| Author |
Message |
|---|
|
|
|
Didn't want to hijack the reservation thread ;)
Will there be a -t switch for multiple core use, as in 321Sieve?
Regards, Peter
____________
There are only 10 kinds of people - those who understand binary and those who don't
 |
|
|
John Honorary cruncher
Send message
Joined: 21 Feb 06
Posts: 2875
ID: 2449
Credit: 2,681,934
RAC: 0
                 
|
Didn't want to hijack the reservation thread ;)
Will there be a -t switch for multiple core use, as in 321Sieve?
Regards, Peter
That is definitely a feature we'd like to add in future versions for Linux. It is unknown right now when this will be available.
____________
|
|
|
geoffVolunteer developer
Send message
Joined: 3 Aug 07
Posts: 99
ID: 10427
Credit: 343,437
RAC: 0
|
|
The psieve-1.2.1-linux-x86.zip archive seems to contain a 64-bit executable.
|
|
|
JohnHonorary cruncher
Send message
Joined: 21 Feb 06
Posts: 2875
ID: 2449
Credit: 2,681,934
RAC: 0
|
The psieve-1.2.1-linux-x86.zip archive seems to contain a 64-bit executable.
Thanks for the heads up. Should be correct now. :)
____________
|
|
|
|
|
|
Bug report, but specific to my configration, I presume.
I'm running psieve on a Laptop with WinXP32, which I put into hibernation from time to time by closing the lid. Sometimes it doenn't wake up properly, presumably causing problems.
One thing: I found after restarting the program that it was crunching a file of 380KB, instead of 9,8MB.
Possible explanation: it was just updating the file when the computer went into hibernation and didn't wake up.
Proposed workaround: Disable the ability to update the file. Doesn't cause any significant speed increases anyway anymore.
Other thing: Once, the checkpoint.txt was smaller than the last factor.
Possible explanation: Checkpoint and factors are different things.
Proposed workaround: Update checkpoint.txt right after updating the factor file.
That's it for the moment; as I will probably not be able to reproduce the errors, I might be completely off. Sorry for that. H. |
|
|
JohnHonorary cruncher
Send message
Joined: 21 Feb 06
Posts: 2875
ID: 2449
Credit: 2,681,934
RAC: 0
|
I'm running psieve on a Laptop with WinXP32, which I put into hibernation from time to time by closing the lid. Sometimes it doenn't wake up properly, presumably causing problems.
One thing: I found after restarting the program that it was crunching a file of 380KB, instead of 9,8MB.
Possible explanation: it was just updating the file when the computer went into hibernation and didn't wake up.
Proposed workaround: Disable the ability to update the file. Doesn't cause any significant speed increases anyway anymore.
First occurrence of this that we know of. It will be interesting to see if you or anyone else can recreate it. If so, this is very problematic. We'll investigate.
Other thing: Once, the checkpoint.txt was smaller than the last factor.
Possible explanation: Checkpoint and factors are different things.
Proposed workaround: Update checkpoint.txt right after updating the factor file.
Not a big issue...just means that a factor was found after the checkpoint.txt was saved. However, researching this did reveal to us that the checkpoint default save value is every 60 minutes. Too much for our comfort level. You can edit the .bat file to include -s#, with # being the number of minutes between saves. Therefore, -s10 will save the checkpoint.txt file every 10 minutes.
Thanks for bringing this up.
-----------------------------------------
For those who haven't checked the help option, here's a print of it:
psieve 1.2.1 -- A sieve for integer sequences of the form n#+1 and n#-1.
Usage: psieve [OPTION ...] <SEQUENCE|FILE ...>
-p --pmin P0 Look for factors p in P0 <= p <= P1 ...
-P --pmax P1
-n --nmin N0
-N --nmax N1
-i --input FILE Read sieve from ABC format FILE (default 'psieve.pfgw').
-o --output FILE Write sieve to ABC format FILE (default 'psieve.pfgw'.
-r --report X Make status reports every X seconds (default 60).
-s --save X Save sieve or write checkpoint every X minutes (default 60).
-S --stop-rate X Stop when it takes X seconds to eliminate a candidate.
-m --minimum X Do not report factors smaller than X (default 100000).
-f --factors Append reported factors to the file factors.txt.
-c --check Check that reported factors really do divide the candidate.
-C --checkpoint Write (and resume from) checkpoint file `checkpoint.txt'.
-k --known-factors FILE Remove factors in FILE from the sieve.
-d --duplicates Report duplicate factors (ones that don't eliminate any n).
-z --idle Run at idle priority. (Default)
-Z --no-idle Don't adjust priority.
-A --affinity N Set affinity to CPU number N.
-v --verbose Be verbose.
-q --quiet Be quiet.
-h --help Print this help.
____________
|
|
|
|
|
|
Follow up:
Indeed it doesn't seem to like hibernation so much. When I came back this evening, I got this screen output:
p=50600818783, 2940 p/sec, 1484 sec/n, 157 terms eliminated, 868251 remain
p=50601018847, 2911 p/sec, 1486 sec/n, 157 terms eliminated, 868251 remain
p=50601058867, 2 p/sec, 2104 sec/n, 157 terms eliminated, 868251 remain
Wrote 868251 terms to PFGW format file psieve.pfgw
p=50601098887, 2724 p/sec, 2105 sec/n, 157 terms eliminated, 868251 remain
Wrote 868251 terms to PFGW format file psieve.pfgw
p=50601138967, 2853 p/sec, 2105 sec/n, 157 terms eliminated, 868251 remain
Wrote 868251 terms to PFGW format file psieve.pfgw
p=50601178967, 2952 p/sec, 2106 sec/n, 157 terms eliminated, 868251 remain
Wrote 868251 terms to PFGW format file psieve.pfgw
p=50601218999, 2993 p/sec, 2106 sec/n, 157 terms eliminated, 868251 remain
Wrote 868251 terms to PFGW format file psieve.pfgw
p=50601238741, 2746 p/sec, 2106 sec/n, 157 terms eliminated, 868251 remain
Wrote 868251 terms to PFGW format file psieve.pfgw
p=50601278747, 3030 p/sec, 2107 sec/n, 157 terms eliminated, 868251 remain
p=50601318749, 3137 p/sec
And then it printed the status line every few seconds, instead of every minute. Obviously, as time has passed during the hibernation, it tries to catch up what it couldn't do when the computer was off.
Another possible exlanation of the problem of the last post would be (I don't remember so well): Hibernation, wake up, continued writing to the disk, I don't look and close the program (perhaps forced), not finished writing, problem.
My opinion is still that as a general guideline, sieve clients should ever only READ the sieve files, and never WRITE to them (with exception of newpgen, of course).
H.
|
|
|
JohnHonorary cruncher
Send message
Joined: 21 Feb 06
Posts: 2875
ID: 2449
Credit: 2,681,934
RAC: 0
|
|
We're looking for a 32 bit Linux user. We need testing on a new psieve which takes advantage of "SSE4.1 and/or SSE4.2 if available".
If anyone is willing to run some timing tests, please PM me.
Thanks!
EDIT: SSE4.1 was introduced with the Penryn core, SSE4.2 with the Nehalem core.
____________
|
|
|
 warddrVolunteer tester
Send message
Joined: 7 Feb 08
Posts: 254
ID: 18735
Credit: 24,054,820
RAC: 0
         
|
|
Can someone make a new sieve-file, the one we use now is quite old (30G).
I made a list from all factors found since then (1,5 Mb), I did not do this manualy but I made a python script to do this.
pfactors_30G-185G.txt
They are not completely in the right order but all factors are in there.
If someone is interested I can send the python file I made to do this (It isn't the best way to do it but it works very quick).
If someone can explain me how to make a new sieve-file I can always make the sieve-file myself and upload it to my server, I have unlimited transfer.
____________
 |
|
|
VatoVolunteer tester
Send message
Joined: 2 Feb 08
Posts: 851
ID: 18447
Credit: 713,640,206
RAC: 1,642,230
                          
|
|
John/Lennart will do this using 'srfile' I believe.
Until then, probably best to continue using the official file.
____________
 |
|
|
|
|
|
Seeing that this project has a Mac client available, I downloaded the psieve app to my PPC OS X box and tried to run it, but was presented with a "cannot execute binary file" error. Any one know how to get around this? Thanks.
____________
|
|
|
rogueVolunteer developer
Send message
Joined: 8 Sep 07
Posts: 1256
ID: 12001
Credit: 18,565,548
RAC: 0
|
|
chmod 755 psieve |
|
|
|
|
|
I just tried that command, and it still is giving me that same error. Is there something different about the zip file. I extracted the zip with stuffit, which works with other command line apps, like the d.net client. I even used sudo, just in case it was a permissions issue.
____________
|
|
|
rogueVolunteer developer
Send message
Joined: 8 Sep 07
Posts: 1256
ID: 12001
Credit: 18,565,548
RAC: 0
|
|
I suspect that the file was not transferred in binary mode. |
|
|
|
|
|
Wow once Happy5214 and my bro Stephen Brown get there's done we will be able to go from psieve_p185Ge0.zip to psieve_p226Ge0.zip!! It makes it faster to sieve thats for sure :)
____________
John M. Johnson "Novex" |
|
|
|
|
|
Yeah this goes back to the time when we all were doing the 321 sieve and we had sieve files that helped a little if at all, and then we had sieve files that when we put them in KA BAM! 3 ninjas flew out and said, it's speed time. :) Toughs are the ones that helped. Just takes more then 2 or 3 sieve files and how happy the complier is at the time. :) But don't loose faith or a rifle. Hope that helps with the question Dar1008 :)
Edit: Are we allowed to stick with what we have now since its been working great on the "at the time" 2 computers I don't touch, but to transfer sieve files over to the server to upload to you all. As long as you guys allow the sieve files to keep on working with what we have now:)
____________
John M. Johnson "Novex" |
|
|
JohnHonorary cruncher
Send message
Joined: 21 Feb 06
Posts: 2875
ID: 2449
Credit: 2,681,934
RAC: 0
|
Edit: Are we allowed to stick with what we have now since its been working great on the "at the time" 2 computers I don't touch, but to transfer sieve files over to the server to upload to you all. As long as you guys allow the sieve files to keep on working with what we have now:)
You can use fsieve or fpsieve. They are the same speed. However, fpsieve has the -t switch which makes it easier to sieve on multicore computers.
