri Jan 21 12:30:54 2011]
Using special division for factor of 2^216091+1
Using B1=250000, B2=126038650, polynomial Dickson(3), sigma=4045060423
dF=1344, k=7, d=12180, d2=11, i0=10
Expected number of curves to find a factor of n digits:
35 40 45 50 55 60 65 70 75 80
4954 70922 1226642 2.5e+007 5.8e+008 1.6e+010 2.7e+013 4e+018 5.3e+023 Inf
Step 1 took 7535160ms
Estimated memory usage: 1698M
Initializing tables of differences for F took 3900ms
Computing roots of F took 89981ms
Building F from its roots took 136782ms
Computing 1/F took 50045ms
Initializing table of differences for G took 25881ms
Computing roots of G took 63164ms
Building G from its roots took 136782ms
Computing roots of G took 63539ms
Building G from its roots took 137047ms
Computing G * H took 35054ms
Reducing G * H mod F took 67751ms
Computing roots of G took 62993ms
Building G from its roots took 135643ms
Computing G * H took 34959ms
Reducing G * H mod F took 67487ms
Computing roots of G took 63040ms
Building G from its roots took 135502ms
Computing G * H took 35084ms
Reducing G * H mod F took 67518ms
Computing roots of G took 63632ms
Building G from its roots took 137188ms
Computing G * H took 35147ms
Reducing G * H mod F took 67751ms
Computing roots of G took 63945ms
Building G from its roots took 135658ms
Computing G * H took 34945ms
Reducing G * H mod F took 67236ms
Computing roots of G took 62915ms
Building G from its roots took 135409ms
Computing G * H took 35131ms
Reducing G * H mod F took 67408ms
Computing polyeval(F,G) took 241303ms
Computing product of all F(g_i) took 11341ms
Step 2 took 2571489ms
Expected time to find a factor of n digits:
35 40 45 50 55 60 65 70 75 80
1.59y 22.73y 393.12y 8024y 186354y 5e+006y 9e+009y 1e+015y 2e+020y Inf

Crystal Pellet

Re: ECM memory usage calculation too low

#55 Ungelesener Beitrag von Crystal Pellet » 21.01.2011 18:29

On the same AMD Phenom I 4GB RAM 2.1GHz Vista64

Code: Alles auswählen

ecm63core2-64-asmredc.exe -v -nn -timestamp -chkpnt checkpnt -inp in -maxmem 1800 250000 > out
Runtime 4 hours and 49 minutes.
Peak working set: 2,082,472K
Max Working Set: 1,866,284K
Peak Virtual Memory: 2,216,080K

Contents out-file:

GMP-ECM 6.3 [configured with GMP 5.0.1 and --enable-asm-redc] [ECM]
Input number is 2486977010...7938509483 (65050 digits)
[Fri Jan 21 13:09:56 2011]
Using special division for factor of 2^216091+1
Using B1=250000, B2=126038650, polynomial Dickson(3), sigma=2242108136
dF=1344, k=7, d=12180, d2=11, i0=10
Expected number of curves to find a factor of n digits:
35 40 45 50 55 60 65 70 75 80
4954 70922 1226642 2.5e+007 5.8e+008 1.6e+010 2.7e+013 4e+018 5.3e+023 Inf
Step 1 took 12995351ms
Estimated memory usage: 1698M
Initializing tables of differences for F took 5429ms
Computing roots of F took 153489ms
Building F from its roots took 234906ms
Computing 1/F took 84038ms
Initializing table of differences for G took 47221ms
Computing roots of G took 108624ms
Building G from its roots took 234516ms
Computing roots of G took 107594ms
Building G from its roots took 230710ms
Computing G * H took 60201ms
Reducing G * H mod F took 108233ms
Computing roots of G took 107610ms
Building G from its roots took 230678ms
Computing G * H took 58750ms
Reducing G * H mod F took 107797ms
Computing roots of G took 107828ms
Building G from its roots took 230553ms
Computing G * H took 59312ms
Reducing G * H mod F took 108311ms
Computing roots of G took 106206ms
Building G from its roots took 229384ms
Computing G * H took 60668ms
Reducing G * H mod F took 107251ms
Computing roots of G took 113272ms
Building G from its roots took 232738ms
Computing G * H took 59749ms
Reducing G * H mod F took 108389ms
Computing roots of G took 110667ms
Building G from its roots took 229290ms
Computing G * H took 57330ms
Reducing G * H mod F took 108499ms
Computing polyeval(F,G) took 407833ms
Computing product of all F(g_i) took 20577ms
Step 2 took 4338185ms
Expected time to find a factor of n digits:
35 40 45 50 55 60 65 70 75 80
2.72y 38.98y 674.22y 13762y 319609y 9e+006y 1e+010y 2e+015y 3e+020y Inf