____________
|
|
|
|
|
|
Question, I'm setting up this fpsieve on my dual core for the first time, but since it lists in the editing part of the psieve as 292G then do I need to download the lasted and also put that into the folder?? It didn't come with for starters. Thats why I ask. If anyone can help fast that would be great since the moment I get the answer and get it up and running it will speed this up big time!!! Thank You
Edit I have it set up all the way it should be and the box comes up and goes away insta fast. so I know somethings wrong. I am running the fpsieve.bat so it can run you know. If anyone can help I would love and appreciate it :)
____________
John M. Johnson "Novex" |
|
|
|
|
|
http://www.primegrid.com/forum_thread.php?id=1154
Can't seam to get the -t2 to work at all the box just blinks like its opening then closing fast.
____________
John M. Johnson "Novex" |
|
|
LumiukkoVolunteer tester
Send message
Joined: 7 Jul 08
Posts: 165
ID: 25183
Credit: 875,332,634
RAC: 114,791
             
|
... box comes up and goes away insta fast....
Sorry about posting to the wrong thread.
If you want the box to remain open after the bat finishes, just open the console box first.
So start Command Prompt: RUN "CMD", or click Start -> All Programs -> Accessories -> Command Prompt.
CD to the directory where the bat is installed: Cd "C:\Program Files\fpsieve" or whatever the directory is.
and run the bat from the console:
Just type the command with possible options: fpsieve.bat -t2, and hit <ENTER>.
When the bat finishes the console will remain open, and you can read possible warnings/errors.
If you are running Windows Vista the problem is most probably related to directory/file permissions.
I do not know what is needed as I have never used these manual sieve programs, but I quess you may need to have write permissions to the installation directory.
In Vista the default is to NOT give user write permissions to directories under "Program Files" directory. So you need to edit the permissions manually.
--
Lumiukko |
|
|
warddrVolunteer tester
Send message
Joined: 7 Feb 08
Posts: 254
ID: 18735
Credit: 24,054,820
RAC: 0
|
http://www.primegrid.com/forum_thread.php?id=1154
Can't seam to get the -t2 to work at all the box just blinks like its opening then closing fast.
With windows I havn't got any problems with this, linux dont seems to support -t2.
Can you paste the text who's in your bat file over here? Did you remember to change the name of the executable to 64bit or 32bit, they doesn't work out of the box.
____________
|
|
|
|
|
|
John,
Check that the sieve file name and the executable file name in the .bat file are correct. You still need to download the sieve file seperately of course.
For instance, the last one I did was using:
fpsieve-x86_64-windows.exe -p501e9 -P504e9 -ipsieve.pfgw -fpfactors_501G-504G.txt -t4
You will need -t2 for dual core and obviously change the parameters.
Best of luck,
Pete.
____________
35 x 2^3587843+1 is prime! |
|
|
|
|
http://www.primegrid.com/forum_thread.php?id=1154
Can't seam to get the -t2 to work at all the box just blinks like its opening then closing fast.
With windows I havn't got any problems with this, linux dont seems to support -t2.
Can you paste the text who's in your bat file over here? Did you remember to change the name of the executable to 64bit or 32bit, they doesn't work out of the box.
./fpsieve -p475e9 -P480e9 -ipsieve.pfgw -fpfactors_475G-480G.txt -y -t4 -c300 |
|
|
|
|
|
John's been pm'ing me trying to help me, I dunno what this is that lennat put out there, I'm trying to try this new fpsieve on my dual core. Just wish it would work right but eh thats coding at its best I suppose, Hope John or someone figures it out or I do myself, I'll be fiddiln with it for a while. Thanks guys Oh the one who asked if I renamed the exe yes did that even did it the other way where you can name the batch to the exe lol but no launch. Anyway off to get sleep or try to fix it who knows
Edit: the last john wanted to know is what ward asked up there and I didn't answer here it is cannot open input file 'psieve_292Ge1.txt'
push any key to continue.....
____________
John M. Johnson "Novex" |
|
|
|
|
|
Tryed that all, didn't work, Johns been emailing me and the guys over on the other thread have been trying to help, its just not working, what am I the first with a dual or something?
____________
John M. Johnson "Novex" |
|
|
|
|
John's been pm'ing me trying to help me, I dunno what this is that lennat put out there, I'm trying to try this new fpsieve on my dual core. Just wish it would work right but eh thats coding at its best I suppose, Hope John or someone figures it out or I do myself, I'll be fiddiln with it for a while. Thanks guys Oh the one who asked if I renamed the exe yes did that even did it the other way where you can name the batch to the exe lol but no launch. Anyway off to get sleep or try to fix it who knows
Do you run in windows on 2 core ?
If so make a fpsieve.bat
Put this in the file. Save it in the folder there you have fpsieve.
fpsieve.exe -p475e9 -P480e9 -ipsieve.pfgw -fpfactors_475G-480G.txt -y -t2
Change -p and -P to min and max your range
Change -fpfactor_minG-maxG
start fpsieve.bat.
Lennart |
|
|
|
|
|
I can't edit the exe only the bat so I don't understand above there?
____________
John M. Johnson "Novex" |
|
|
warddrVolunteer tester
Send message
Joined: 7 Feb 08
Posts: 254
ID: 18735
Credit: 24,054,820
RAC: 0
|
I can't edit the exe only the bat so I don't understand above there?
You have to edit the bat file, and you have to double click that very same file to start the sieve.
____________
|
|
|
JohnHonorary cruncher
Send message
Joined: 21 Feb 06
Posts: 2875
ID: 2449
Credit: 2,681,934
RAC: 0
|
|
Just to clear things up, we are using the actual sieve file name now. Therefore, John, your bat file should look like this:
fpsieve-x86-windows.exe -p456e9 -P457e9 -ipsieve_292Ge1.txt -fpfactors_456G-457G.txt -t2
If you are using the new sieve file, then it should look like this:
fpsieve-x86-windows.exe -p456e9 -P457e9 -ipsieve_337G.txt -fpfactors_456G-457G.txt -t2
If you renamed fpsieve-x86-windows.exe to fpsieve.exe then you bat should look like this:
fpsieve.exe -p456e9 -P457e9 -ipsieve_292Ge1.txt -fpfactors_456G-457G.txt -t2
If you are using the new sieve file, then it should look like this:
fpsieve.exe -p456e9 -P457e9 -ipsieve_337G.txt -fpfactors_456G-457G.txt -t2
Then just double click on the fpsieve.bat file to start your sieve.
____________
|
|
|
|
|
I can't edit the exe only the bat so I don't understand above there?
Download fpsieve.bat from here http://pgllr.mine.nu/sieves/psieve/sievefile
In that file you will see this
ECHO off
fpsieve.exe -p##e9 -P##e9 -ipsieve_##G.txt -fpfactors_##G-##G.txt
EXIT
Lets say you reserve 510G -515G and your sievefile is psieve_337G.txt
Edit the line abowe to
ECHO off
fpsieve.exe -p510e9 -P515e9 -ipsieve_337G.txt -fpfactors_510G-515G.txt -t2
EXIT
You see that i added -t2 at the end to let it use 2 core.
Put this file in the folder you have fpsieve.exe and sievefile
click on fpsieve.bat and it should start.
/Lennart |
|
|
HAmstyVolunteer tester
Send message
Joined: 26 Dec 08
Posts: 132
ID: 33421
Credit: 12,510,712
RAC: 0
          
|
|
i tested factorial sieve with one machine at work (had other workload running during this test (12 threads), this might have influenced the test)
Sample Rates - fsieve_48G sieve file @ 235G
4x Intel Xeon DP X5560, 2,8 GHz, Centos 64bit, 16 threads possible
4 Threads: 33,38k p/s
1 Thread: 11,09k p/s
the "1 Thread"-Result could be influenced by the turbo boost feature
____________
 |
|
|
|
|
|
I just noticed this on my Win Server 2003 x64 machine when running factorial sieve. When I use the 64 bit apps with any or none of the -t flags, I get half the speed compared to running it with the 32 bit app using the same -t flag.
Then i tested the primorial sieve in the same way, since they use the same apps, just changing the sieve file input flag and tweaking the range slightly, and oddly enough when running it that way, using the same -t flags against each other, the 64 bit app has almost a 40% advantage.
Any ideas why the same apps on the same machine would have such a drastic change just changing the sieve input file?
Its running on a Xeon 5030 Dual core w/ HT on, and 2 GB ram.
____________
|
|
|
geoffVolunteer developer
Send message
Joined: 3 Aug 07
Posts: 99
ID: 10427
Credit: 343,437
RAC: 0
|
Its running on a Xeon 5030 Dual core w/ HT on, and 2 GB ram.
My first guess would be that when you ran the 64-bit executable both threads were running as virtual threads on the same physical core, but when you ran the 32-bit executable they ran on seperate physical cores.
The program doesn't know anything about this, it just starts two child threads and leaves it up to the operating system as to which core they run on.
|
|
|
|
|
|
Even when running with a -t4 results in the same effect, which would guarentee virtual cores were in use.
____________
|
|
|
warddrVolunteer tester
Send message
Joined: 7 Feb 08
Posts: 254
ID: 18735
Credit: 24,054,820
RAC: 0
|
Even when running with a -t4 results in the same effect, which would guarentee virtual cores were in use.
Same problem here.
I have a c2q q8300 (2,5Ghz) 4GB ram
With linux everything is running fine
____________
|
|
|
geoffVolunteer developer
Send message
Joined: 3 Aug 07
Posts: 99
ID: 10427
Credit: 343,437
RAC: 0
|
Even when running with a -t4 results in the same effect, which would guarentee virtual cores were in use.
OK thanks. I have been checking the code and I can't find anything wrong yet, but as I can't test this myself (no Win64 machine) it will need some help to find the problem.
I have uploaded fpsieve version 0.2.2 at http://www.geocities.com/g_w_reynolds/testing/, if you could run it and see if the problem persists that could be great. I have compiled the Win64 version with a lower optimisation setting to avoid a possible compiler bug that affects some other code. There are a few other misc changes, but they don't affect performance.
|
|
|
warddrVolunteer tester
Send message
Joined: 7 Feb 08
Posts: 254
ID: 18735
Credit: 24,054,820
RAC: 0
|
Even when running with a -t4 results in the same effect, which would guarentee virtual cores were in use.
OK thanks. I have been checking the code and I can't find anything wrong yet, but as I can't test this myself (no Win64 machine) it will need some help to find the problem.
I have uploaded fpsieve version 0.2.2 at http://www.geocities.com/g_w_reynolds/testing/, if you could run it and see if the problem persists that could be great. I have compiled the Win64 version with a lower optimisation setting to avoid a possible compiler bug that affects some other code. There are a few other misc changes, but they don't affect performance.
C:\Documents and Settings\Administrator\Desktop\pfsieve>fpsieve-x86_64-windows.e
xe -p245e9 -P246e9 -ifsieve_48G.txt -fffactors_245G-246G.txt -t4
fpsieve version 0.2.2 (testing)
Compiled Jun 24 2009 with GCC 4.3.0 20070924 (experimental)
fpconfig.txt: out of range argument blocksize 32
____________
|
|
|
|
|
|
I am still seeing the same slow down using the x64 0.2.2 version compared to 32 bit.
____________
|
|
|
geoffVolunteer developer
Send message
Joined: 3 Aug 07
Posts: 99
ID: 10427
Credit: 343,437
RAC: 0
|
|
OK I have found the bug, it affected the Win64 factorial sieve code in both fsieve and fpsieve. It caused the factorial sieve to run very slowly, but it didn't cause factors to be missed.
It is fixed in fsieve version 1.0.1 and fpsieve version 0.2.3
|
|
|
HonzaVolunteer moderator
Volunteer tester
Project scientist
Send message
Joined: 15 Aug 05
Posts: 1957
ID: 352
Credit: 6,139,270,864
RAC: 2,271,114
                                  
|
OK I have found the bug, it affected the Win64 factorial sieve code in both fsieve and fpsieve. It caused the factorial sieve to run very slowly, but it didn't cause factors to be missed.
Wow, is it just me or is new version really way fatser?
____________
My stats |
|
|
HAmstyVolunteer tester
Send message
Joined: 26 Dec 08
Posts: 132
ID: 33421
Credit: 12,510,712
RAC: 0
|
OK I have found the bug, it affected the Win64 factorial sieve code in both fsieve and fpsieve. It caused the factorial sieve to run very slowly, but it didn't cause factors to be missed.
Wow, is it just me or is new version really way fatser?
on my t2300(win32) i got an increase of 0.5K p/s !!! 25% !!!
no improvements on my opteron(linux64) :/
____________
|
|
|
HonzaVolunteer moderator
Volunteer tester
Project scientist
Send message
Joined: 15 Aug 05
Posts: 1957
ID: 352
Credit: 6,139,270,864
RAC: 2,271,114
|
|
I'm getting like 25K instead of ~11K with -t3 on newer quad with larger cache :-)
____________
My stats |
|
|
HAmstyVolunteer tester
Send message
Joined: 26 Dec 08
Posts: 132
ID: 33421
Credit: 12,510,712
RAC: 0
|
I'm getting like 25K instead of ~11K with -t3 on newer quad with larger cache :-)
Wow! Windows or Linux System?
____________
|
|
|
Scott BrownVolunteer moderator
Project administrator
Volunteer tester
Project scientist
Send message
Joined: 17 Oct 05
Posts: 2392
ID: 1178
Credit: 18,654,341,482
RAC: 6,969,388
                                          
|
|
0.2.3 change over 0.2.1
T8100 on Vista (32-bit) - no change.
Athlon x2 5600+ on Vista (32-bit) - increase of about 0.4k p/s
Athlon x2 4200+ on XP Media Ctr - increase of about 0.2k p/s
*all with -t2 option in use
____________
141941*2^4299438-1 is prime!
|
|
|
|
|
|
WOW! I would say you found it, I went from 5.2K to 7.3K with -t4. Thats a 40% increase. Thanks for looking into that.
If you have a Win64 machine, you NEED to get this downloaded.
____________
|
|
|
Scott BrownVolunteer moderator
Project administrator
Volunteer tester
Project scientist
Send message
Joined: 17 Oct 05
Posts: 2392
ID: 1178
Credit: 18,654,341,482
RAC: 6,969,388
|
0.2.3 change over 0.2.1
T8100 on Vista (32-bit) - no change.
Athlon x2 5600+ on Vista (32-bit) - increase of about 0.4k p/s
Athlon x2 4200+ on XP Media Ctr - increase of about 0.2k p/s
*all with -t2 option in use
Pent D 830 on XP Pro (32-bit) - no change
Pent 4 3.0Ghz HT on Win2k - no change
so, looks to give a little boost to Athlons in 32-bit, but not so much for intel chips (except HAmsty's T2300).
____________
141941*2^4299438-1 is prime!
|
|
|
HonzaVolunteer moderator
Volunteer tester
Project scientist
Send message
Joined: 15 Aug 05
Posts: 1957
ID: 352
Credit: 6,139,270,864
RAC: 2,271,114
|
|
Looks like it gives max boots on large-cache CPU.
It's Win64 affected as was stated before.
____________
My stats |
|
|
HonzaVolunteer moderator
Volunteer tester
Project scientist
Send message
Joined: 15 Aug 05
Posts: 1957
ID: 352
Credit: 6,139,270,864
RAC: 2,271,114
|
|
During last 48 hours, I had 3 restarts on two servers running latest fpsieve -t3 (they are quads).
2 restarts during 2 nights on Oracle server, one restart on domain controller during working hours...not pleasant.
Anyone else experience seldom restarts on machines running fpsieve...due to high optimalization or anything else?
I believe those HP servers are stable, no heating issues (air conditioning in server room), behind UPS etc.
In the meantime, I postponed runnig fpsieve on those machines...
____________
My stats |
|
|
|
|
|
Hi John,
Re: Primorial Sieve
I know that the thread says that manual BOINC Cobblestones are available for this sieve but could you let us know how many per G we would probably get? I'd like to compare it with the PPS and PSP sieving that I do occasionally under BOINC.
Also, how many factors do you think we need to find before a new sieve file is worth while? We are currently up to c. 650G and the last file was built when we were at c. 300G
Thanks,
Pete.
____________
35 x 2^3587843+1 is prime! |
|
|
JohnHonorary cruncher
Send message
Joined: 21 Feb 06
Posts: 2875
ID: 2449
Credit: 2,681,934
RAC: 0
|
Also, how many factors do you think we need to find before a new sieve file is worth while? We are currently up to c. 650G and the last file was built when we were at c. 300G
Thanks,
Pete.
We can do a new one at 650G. Currently I'm not showing all ranges below 650G complete...so we'll wait for the rest. We're expecting to see about 15K factors removed...which will lower the master file by 2%. We'll have to see if that translates into any improvement in sieve speed.
____________
|
|
|
HonzaVolunteer moderator
Volunteer tester
Project scientist
Send message
Joined: 15 Aug 05
Posts: 1957
ID: 352
Credit: 6,139,270,864
RAC: 2,271,114
|
|
New factorial input sieve file (500G) has ~1% speed improvement comparing to 45G file, IIRC.
____________
My stats |
|
|
geoffVolunteer developer
Send message
Joined: 3 Aug 07
Posts: 99
ID: 10427
Credit: 343,437
RAC: 0
|
During last 48 hours, I had 3 restarts on two servers running latest fpsieve -t3 (they are quads).
2 restarts during 2 nights on Oracle server, one restart on domain controller during working hours...not pleasant.
Anyone else experience seldom restarts on machines running fpsieve...due to high optimalization or anything else?
I believe those HP servers are stable, no heating issues (air conditioning in server room), behind UPS etc.
In the meantime, I postponed runnig fpsieve on those machines...
It shouldn't be possible for a user program to cause a reboot, so I think a hardware or operating system fault must be the underlying problem. Of course there could be a bug in fpsieve that triggers the problem, but the operating system should not allow any bug to cause a reboot.
The primorial sieve (and other PrimeGrid sieves) usually access the cache every loop iteration, and so the CPU probably gets regular rest cycles while waiting for data to arrive, and the work is likely to be spread across the CPU.
The 32-bit factorial sieve also has to keep some data in cache because there are not enough registers in 32-bit mode.
But the 64-bit factorial sieve spends long periods in a tight loop with all data held in registers, and so it might stress the CPU differently than the other sieves.
Can you confirm that the reboots only happen with the 64-bit factorial sieve, not with the 32-bit factorial sieve or the 64-bit primorial sieve?
Edit: If the factorial code does cause enough stress to result in problems that don't occur with the other sieves, then there are two ways to address that:
1. Write a torture test program based on the factorial sieve inner loop, to check for CPU faults.
2. Add some no-op instructions to the factorial code to give the CPU a few rest cycles each loop iteration. |
|
|
HonzaVolunteer moderator
Volunteer tester
Project scientist
Send message
Joined: 15 Aug 05
Posts: 1957
ID: 352
Credit: 6,139,270,864
RAC: 2,271,114
|
Edit: If the factorial code does cause enough stress to result in problems that don't occur with the other sieves, then there are two ways to address that: Thanks for reply.
Now it is almost certain is it HP iLO and some HP drivers to blame for seldom restarts. We are upgrading firmware and drivers there days...
Looks like false correlation - I run fpsieve on these 64-bit servers and only those were restarting. 32-bit servers runs PRPNet or nothing BUT also have *different* version of HP drivers and iLO/firmware.
I'll monitor this closely and report back if somethings wrong.
Now I assume fpsieve is just fine.
Thanks again for reply.
____________
My stats |
|
|
|
|
|
I have a question for [SG]Arsenic and vkojlf And any others. The question is have you noticed at all 5% or 10% or any difference in speed with the new psieve file?
____________
John M. Johnson "Novex" |
|
|
vkojlfVolunteer tester
Send message
Joined: 10 Oct 08
Posts: 337
ID: 30361
Credit: 41,827,004
RAC: 0
      
|
I have a question for [SG]Arsenic and vkojlf And any others. The question is have you noticed at all 5% or 10% or any difference in speed with the new psieve file?
No change at all, at least on a i7 920. About the same speed with both 650G and 1360G sieve files.
Linux 64 bit, Intel i7 920 @ 2673 MHz, 64Kp/s with -t8 |
|
|
rogueVolunteer developer
Send message
Joined: 8 Sep 07
Posts: 1256
ID: 12001
Credit: 18,565,548
RAC: 0
|
I have a question for [SG]Arsenic and vkojlf And any others. The question is have you noticed at all 5% or 10% or any difference in speed with the new psieve file?
You shouldn't see any speed increase. The reason is that this sieve does not rely on a discrete log, which is what the fixed base sieves can use. This sieve has to test all values in succession. It cannot skip any values of p because each step must be computed for the step that follows. It is possible to combine some steps and ignore the intermediate product, but because so few values are removed by sieving, the number of steps that could be combined would be extremely low, probably less than 1%.