Crystal Pellet

Re: ECM memory usage calculation too low

#56 Ungelesener Beitrag von Crystal Pellet » 21.01.2011 21:28

Final test on Intel Q8300 2.5GHz 8GB RAM Vista64

Code: Alles auswählen

ecm63core2-64-asmredc.exe -v -nn -timestamp -chkpnt checkpnt -inp in -maxmem 1800 250000 > out
Runtime 4 hours 15 Minutes
Peak working set: 2,082,636K
Max Working Set: 1,866,116K
Peak Virtual Memory: 2,216,096K

outcome:

GMP-ECM 6.3 [configured with GMP 5.0.1 and --enable-asm-redc] [ECM]
Input number is 2486977010...7938509483 (65050 digits)
[Fri Jan 21 15:40:54 2011]
Using special division for factor of 2^216091+1
Using B1=250000, B2=126038650, polynomial Dickson(3), sigma=3680843894
dF=1344, k=7, d=12180, d2=11, i0=10
Expected number of curves to find a factor of n digits:
35 40 45 50 55 60 65 70 75 80
4954 70922 1226642 2.5e+007 5.8e+008 1.6e+010 2.7e+013 4e+018 5.3e+023 Inf
Step 1 took 11446293ms
Estimated memory usage: 1698M
Initializing tables of differences for F took 5694ms
Computing roots of F took 133599ms
Building F from its roots took 202661ms
Computing 1/F took 75645ms
Initializing table of differences for G took 38719ms
Computing roots of G took 94927ms
Building G from its roots took 202988ms
Computing roots of G took 95020ms
Building G from its roots took 203114ms
Computing G * H took 52323ms
Reducing G * H mod F took 99045ms
Computing roots of G took 94958ms
Building G from its roots took 202614ms
Computing G * H took 52229ms
Reducing G * H mod F took 98983ms
Computing roots of G took 94911ms
Building G from its roots took 202536ms
Computing G * H took 52276ms
Reducing G * H mod F took 99263ms
Computing roots of G took 94989ms
Building G from its roots took 203113ms
Computing G * H took 52136ms
Reducing G * H mod F took 98717ms
Computing roots of G took 96393ms
Building G from its roots took 202864ms
Computing G * H took 51902ms
Reducing G * H mod F took 98826ms
Computing roots of G took 94989ms
Building G from its roots took 202614ms
Computing G * H took 52369ms
Reducing G * H mod F took 98718ms
Computing polyeval(F,G) took 363201ms
Computing product of all F(g_i) took 16817ms
Step 2 took 3829591ms
Expected time to find a factor of n digits:
35 40 45 50 55 60 65 70 75 80
2.40y 34.35y 594.18y 12128y 281669y 8e+006y 1e+010y 2e+015y 3e+020y Inf

Benutzeravatar
yoyo
Vereinsvorstand
Vereinsvorstand
Beiträge: 8045
Registriert: 17.12.2002 14:09
Wohnort: Berlin
Kontaktdaten:

Re: ECM memory usage calculation too low

#57 Ungelesener Beitrag von yoyo » 23.01.2011 09:16

ATH im mersenneforum hat ecm mal mit MPIR statt GMP übersetzt. Diese Version ist wesentlich schneller. Wäre schön mal ein Test auf Intel und AMD auf Win64 zu bekommen.
Hier ist die Version http://www.hoegge.dk/mersenne/ecm63mpir ... smredc.zip
yoyo
HILF mit im Rechenkraft-WiKi, dies gibts zu tun.
Wiki - FAQ - Verein - Chat

Bild Bild

Crystal Pellet

Re: ECM memory usage calculation too low

#58 Ungelesener Beitrag von Crystal Pellet » 23.01.2011 12:44

Muß ich auf einmal Deutsch reden anstatt English. Na ja, auch OK. ;)

Habe 2 tests gestartet:

Code: Alles auswählen

ecm.exe -v -nn -timestamp -chkpnt checkpnt -inp in -maxmem 1800 250000 > out
1 x AMD Athlon II Window7-64 2.95GHz 4GB RAM
1x Intel Q8300 Vista64 2.5GHz 8GB RAM

Beide natürlich mit dem Problem-IN-datei von 65050 charactere aus dem MP-batch.
Resultate erfolgen später.