That being said, it might be possible to do P-1 or ECM on these numbers and remove many more by factoring than by sieving, but I have not done any tests. Hundreds, if not thousands of factoring runs would need to be run (probably using the gwnum version of GMP-ECM) to see how many candidates get removed. The results and time needed per run would need to run though some calculations to see how much "bang for the buck" such factoring provides. It is very possible that at some point P-1 and ECM will remove factors faster than sieving. The challenge will then be to guess which B1 values are optimal. Since sieving will go to about 16 digits, I think that aiming for 20 to 25 digits factors with P-1/ECM would be optimal, but that is an absolute guess. |
|
|
Scott BrownVolunteer moderator
Project administrator
Volunteer tester
Project scientist
Send message
Joined: 17 Oct 05
Posts: 2392
ID: 1178
Credit: 18,654,341,482
RAC: 6,969,388
|
|
I am having trouble with the -t option on 64-bit windows Vista with my i7. It runs with -t2 or -t3, but anything higher freezes up with fpsieve 0.2.4. A bug perhaps? There is no problem if I run 0.2.3.
____________
141941*2^4299438-1 is prime!
|
|
|
|
|
|
Both Factorial and that 27121 have blue goal marks on what the goal is before possible next step insertion into maybe Boinc or the next step at least. What is Primorial's?
____________
John M. Johnson "Novex" |
|
|
JohnHonorary cruncher
Send message
Joined: 21 Feb 06
Posts: 2875
ID: 2449
Credit: 2,681,934
RAC: 0
|
Both Factorial and that 27121 have blue goal marks on what the goal is before possible next step insertion into maybe Boinc or the next step at least. What is Primorial's?
Right now, the Primorial Prime Search will remain in PRPNet...as will the Factorial Prime Search. fpsieve and pfgw have not been integrated into BOINC.
As for goals, they were established at the beginning of the sieves rather arbitrarily thinking that's as far as we'd need to sieve. Since then we have learned and we can sieve substantially deeper.
As for progress goals, I've just updated both Primorial and Factorial. They will progress in 500G increments. At each mark, we'll take new factor timings to see how primality timings match up.
____________
|
|
|
rogueVolunteer developer
Send message
Joined: 8 Sep 07
Posts: 1256
ID: 12001
Credit: 18,565,548
RAC: 0
|
That being said, it might be possible to do P-1 or ECM on these numbers and remove many more by factoring than by sieving, but I have not done any tests. Hundreds, if not thousands of factoring runs would need to be run (probably using the gwnum version of GMP-ECM) to see how many candidates get removed. The results and time needed per run would need to run though some calculations to see how much "bang for the buck" such factoring provides. It is very possible that at some point P-1 and ECM will remove factors faster than sieving. The challenge will then be to guess which B1 values are optimal. Since sieving will go to about 16 digits, I think that aiming for 20 to 25 digits factors with P-1/ECM would be optimal, but that is an absolute guess.
I started with some preliminary testing of this idea. Things aren't looking too favorable, at least for candidates in the 200000 digit range. I estimated that one ECM curve at 11e3 (20 digit target) would take longer than a PRP test for numbers of this size, and that is comparing 64-bit GMP-ECM/gwnum compared with 32-bit PFGW on the same Core2 Duo. |
|
|
|
|
|
This could be a potentially stupid question depending on whether or not the sieve works how I assume it does. I just saw the following line pop up while running factorial sieve:
p=1236331093505, 7.993K p/sec, 1.87x2394MHz CPU, 26.6% done
This value of p is clearly not a prime number, so why are we passing it though the sieve? Any number that it divides is also divisible by 5 so it should have already been removed from the efforts. Wouldn't it make the most sense to see if only prime numbers are divisors of any of the candidates in the sieve file?
____________
 |
|
|
rogueVolunteer developer
Send message
Joined: 8 Sep 07
Posts: 1256
ID: 12001
Credit: 18,565,548
RAC: 0
|
This could be a potentially stupid question depending on whether or not the sieve works how I assume it does. I just saw the following line pop up while running factorial sieve:
p=1236331093505, 7.993K p/sec, 1.87x2394MHz CPU, 26.6% done
This value of p is clearly not a prime number, so why are we passing it though the sieve? Any number that it divides is also divisible by 5 so it should have already been removed from the efforts. Wouldn't it make the most sense to see if only prime numbers are divisors of any of the candidates in the sieve file?
Which program? fpsieve? If so, then I'll leave it to Geoff to address. |
|
|
|
|
|
Yes, fpsieve.
____________
|
|
|
geoffVolunteer developer
Send message
Joined: 3 Aug 07
Posts: 99
ID: 10427
Credit: 343,437
RAC: 0
|
p=1236331093505, 7.993K p/sec, 1.87x2394MHz CPU, 26.6% done
This value of p is clearly not a prime number, so why are we passing it though the sieve?
This is normal for programs based on my 'testing' code (tpsieve, fpsieve etc.), there are two issues:
1. In these programs the p= number displayed doesn't always refer to the prime number being tested, it is just the end of the next block of odd numbers from which the test primes are being sieved. This is because the thread that prints the display message doesn't know which number the other threads are actually testing at any given time, it only knows which block of odd numbers was most recently allocated.
2. In some cases it is faster not to bother ensuring that the test number p is actually prime [by sieving for factors up to sqrt(p)] before testing, but instead just check that it has no factors up to some lower limit. That happens when the --qmax switch is used. (but doesn't apply in this particular case because 5 is a very small factor and would always be checked).
|
|
|
|
|
|
2090G-2091G John M. Johnson "Novex" COMPLETE (11 Factors)
On the user stats page this 90 to 91 makes 64 by my counting the archives. It says it was updated on the 25th of Jan. so Yesterday. Just wanted to make sure everything is ok. If you only count when they get into the archive wait for this one then and at least bring mine up to 63 please. Thanks :)
Still working on the bigger one below it. The 2009-2030G
____________
John M. Johnson "Novex" |
|
|
|
|
2090G-2091G John M. Johnson "Novex" COMPLETE (11 Factors)
On the user stats page this 90 to 91 makes 64 by my counting the archives. It says it was updated on the 25th of Jan. so Yesterday. Just wanted to make sure everything is ok. If you only count when they get into the archive wait for this one then and at least bring mine up to 63 please. Thanks :)
Still working on the bigger one below it. The 2009-2030G
Actually now with 2005G-2009G John M. Johnson "Novex" COMPLETE (44 Factors ) makes mine to be up to 68 please. I emailed you like it says to do as well. :)
____________
John M. Johnson "Novex" |
|
|
|
|
2090G-2091G John M. Johnson "Novex" COMPLETE (11 Factors)
On the user stats page this 90 to 91 makes 64 by my counting the archives. It says it was updated on the 25th of Jan. so Yesterday. Just wanted to make sure everything is ok. If you only count when they get into the archive wait for this one then and at least bring mine up to 63 please. Thanks :)
Still working on the bigger one below it. The 2009-2030G
Actually now with 2005G-2009G John M. Johnson "Novex" COMPLETE (44 Factors ) makes mine to be up to 68 please. I emailed you like it says to do as well. :)
I got a post saying I counted wrong so I copied and pasted each one from the archives and saw the mistake, 1200G-1205G John M. Johnson "Novex" COMPLETE (Factors 113)
Listed twice in the Archive. Needs to be fixed. Along with Menipes 1046G. Thats how I was off. I didn't catch it since I was only counting my name not the fact that it was doubled by spaces in there.. along with Menipes. My apology's.
____________
John M. Johnson "Novex" |
|
|
JohnHonorary cruncher
Send message
Joined: 21 Feb 06
Posts: 2875
ID: 2449
Credit: 2,681,934
RAC: 0
|
saw the mistake, 1200G-1205G John M. Johnson "Novex" COMPLETE (Factors 113)
Listed twice in the Archive. Needs to be fixed. Along with Menipes 1046G.
Duplicates deleted. Thanks for reporting.
____________
|
|
|
|
|
|
Problem when uploading the factors on Factor upload page:
"ffactors_1741G-1742G.txt: FAILED --- Error creating file: Z:\host\sieves\factorial\factors\"ffactors_1741G-1742G.txt
What should I do?
____________
 |
|
|
JohnHonorary cruncher
Send message
Joined: 21 Feb 06
Posts: 2875
ID: 2449
Credit: 2,681,934
RAC: 0
|
Problem when uploading the factors on Factor upload page:
"ffactors_1741G-1742G.txt: FAILED --- Error creating file: Z:\host\sieves\factorial\factors\"ffactors_1741G-1742G.txt
What should I do?
Retry, it should be up now.
____________
|
|
|
|
|
Problem when uploading the factors on Factor upload page:
"ffactors_1741G-1742G.txt: FAILED --- Error creating file: Z:\host\sieves\factorial\factors\"ffactors_1741G-1742G.txt
What should I do?
Retry, it should be up now.
Still nothing, still the same problem...am I doing sth wrong??
____________
|
|
|
|
|
|
John,
Could you let us know if you think it would be worth having a new sieve file for Primorial as we are over 2500G now? Of course, I will happily accept your judgement if you think the 'gain' would be minimal.
Thanks,
Pete.
____________
35 x 2^3587843+1 is prime! |
|
|
JohnHonorary cruncher
Send message
Joined: 21 Feb 06
Posts: 2875
ID: 2449
Credit: 2,681,934
RAC: 0
|
John,
Could you let us know if you think it would be worth having a new sieve file for Primorial as we are over 2500G now? Of course, I will happily accept your judgement if you think the 'gain' would be minimal.
Thanks,
Pete.
We can prepare a new sieve file at 2500G but again, the speed improvement will be negligible. Expect a new one sometime this week.
____________
|
|
|
|
|
New Sieve file!
psieve_2630G is the current sieve file and can be found here: Primorial Sieve file.
Updating the sieve file
Stop fpsieve. Download the new sieve file and unzip. Edit fpsieve.bat -i switch to new sieve file name. Restart fpsieve.
Has anyone noticed any speed increase with this new sieve file? or any difference at all? Thanks so much guys !
____________
John M. Johnson "Novex" |
|
|
|
|
John,
Could you let us know if you think it would be worth having a new sieve file for Primorial as we are over 2500G now? Of course, I will happily accept your judgement if you think the 'gain' would be minimal.
Thanks,
Pete.
We can prepare a new sieve file at 2500G but again, the speed improvement will be negligible. Expect a new one sometime this week.
Now that factorial sieve is also close to finishing up to 2500G, will it also receive a new sieve file in the near future?
____________
|
|
|
JohnHonorary cruncher
Send message
Joined: 21 Feb 06
Posts: 2875
ID: 2449
Credit: 2,681,934
RAC: 0
|
Now that factorial sieve is also close to finishing up to 2500G, will it also receive a new sieve file in the near future?
Yes, we can prepare a new sieve file at 2500G. As with Primorial, the speed improvement will be negligible. Expect a new one sometime this week.
____________
|
|
|
|
|
|
"BiBi" wrote: It's running at 4.296K p/sec, on 0.93x1829MHz, I think I need a faster machine ;)
My 3.99x3000MHz push it to 59Kp/sec. But then, output = rate * time, and I won't have the Core2Extreme machine for too much longer. In total, your machines already create a significantly higher output than my few cores.
____________
|
|
|
valtercVolunteer tester
Send message
Joined: 30 May 07
Posts: 121
ID: 8810
Credit: 20,305,932,457
RAC: 5,733,240
           
|
|
What is the status of the Factorial Sieve Project?
(it seems that I am the only one doing something with it .... just wondering if it is worthy to continue...) |
|
|
VatoVolunteer tester
Send message
Joined: 2 Feb 08
Posts: 851
ID: 18447
Credit: 713,640,206
RAC: 1,642,230
|
|
I don't know at what level the sieve becomes optimal, but I believe it's definitely worthwhile to continue - especially since 94550!-1 hit the jackpot. And now that the GPUs are dominating the PPSE sieve, I'll be bringing my CPUs here (and 27121 and ESP).
PS the sievefiles are missing from the server, so I can't start work here yet...
____________
|
|
|
BiBiVolunteer tester
Send message
Joined: 6 Mar 10
Posts: 151
ID: 56425
Credit: 34,290,031
RAC: 0
        
|
|
How much PSA credit is received by Factorial and Primorial sieving? |
|
|
|
|
|
I don´t know the right numbers, but last time I was computing it, I got about 2700 cr./1G in Factorial Sieving.
I´m not participating in another subprojects, so I can´t tell you the amount of credit award.
____________
|
|
|
BiBiVolunteer tester
Send message
Joined: 6 Mar 10
Posts: 151
ID: 56425
Credit: 34,290,031
RAC: 0
|
|
What is the purpose of the sieve file and who generates them? |
|
|
|
|
What is the purpose of the sieve file and who generates them?
Please correct me if I'm wrong.
Sieving is about removing candidates, a sieve file contains candidates of the form of the particular primes we are searching for. These are generated using the specification of the prime form, in this case Factorial/Primorial. I don't know who in particular generated the original sieve files but I suggest it was Lennart or John.
____________
35 x 2^3587843+1 is prime! |
|
|
BiBiVolunteer tester
Send message
Joined: 6 Mar 10
Posts: 151
ID: 56425
Credit: 34,290,031
RAC: 0
|
|
So the factors that we find need to be removed from the sieve file?
And a further question, will this reduce the computation time of the current sieve work?
|
|
|
|
|
|
yes.. the factor found are removd from the sieve file.
i believe candidiates that have been LLR'd already are removed too. (not sure on that)
and yes it _usually_ make the sieve faster.. but only very small amounts ( under 1%) |
|
|
|
|
|
1% is not that low...imagine factorial sieving 50G - on my CPU, it may take 20 days - but it is X6@3.8GHz running 24/7...1% on a dualcore it can reduce the time about half/one day :-) - it saves energy, too ;-). On 200W system running 24/7 it isn´t a small value.
____________
|
|
|
BiBiVolunteer tester
Send message
Joined: 6 Mar 10
Posts: 151
ID: 56425
Credit: 34,290,031
RAC: 0
|
|
Let's get this sieve file updated each 500G. I move my savings to another bank for 1% (extra interest ;) ) |
|
|
|
|
|
Agreed.
____________
|
|
|
BiBiVolunteer tester
Send message
Joined: 6 Mar 10
Posts: 151
ID: 56425
Credit: 34,290,031
RAC: 0
|
|
Did someone start the update? All the results are gone from the upload location.
And did I notice it correctly, the prp-client is faster in reducing terms than the fpsieve? |
|
|
JohnHonorary cruncher
Send message
Joined: 21 Feb 06
Posts: 2875
ID: 2449
Credit: 2,681,934
RAC: 0
|
Did someone start the update? All the results are gone from the upload location.
And did I notice it correctly, the prp-client is faster in reducing terms than the fpsieve?
Yes, I am working on a new sieve file. However, I suspect that a 1% improvement expectation is much too optimistic. I'll be surprised if any improvement can be measured at all. As for sieving vs. prp testing, sieving is still beneficial for the entire data set. Remember, the data set goes up to 10M. Our current n is about 875K.
____________
|
|
|
|
|
... I suspect that a 1% improvement expectation is much too optimistic. I'll be surprised if any improvement can be measured at all...
I agree, I've never seen any difference with newer sieve files.
____________
35 x 2^3587843+1 is prime! |
|
|
BiBiVolunteer tester
Send message
Joined: 6 Mar 10
Posts: 151
ID: 56425
Credit: 34,290,031
RAC: 0
|
|
Great a new sieve file, I will get it en see how much speed improvement it gets. (If it is even measurable .... ;) )
As for sieving vs. prp testing, sieving is still beneficial for the entire data set. Remember, the data set goes up to 10M. Our current n is about 875K.
I get it. The current N that is tested by prp has lower runtimes than the higher Ns. Is there a way to start pfgw at highest N to see the run time? Or did someone else do that....
The boon for PRP testing is that you will occasionally find a prime and with sieving you increase the chance of finding one. (Assuming a prime is left upto N=10M)
|
|
|
BiBiVolunteer tester
Send message
Joined: 6 Mar 10
Posts: 151
ID: 56425
Credit: 34,290,031
RAC: 0
|
|
Did someone look into creating a CUDA version of the fpsieve? Seeing the success in the other sieves it could be helpful to reduce the factors. |
|
|
|
|
|
i did say "under 1%" as in "less than 1%"
perahps i shoul dhave been less generaous and say "less then 0.1%"
i seem to rememebr the SOB sieve reduced by that amont.. but possible only when an entire K was eliminated.
|
|
|
|
|
|
Where can I download or view the fpsieve.bat or fpsieve.sh?
I can't find them anywhere. |
|
|
HonzaVolunteer moderator
Volunteer tester
Project scientist
Send message
Joined: 15 Aug 05
Posts: 1957
ID: 352
Credit: 6,139,270,864
RAC: 2,271,114
|
Where can I download or view the fpsieve.bat or fpsieve.sh?
I can't find them anywhere.
Before reserving a range, please visit the Welcome thread for instructions.
____________
My stats |
|
|
BiBiVolunteer tester
Send message
Joined: 6 Mar 10
Posts: 151
ID: 56425
Credit: 34,290,031
RAC: 0
|
|
Why are we stopping the primorial sieve at 6000G? |
|
|
|
|
|
I am guessing that 6000G has been determined to be optimal depth for the Sieve. I'm glad that I switched over to it while there were still ranges available, just so that I can say that I helped out.
____________
|
|
|
|
|
Why are we stopping the primorial sieve at 6000G?
As Michael says...
I am guessing that 6000G has been determined to be optimal depth for the Sieve.
There comes a time when the difference in time to perform a prime test and the time it takes to eliminate a candidate from the sieve file is negligible, and seeing that we can't find a prime by sieving we may as well prime test. I guess we are approaching that situation for Primorial?
____________
35 x 2^3587843+1 is prime! |
|
|
|
|
|
sorry for my stupid question, but what is Primorial prime search?
just completed a small range for Primorial sieving, and wanted to check the prime search by itself.
but there's no such subproject, lol
where I can find info?
____________
wbr, Me. Dead J. Dona
 |
|
|
|
|
|
Primality testing for Primorial primes is part of PRPnet. You can find more information in the post I have linked below.
http://www.primegrid.com/forum_thread.php?id=3008&nowrap=true#30720
____________
|
|
|
|
|
|
I believe we just completed the primorial sieving up to 6000G, which is listed as the final goal. Great job everyone! Now to finish factorial sieve also...
____________
|
|
|
GennVolunteer tester
Send message
Joined: 16 Jul 09
Posts: 50
ID: 43504
Credit: 91,204,289
RAC: 0
             
|
I believe we just completed the primorial sieving up to 6000G, which is listed as the final goal. Great job everyone! Now to finish factorial sieve also...
Great work!!
Interesting:
3703071468211 | 10156!+1
That's not very useful factor.
Factorial sieve file contains candidates that were already checked in prpnet (even 103040!-1 (yes, it's prime)). Is it necessarily to sieve candidates that are proven to be composite? |
|
|
|
|
|
I've always wondered this too.
____________
|
|
|
NeoVolunteer tester
Send message
Joined: 28 Oct 10
Posts: 710
ID: 71509
Credit: 91,178,992
RAC: 0
       
|
I believe we just completed the primorial sieving up to 6000G, which is listed as the final goal. Great job everyone! Now to finish factorial sieve also...
Great work!!
Interesting:
3703071468211 | 10156!+1
That's not very useful factor.
Factorial sieve file contains candidates that were already checked in prpnet (even 103040!-1 (yes, it's prime)). Is it necessarily to sieve candidates that are proven to be composite?
I just volunteered for a 150G factorial sieve effort. I produced a factor like the one above and wondered the same thing. Should we not be modifying the "N" to eliminate some of the work??
Neo
|
|
|
NeoVolunteer tester
Send message
Joined: 28 Oct 10
Posts: 710
ID: 71509
Credit: 91,178,992
RAC: 0
|
|
Quick question:
I reserved 4350-4500G for factorial sieving.
I am not producing as many factors as I had expected (I'm only at .4% done with 3 terms eliminated)..but..I just want to make sure I'm entering the command line correctly.
fpsieve-x86_64-windows -p4350e9 -P4500e9 -ifsieve_2530G.txt -fffactors_4350G-4500G.txt -t2
Thanks in advance.
Neo
|
|
|
|
|
|
I can't find the Factorial Sieve software at the suggested address:
http://uwin.mine.nu/software/fsieve/
Can anyone point me a the correct location please?