Crystal Pellet

Re: ECM memory usage calculation too low

#59 Ungelesener Beitrag von Crystal Pellet » 23.01.2011 14:14

Resultat test erfolgreich Athlon II 2.95GHz 4GB RAM Win7 64bit
Runtime: 1 hour and 31 minutes.
Out:
GMP-ECM 6.3 [configured with GMP 5.0.1 and --enable-asm-redc] [ECM]
Input number is 2486977010...7938509483 (65050 digits)
[Sun Jan 23 12:26:45 2011]
Using special division for factor of 2^216091+1
Using B1=250000, B2=126038650, polynomial Dickson(3), sigma=3248293191
dF=1344, k=7, d=12180, d2=11, i0=10
Expected number of curves to find a factor of n digits:
35 40 45 50 55 60 65 70 75 80
4954 70922 1226642 2.5e+007 5.8e+008 1.6e+010 2.7e+013 4e+018 5.3e+023 Inf
Step 1 took 3513767ms
Estimated memory usage: 1698M
Initializing tables of differences for F took 1982ms
Computing roots of F took 45178ms
Building F from its roots took 116034ms
Computing 1/F took 37518ms
Initializing table of differences for G took 12605ms
Computing roots of G took 30451ms
Building G from its roots took 116845ms
Computing roots of G took 30483ms
Building G from its roots took 116127ms
Computing G * H took 29079ms
Reducing G * H mod F took 50840ms
Computing roots of G took 30733ms
Building G from its roots took 116829ms
Computing G * H took 28751ms
Reducing G * H mod F took 51386ms
Computing roots of G took 30717ms
Building G from its roots took 116174ms
Computing G * H took 28845ms
Reducing G * H mod F took 51012ms
Computing roots of G took 30717ms
Building G from its roots took 118373ms
Computing G * H took 30077ms
Reducing G * H mod F took 51636ms
Computing roots of G took 31107ms
Building G from its roots took 118732ms
Computing G * H took 29890ms
Reducing G * H mod F took 50123ms
Computing roots of G took 30435ms
Building G from its roots took 114770ms
Computing G * H took 28252ms
Reducing G * H mod F took 47502ms
Computing polyeval(F,G) took 179074ms
Computing product of all F(g_i) took 10187ms
Step 2 took 1912916ms
Expected time to find a factor of n digits:
35 40 45 50 55 60 65 70 75 80
311.19d 12.20y 211.08y 4308y 100062y 3e+006y 5e+009y 7e+014y 9e+019y Inf

Strange is that in the outfile it says: GMP 5.0.1 and I would expect something like MPIR 2.1.1
Exe used is from ecm63mpircore2-64-asmredc.zip and created 22-1-2011 21:43

Crystal Pellet

Re: ECM memory usage calculation too low

#60 Ungelesener Beitrag von Crystal Pellet » 23.01.2011 14:59

From Intel Q8300 8GB RAM 2.5GHz Vista64
Runtime 2 hours and 6 minutes.

out:
GMP-ECM 6.3 [configured with GMP 5.0.1 and --enable-asm-redc] [ECM]
Input number is 2486977010...7938509483 (65050 digits)
[Sun Jan 23 12:33:24 2011]
Using special division for factor of 2^216091+1
Using B1=250000, B2=126038650, polynomial Dickson(3), sigma=4167528448
dF=1344, k=7, d=12180, d2=11, i0=10
Expected number of curves to find a factor of n digits:
35 40 45 50 55 60 65 70 75 80
4954 70922 1226642 2.5e+007 5.8e+008 1.6e+010 2.7e+013 4e+018 5.3e+023 Inf
Step 1 took 5071187ms
Estimated memory usage: 1698M
Initializing tables of differences for F took 2870ms
Computing roots of F took 64569ms
Building F from its roots took 152725ms
Computing 1/F took 47034ms
Initializing table of differences for G took 17784ms
Computing roots of G took 43119ms
Building G from its roots took 153099ms
Computing roots of G took 43119ms
Building G from its roots took 152928ms
Computing G * H took 36566ms
Reducing G * H mod F took 58048ms
Computing roots of G took 43166ms
Building G from its roots took 152615ms
Computing G * H took 36536ms
Reducing G * H mod F took 58297ms
Computing roots of G took 43088ms
Building G from its roots took 152225ms
Computing G * H took 36832ms
Reducing G * H mod F took 58110ms
Computing roots of G took 43338ms
Building G from its roots took 152881ms
Computing G * H took 36458ms
Reducing G * H mod F took 58297ms
Computing roots of G took 43883ms
Building G from its roots took 152850ms
Computing G * H took 36598ms
Reducing G * H mod F took 57986ms
Computing roots of G took 43212ms
Building G from its roots took 152663ms
Computing G * H took 36520ms
Reducing G * H mod F took 57985ms
Computing polyeval(F,G) took 236420ms
Computing product of all F(g_i) took 14196ms
Step 2 took 2476422ms
Expected time to find a factor of n digits:
35 40 45 50 55 60 65 70 75 80
1.19y 16.97y 293.58y 5992y 139169y 4e+006y 6e+009y 1e+015y 1e+020y Inf


Den Test, den ich vorgestern machte auf der gleiche Maschine mit ecm63core2-64-asmredc.exe dauerte 4 Stunden und 15 Minuten.
Dieses "configured with GMP 5.0.1" im Output verstehe ich das falsch, oder hat ATH das falsch mit kompiliert?

Antworten

Zurück zu „Number crunching“