Peter
____________
35 x 2^3587843+1 is prime! |
|
|
JohnHonorary cruncher
Send message
Joined: 21 Feb 06
Posts: 2875
ID: 2449
Credit: 2,681,934
RAC: 0
|
I can't find the Factorial Sieve software at the suggested address:
http://uwin.mine.nu/software/fsieve/
Can anyone point me a the correct location please?
Peter
You can download it from here: http://uwin.mine.nu/software/fpsieve/
____________
|
|
|
JohnHonorary cruncher
Send message
Joined: 21 Feb 06
Posts: 2875
ID: 2449
Credit: 2,681,934
RAC: 0
|
I am not producing as many factors as I had expected (I'm only at .4% done with 3 terms eliminated)..but..I just want to make sure I'm entering the command line correctly.
fpsieve-x86_64-windows -p4350e9 -P4500e9 -ifsieve_2530G.txt -fffactors_4350G-4500G.txt -t2
Yes, this is correct. I presume you wish to run it across 2 cores.
We are pretty deep into the sieve file. Therefore, we're not expecting many factors. However, the sieve file goes up to n=1M which means we'll be sieving for quite some time.
____________
|
|
|
JohnHonorary cruncher
Send message
Joined: 21 Feb 06
Posts: 2875
ID: 2449
Credit: 2,681,934
RAC: 0
|
Interesting:
3703071468211 | 10156!+1
That's not very useful factor.
Factorial sieve file contains candidates that were already checked in prpnet (even 103040!-1 (yes, it's prime)). Is it necessarily to sieve candidates that are proven to be composite?
I just volunteered for a 150G factorial sieve effort. I produced a factor like the one above and wondered the same thing. Should we not be modifying the "N" to eliminate some of the work??
I was once told by a great sieve programmer that "primality testing doesn't prove that a number is composite, only that it is prime. To prove a number composite, you must find a factor." :)
Semantics, right? Nevertheless, removing the n already primality tested currently has a neglible effect on the sieve speed. Maybe at some point down the road we can remove the lower tests. Until then, they are practically free AND help cut down the double check effort should one start in the future.
____________
|
|
|
rogueVolunteer developer
Send message
Joined: 8 Sep 07
Posts: 1256
ID: 12001
Credit: 18,565,548
RAC: 0
|
|
I have posted http://home.roadrunner.com/~mrodenkirch/fsievecl_1.0.0.zip, an OpenCL version of the factorial sieve software. The sources include a Mac build and the necessary files for a Windows build. (Note that building on Windows might require some tweaks of the included project files.) I have done some testing and it appears to be working correctly, but it still needs more testing.
With this build, I am able to do a range of 1e9 in about 4 hours on my Mac Pro which has an AMD ATI Radeon HD 5870. I haven't run fpsieve, so I don't know how long that takes in comparison on this machine, so any comparisons would be interesting. This version does not require an input file, so it will output factors for factorial prime candidates that have a smaller factor. If you discover that it outputs factors not shown by fpsieve, then let me know. That doesn't necessarily mean a bug in the software. I can run them against the calc program that is included with GMP to verify. 4522267530563 | 34048!-1 is one such factor.
If you choose to test, this will tax your GPU much more than wwwwcl. You should play around with the settings of -b and -s to find what you can live with. The -b switch, like wwwwcl, will limit how large a workgroup you send to the GPU. The -s switch is specific to fsieve. It will make your system more responsive by doing the GPU work in smaller chunks. -s defaults to 1000, which could appear to make your system hang. You will probably want to reduce that, although doing so will impact performance. |
|
|
rogueVolunteer developer
Send message
Joined: 8 Sep 07
Posts: 1256
ID: 12001
Credit: 18,565,548
RAC: 0
|
|
I forgot to mention. If this works as well I hope it works, then I will write a psieve program for primorials. It shouldn't be much more work than fsieve. |
|
|
rogueVolunteer developer
Send message
Joined: 8 Sep 07
Posts: 1256
ID: 12001
Credit: 18,565,548
RAC: 0
|
|
I updated the zip file because of problems with the VS solution that was included. I also forgot that some OS's are case agnostic WRT file names, so I had to rename a source file to address the problem. |
|
|
BiBiVolunteer tester
Send message
Joined: 6 Mar 10
Posts: 151
ID: 56425
Credit: 34,290,031
RAC: 0
|
|
Tried to compile for Nvidia in Linux system but get the following error:
~/fsievecl_1.0.0$ ./fsievecl -v -p 4540G -P 4541G -n 1000000 -N 10000000
fsievecl v1.0.0, a GPU program to find factors numbers of the form n!+/-1
Compiled Oct 26 2012 with GCC 4.6.3
Platform 0 is a NVIDIA Corporation NVIDIA CUDA, version OpenCL 1.1 CUDA 4.2.1
Device 0 is a NVIDIA Corporation GeForce GTX 460
OpenCL Error: Program build failure
in call to clBuildProgram
:1:26: warning: unknown '#pragma OPENCL EXTENSION' - ignored
#pragma OPENCL EXTENSION cl_khr_global_int64_base_atomics : enable
^
:77:20: error: must specify '#pragma OPENCL EXTENSION cl_khr_int64_base_atomics: enable' before using this atomic operation
long old = atom_inc(counter);
^
|
|
|
|
|
Tried to compile for Nvidia in Linux system but get the following error:
~/fsievecl_1.0.0$ ./fsievecl -v -p 4540G -P 4541G -n 1000000 -N 10000000
fsievecl v1.0.0, a GPU program to find factors numbers of the form n!+/-1
Compiled Oct 26 2012 with GCC 4.6.3
Platform 0 is a NVIDIA Corporation NVIDIA CUDA, version OpenCL 1.1 CUDA 4.2.1
Device 0 is a NVIDIA Corporation GeForce GTX 460
OpenCL Error: Program build failure
in call to clBuildProgram
:1:26: warning: unknown '#pragma OPENCL EXTENSION' - ignored
#pragma OPENCL EXTENSION cl_khr_global_int64_base_atomics : enable
^
:77:20: error: must specify '#pragma OPENCL EXTENSION cl_khr_int64_base_atomics: enable' before using this atomic operation
long old = atom_inc(counter);
^
Change cl_khr_global_int64_base_atomics to cl_khr_int64_base_atomics in fsievecl.h. This was a fix that rogue gave me for windows. |
|
|
|
|
|
Lets start some testing.
Download:
32bit
64bit
Requirements for windows users:
-Visual C++ Redistributable for Visual Studio 2012
-OpenCl2.0 or greater (I dont think OpenCl1.1 will work but you can try out)
-AMD or NVIDIA card
my first test with GTX670 and i5-2500k@4Ghz
fsievecl64.exe -p4500495823751 -P4500746583419 -n10 -N1e6 -t1 -b4000 -s2000
4500499942973 | 59736!-1
4500495823751 | 236399!-1
4500505759129 | 286081!+1
4500616334017 | 139076!-1
4500637557179 | 718880!-1
4500730804397 | 641251!-1
4500746583419 | 649090!-1
Primes tested 8605918. Factors found 7. Time 1330.791863 seconds
6.454K p/sec, 0.98 CPU cores
Be aware by setting a large -t and -s value, can crash the client or freezing windows. Can also lagging. Try out different values in a mix.
Post your results here with specs. |
|
|
rogueVolunteer developer
Send message
Joined: 8 Sep 07
Posts: 1256
ID: 12001
Credit: 18,565,548
RAC: 0
|
Be aware by setting a large -t and -s value, can crash the client or freezing windows. Can also lagging. Try out different values in a mix..
Don't forget -b as well. -s is likely to impact your machine the most. |
|
|
RogerVolunteer developer
Volunteer tester
Send message
Joined: 27 Nov 11
Posts: 1138
ID: 120786
Credit: 268,668,824
RAC: 0
        
|
|
I installed the Visual C++ Redistributable for Visual Studio 2012 and AMD APP SDK v2.7 with OpenCL1.2 support.
I am running Windows 8 Release Preview 64 bit and have a HD7970 GPU.
When I try to run Rebirther's 64bit compile, or compile myself with VS2010 Express I get the following:
fsievecl v1.0.0, a GPU program to find factors numbers of the form n!+/-1
OpenCL Error: Program build failure
in call to clBuildProgram
"C:\Users\Roger\AppData\Local\Temp\OCL12BD.tmp.cl", line 1: error: can't
enable all OpenCL extensions or unrecognized OpenCL extension
#pragma OPENCL EXTENSION cl_khr_int64_base_atomics : enable
^
"C:\Users\Roger\AppData\Local\Temp\OCL12BD.tmp.cl", line 61: error: vector
operation requires identical element types
xa_rem = convert_long2(xa_low - (xa_quot * pl));
^
"C:\Users\Roger\AppData\Local\Temp\OCL12BD.tmp.cl", line 77: error: function
"atom_inc" declared implicitly
long old = atom_inc(counter);
^
3 errors detected in the compilation of "C:\Users\Roger\AppData\Local\Temp\OCL12BD.tmp.cl".
Internal error: clc compiler invocation failed.
Where can you get the OpenCL2.0?
____________
 |
|
|
rogueVolunteer developer
Send message
Joined: 8 Sep 07
Posts: 1256
ID: 12001
Credit: 18,565,548
RAC: 0
|
I installed the Visual C++ Redistributable for Visual Studio 2012 and AMD APP SDK v2.7 with OpenCL1.2 support.
I am running Windows 8 Release Preview 64 bit and have a HD7970 GPU.
When I try to run Rebirther's 64bit compile, or compile myself with VS2010 Express I get the following:
fsievecl v1.0.0, a GPU program to find factors numbers of the form n!+/-1
OpenCL Error: Program build failure
in call to clBuildProgram
"C:\Users\Roger\AppData\Local\Temp\OCL12BD.tmp.cl", line 1: error: can't
enable all OpenCL extensions or unrecognized OpenCL extension
#pragma OPENCL EXTENSION cl_khr_int64_base_atomics : enable
^
"C:\Users\Roger\AppData\Local\Temp\OCL12BD.tmp.cl", line 61: error: vector
operation requires identical element types
xa_rem = convert_long2(xa_low - (xa_quot * pl));
^
"C:\Users\Roger\AppData\Local\Temp\OCL12BD.tmp.cl", line 77: error: function
"atom_inc" declared implicitly
long old = atom_inc(counter);
^
3 errors detected in the compilation of "C:\Users\Roger\AppData\Local\Temp\OCL12BD.tmp.cl".
Internal error: clc compiler invocation failed.
Where can you get the OpenCL2.0?
I have to modify the code to remove the OpenCL 1.2 requirements. The impact should be fairly small. |
|
|
|
|
I installed the Visual C++ Redistributable for Visual Studio 2012 and AMD APP SDK v2.7 with OpenCL1.2 support.
I am running Windows 8 Release Preview 64 bit and have a HD7970 GPU.
When I try to run Rebirther's 64bit compile, or compile myself with VS2010 Express I get the following:
fsievecl v1.0.0, a GPU program to find factors numbers of the form n!+/-1
OpenCL Error: Program build failure
in call to clBuildProgram
"C:\Users\Roger\AppData\Local\Temp\OCL12BD.tmp.cl", line 1: error: can't
enable all OpenCL extensions or unrecognized OpenCL extension
#pragma OPENCL EXTENSION cl_khr_int64_base_atomics : enable
^
"C:\Users\Roger\AppData\Local\Temp\OCL12BD.tmp.cl", line 61: error: vector
operation requires identical element types
xa_rem = convert_long2(xa_low - (xa_quot * pl));
^
"C:\Users\Roger\AppData\Local\Temp\OCL12BD.tmp.cl", line 77: error: function
"atom_inc" declared implicitly
long old = atom_inc(counter);
^
3 errors detected in the compilation of "C:\Users\Roger\AppData\Local\Temp\OCL12BD.tmp.cl".
Internal error: clc compiler invocation failed.
Where can you get the OpenCL2.0?
My fault, you need OpenCL1.2 (latest version) what you have, only 1.1 doesnt work yet. If you want to compile with vs2010 you need vc redist 2010. |
|
|
HonzaVolunteer moderator
Volunteer tester
Project scientist
Send message
Joined: 15 Aug 05
Posts: 1957
ID: 352
Credit: 6,139,270,864
RAC: 2,271,114
|
|
My first test with GTX580 and i5-2500k @ stock speed
c:\_PG\FPSieve>fsievecl64.exe -p4500495823751 -P4500746583419 -n10 -N1e6 -t1 -b4000 -s2000
fsievecl v1.0.0, a GPU program to find factors numbers of the form n!+/-1
Sieve started: 4500495823751 <= p < 4500746583419
4500499942973 | 59736!-1
4500495823751 | 236399!-1
4500505759129 | 286081!+1
4500616334017 | 139076!-1 p/sec, 1.00 CPU cores, 47.6% done. ETA 27 Oct 13:51
4500730804397 | 641251!-1
4500637557179 | 718880!-1
4500746583419 | 649090!-1
Sieve complete: 4500495823751 <= p < 4500746583419 8605918 primes tested
Elapsed time: 1048.50 sec. (1.20 init + 968.83 sieve) at 8208 p/sec.
Processor time: 1046.66 sec. (0.51 init + 1046.14 sieve).
Seconds spent in CPU and GPU: 0.05 (cpu), 1046.92 (gpu)
Percent of time spent in CPU vs. GPU: 0.00 (cpu), 1.00 (gpu)
CPU/GPU utilization: 0.00 (cores), 0.00 (devices)
____________
My stats |
|
|
HonzaVolunteer moderator
Volunteer tester
Project scientist
Send message
Joined: 15 Aug 05
Posts: 1957
ID: 352
Credit: 6,139,270,864
RAC: 2,271,114
|
|
And as with WWWW on AMD HD7950, no luck with factorial/primorial.
Driver date is from April 2012 sonot the very recent one...but everything I need is working so I'm reluctant to update.
>fsievecl64.exe -p4500495823751 -P4500746583419 -n10 -N1e6 -t1 -b4000 -s2000
fsievecl v1.0.0, a GPU program to find factors numbers of the form n!+/-1
OpenCL Error: Program build failure
in call to clBuildProgram
C:\Users\ADMINI~1\AppData\Local\Temp\OCLBA60.tmp.cl(1): error: can't enable
all OpenCL extensions or unrecognized OpenCL extension
#pragma OPENCL EXTENSION cl_khr_int64_base_atomics : enable
^
C:\Users\ADMINI~1\AppData\Local\Temp\OCLBA60.tmp.cl(61): error: vector
operation requires identical element types
xa_rem = convert_long2(xa_low - (xa_quot * pl));
^
C:\Users\ADMINI~1\AppData\Local\Temp\OCLBA60.tmp.cl(77): error: function
"atom_inc" declared implicitly
long old = atom_inc(counter);
^
3 errors detected in the compilation of "C:\Users\ADMINI~1\AppData\Local\Temp\OCLBA60.tmp.cl".
Internal error: clc compiler invocation failed.
____________
My stats |
|
|
rogueVolunteer developer
Send message
Joined: 8 Sep 07
Posts: 1256
ID: 12001
Credit: 18,565,548
RAC: 0
|
And as with WWWW on AMD HD7950, no luck with factorial/primorial.
Driver date is from April 2012 sonot the very recent one...but everything I need is working so I'm reluctant to update.
These are two different issues. The problem with fsievecl is that it requires a card that supports OpenCL 1.2. It is not a driver issue. As for wwwwcl, I don't know yet. I thought you were going to get me access to your machine so that I could try to debug it. |
|
|
HonzaVolunteer moderator
Volunteer tester
Project scientist
Send message
Joined: 15 Aug 05
Posts: 1957
ID: 352
Credit: 6,139,270,864
RAC: 2,271,114
|
I thought you were going to get me access to your machine so that I could try to debug it.
I already had MS VS installed in trial version and it expired so it would be of no help...or at least this is what I think.
Otherwise I have no trouble providing remote access, we have already experimented with Rebirther.
____________
My stats |
|
|
|
|
I thought you were going to get me access to your machine so that I could try to debug it.
I already had MS VS installed in trial version and it expired so it would be of no help...or at least this is what I think.
Otherwise I have no trouble providing remote access, we have already experimented with Rebirther.
You can get a vs2012 trial version too. |
|
|
HonzaVolunteer moderator
Volunteer tester
Project scientist
Send message
Joined: 15 Aug 05
Posts: 1957
ID: 352
Credit: 6,139,270,864
RAC: 2,271,114
|
These are two different issues. The problem with fsievecl is that it requires a card that supports OpenCL 1.2. It is not a driver issue.
HD 77xx-79xx models support DirectX 11.1, OpenGL 4.3 and OpenCL 1.2
Updated driver to latest 12.10
____________
My stats |
|
|
HonzaVolunteer moderator
Volunteer tester
Project scientist
Send message
Joined: 15 Aug 05
Posts: 1957
ID: 352
Credit: 6,139,270,864
RAC: 2,271,114
|
You can get a vs2012 trial version too.
OK, downloaded and installed.
PM me so we can arrange time.
____________
My stats |
|
|
rogueVolunteer developer
Send message
Joined: 8 Sep 07
Posts: 1256
ID: 12001
Credit: 18,565,548
RAC: 0
|
These are two different issues. The problem with fsievecl is that it requires a card that supports OpenCL 1.2. It is not a driver issue.
HD 77xx-79xx models support DirectX 11.1, OpenGL 4.3 and OpenCL 1.2
Okay, but something is not allowing OpenCL 1.2 function calls. Are you using AMD APP SDK? Do you have the latest version of it? I wonder if that is the problem. |
|
|
rogueVolunteer developer
Send message
Joined: 8 Sep 07
Posts: 1256
ID: 12001
Credit: 18,565,548
RAC: 0
|
|
BTW, when comparing the final rate (p/sec) of fpsieve to fsievecl I believe that fpsieve counts all n in the range sieved. fsievecl counts only primes in that range. You need to divide the rate given by fpsieve by about 25 to 30 to get the actual number of p it tests each second. |
|
|
BiBiVolunteer tester
Send message
Joined: 6 Mar 10
Posts: 151
ID: 56425
Credit: 34,290,031
RAC: 0
|
|
Ok, I got some problems running this for 1G. The terminal disconnected and this stopped fsievecl. It wasn't able to recocver from the checkpoint. In the second run the application was not able to exit properly. Here's the output of the second run:
~/fsievecl_1.0.0$ ./fsievecl -v -p4540e9 -P4541e9 -n10 -N1e6 -t1 -b32000 -s4000
fsievecl v1.0.0, a GPU program to find factors numbers of the form n!+/-1
Compiled Oct 27 2012 with GCC 4.6.3
Platform 0 is a NVIDIA Corporation NVIDIA CUDA, version OpenCL 1.1 CUDA 4.2.1
Device 0 is a NVIDIA Corporation GeForce GTX 460
workGroupSize = 7168000 = 32000 * 32 * 7 (blocks * workGroupSizeMultiple * deviceComputeUnits)
Running with 1 threads
Allocated memory (prior to sieving): 546 MB in CPU, 546 MB in GPU
Sieve started: 4540000000000 <= p < 4541000000000
4540121050379 | 36573!-1
4540061523619 | 100689!+1
4540153820803 | 206748!+1
4540048564927 | 466671!+1
4540194646751 | 503337!+1
4540064753687 | 699812!-1
4540558922227 | 40689!-1K p/sec, 1.02 CPU cores, 41.8% done. ETA 28 Oct 00:04
4540518744161 | 227136!+1
4540764700489 | 535988!+1
4540775205193 | 545000!-1
4540792791313 | 549106!+1
4540631158247 | 650205!-1
4540453203137 | 679027!-1
4540955099243 | 176838!+1
1 threads didn't stop after 60 seconds
Thread 0: Worker Status: 3 Sieve Status: 3
~/fsievecl_1.0.0$ cat fsieve.log
4540121050379 | 36573!-1
4540061523619 | 100689!+1
4540153820803 | 206748!+1
4540048564927 | 466671!+1
4540194646751 | 503337!+1
4540064753687 | 699812!-1
4540334883167 | 988408!-1
4540558922227 | 40689!-1
4540518744161 | 227136!+1
4540764700489 | 535988!+1
4540775205193 | 545000!-1
4540792791313 | 549106!+1
4540631158247 | 650205!-1
4540453203137 | 679027!-1
4540955099243 | 176838!+1
|
|
|
BiBiVolunteer tester
Send message
Joined: 6 Mar 10
Posts: 151
ID: 56425
Credit: 34,290,031
RAC: 0
|
|
Compiled Oct 27 2012 with GCC 4.6.3
Platform 0 is a NVIDIA Corporation NVIDIA CUDA, version OpenCL 1.1 CUDA 4.2.1
Device 0 is a NVIDIA Corporation GeForce GTX 460
workGroupSize = 896000 = 4000 * 32 * 7 (blocks * workGroupSizeMultiple * deviceComputeUnits)
Running with 1 threads
Allocated memory (prior to sieving): 68 MB in CPU, 68 MB in GPU
Sieve started: 4500495823751 <= p < 4500746583419
fsievecl v1.0.0, a GPU program to find factors numbers of the form n!+/-1
Invalid checkpoint. Starting from beginning of range
4500499942973 | 59736!-1
4500495823751 | 236399!-1
4500505759129 | 286081!+1
4500616334017 | 139076!-1 p/sec, 1.04 CPU cores, 41.6% done. ETA 28 Oct 00:48
4500637557179 | 718880!-1
p=4500652435421, 3.691K p/sec, 1.04 CPU cores, 62.5% done. ETA 28 Oct 00:48
Sieve complete: 4500495823751 <= p < 4500746583419 8605918 primes tested
Elapsed time: 2339.07 sec. (0.02 init + 1941.96 sieve) at 3679 p/sec.
Processor time: 2444.42 sec. (0.02 init + 2444.39 sieve).
Seconds spent in CPU and GPU: 0.10 (cpu), 2338.61 (gpu)
Percent of time spent in CPU vs. GPU: 0.00 (cpu), 1.00 (gpu)
CPU/GPU utilization: 1.27 (cores), 30648.18 (devices)
4500730804397 | 641251!-1
4500746583419 | 649090!-1
|
|
|
rogueVolunteer developer
Send message
Joined: 8 Sep 07
Posts: 1256
ID: 12001
Credit: 18,565,548
RAC: 0
|
Ok, I got some problems running this for 1G. The terminal disconnected and this stopped fsievecl. It wasn't able to recocver from the checkpoint. In the second run the application was not able to exit properly.
I am aware of second problem. I have a simple fix for it. I will have to look at the checkpoint logic. |
|
|
|
|
|
I appear to have a working version for Linux. Compiled on Ubuntu 11.10, GCC 4.6.1. Running on a GTX 570 with a mild o/c (786), and 2600K CPU.
gary@falcon:~/Downloads/fsievecl_1.0.0$ ~/Downloads/fsievecl_1.0.0/fsievecl_1.0.0 -p4500495823751 -P4500746583419 -n10 -N1e6 -t1 -b4000 -s2000
fsievecl v1.0.0, a GPU program to find factors numbers of the form n!+/-1
Sieve started: 4500495823751 <= p < 4500746583419
4500499942973 | 59736!%ld
4500495823751 | 236399!%ld
4500505759129 | 286081!%ld
4500616334017 | 139076!%ldp/sec, 1.03 CPU cores, 44.6% done. ETA 27 Oct 17:23
4500637557179 | 718880!%ld
4500730804397 | 641251!%ld
4500746583419 | 649090!%ld
Sieve complete: 4500495823751 <= p < 4500746583419 8605918 primes tested
Elapsed time: 1088.09 sec. (0.69 init + 949.38 sieve) at 7910 p/sec.
Processor time: 1130.29 sec. (0.40 init + 1129.89 sieve).
Seconds spent in CPU and GPU: 0.05 (cpu), 1087.10 (gpu)
Percent of time spent in CPU vs. GPU: 0.00 (cpu), 1.00 (gpu)
CPU/GPU utilization: 0.00 (cores), 0.00 (devices)
gary@falcon:~/Downloads/fsievecl_1.0.0$
Note "%ld" errors in output text, and overwriting; I didn't try to fix those. I had to make various changes to #include directives in order to compile at all, and basically conjured a makefile "on the fly". I can provide the source code diff's if/when requested.
I just let it run; no stop/restarts or checkpointing.
--Gary |
|
|
rogueVolunteer developer
Send message
Joined: 8 Sep 07
Posts: 1256
ID: 12001
Credit: 18,565,548
RAC: 0
|
Note "%ld" errors in output text, and overwriting; I didn't try to fix those. I had to make various changes to #include directives in order to compile at all, and basically conjured a makefile "on the fly". I can provide the source code diff's if/when requested.
When did you d/l the source? That was fixed. I'll have to fix the \r issue with the status message as the next factor will overwrite that line. Please send me the diffs via e-mail, not PM. |
|
|
HonzaVolunteer moderator
Volunteer tester
Project scientist
Send message
Joined: 15 Aug 05
Posts: 1957
ID: 352
Credit: 6,139,270,864
RAC: 2,271,114
|
Okay, but something is not allowing OpenCL 1.2 function calls. Are you using AMD APP SDK? Do you have the latest version of it? I wonder if that is the problem.
Downloaded and installed latest AMD APP SDK.
http://developer.amd.com/tools/hc/AMDAPPSDK/downloads/Pages/default.aspx
Still the same error.
____________
My stats |
|
|
BiBiVolunteer tester
Send message
Joined: 6 Mar 10
Posts: 151
ID: 56425
Credit: 34,290,031
RAC: 0
|
Note "%ld" errors in output text, and overwriting; I didn't try to fix those. I had to make various changes to #include directives in order to compile at all, and basically conjured a makefile "on the fly". I can provide the source code diff's if/when requested.
When did you d/l the source? That was fixed. I'll have to fix the \r issue with the status message as the next factor will overwrite that line. Please send me the diffs via e-mail, not PM.
I fixed the %ld error, do you need the diffs as well? (I got the newest 1.0.0 zip) |
|
|
|
|
Note "%ld" errors in output text, and overwriting; I didn't try to fix those. I had to make various changes to #include directives in order to compile at all, and basically conjured a makefile "on the fly". I can provide the source code diff's if/when requested.
When did you d/l the source? That was fixed. I'll have to fix the \r issue with the status message as the next factor will overwrite that line. Please send me the diffs via e-mail, not PM.
Downloaded:
wget http://home.roadrunner.com/~mrodenkirch/fsievecl_1.0.0.zip
yesterday (Sat.) afternoon. Will put together the diffs and email later today.
--Gary |
|
|
rogueVolunteer developer
Send message
Joined: 8 Sep 07
Posts: 1256
ID: 12001
Credit: 18,565,548
RAC: 0
|
|
I probably didn't post the source with the %ld fix. In case I missed something, if anyone has made changes to fix bugs or to get it to compile on their platform, please e-mail those changes to me. |
|
|
|
|
I probably didn't post the source with the %ld fix. In case I missed something, if anyone has made changes to fix bugs or to get it to compile on their platform, please e-mail those changes to me.
Sent. |
|
|
|
|
|
fsieve 1.0.1
32bit
64bit
Changelog:
- Removed OpenCL 1.2 requirement.
- Read/write file of terms.
- Status now reports factor removal rate.
- Added new options: -m, -f, -i, -o. |
|
|
rogueVolunteer developer
Send message
Joined: 8 Sep 07
Posts: 1256
ID: 12001
Credit: 18,565,548
RAC: 0
|
|
I believe there is a problem with 1.0.1. Please hold off on testing with it until I get a chance to investigate. |
|
|
rogueVolunteer developer
Send message
Joined: 8 Sep 07
Posts: 1256
ID: 12001
Credit: 18,565,548
RAC: 0
|
|
I've posted v1.0.2 to my website fixing the issues I found in 1.0.1. You will have to wait for rebirther to post a Windows build. |
|
|
|
|
I've posted v1.0.2 to my website fixing the issues I found in 1.0.1. You will have to wait for rebirther to post a Windows build.
I have updated apps, pls use links below but missing changelog. |
|
|
|
|
|
I've given it a go with this command:
fsievecl64.exe -p4500495823751 -P4500746583419 -n10 -N1e6 -t1 -b4000 -s2000
This produces a lot of lag, as expected. Changing to -s1000 already made it a lot better. However, the app does not respond to ctrl+c. I have to double tap ctrl+c in order to have it stop, but that also kills my display driver.
I'm on Win7 with a GTX 570 and i5-2500k, latest updates for both windows and videodrivers.
____________
PrimeGrid Challenge Overall standings --- Last update: From Pi to Paddy (2016)
 |
|
|
rogueVolunteer developer
Send message
Joined: 8 Sep 07
Posts: 1256
ID: 12001
Credit: 18,565,548
RAC: 0
|
I've given it a go with this command:
[code]fsievecl64.exe -p4500495823751 -P4500746583419 -n10 -N1e6 -t1 -b4000 -s2000[/]
This produces a lot of lag, as expected. Changing to -s1000 already made it a lot better. However, the app does not respond to ctrl+c. I have to double tap ctrl+c in order to have it stop, but that also kills my display driver.
I'm on Win7 with a GTX 570 and i5-2500k, latest updates for both windows and videodrivers.
The problem with ^C is that the preferred behavior of the app is to complete testing of a block of primes then exit and say that it completed testing all primes below a given limit. If I modify the program to accept ^C to exit while processing primes, then it is nearly impossible to guarantee that all primes below the current prime were tested to completion. After hitting ^C, you might need to wait for a couple of minutes for it to complete the current batch of primes.
For testing purposes I suggesting setting the -b value much lower, maybe as low as 100, then scale up -s to improve throughput while trying to avoid slowing down your system too much. When you hit ^C, it will take less time to complete the block of primes and exit. It might still take a number of seconds, but it shouldn't take minutes. |
|
|
|
|
I've given it a go with this command:
[code]fsievecl64.exe -p4500495823751 -P4500746583419 -n10 -N1e6 -t1 -b4000 -s2000[/]
This produces a lot of lag, as expected. Changing to -s1000 already made it a lot better. However, the app does not respond to ctrl+c. I have to double tap ctrl+c in order to have it stop, but that also kills my display driver.
I'm on Win7 with a GTX 570 and i5-2500k, latest updates for both windows and videodrivers.
The problem with ^C is that the preferred behavior of the app is to complete testing of a block of primes then exit and say that it completed testing all primes below a given limit. If I modify the program to accept ^C to exit while processing primes, then it is nearly impossible to guarantee that all primes below the current prime were tested to completion. After hitting ^C, you might need to wait for a couple of minutes for it to complete the current batch of primes.
For testing purposes I suggesting setting the -b value much lower, maybe as low as 100, then scale up -s to improve throughput while trying to avoid slowing down your system too much. When you hit ^C, it will take less time to complete the block of primes and exit. It might still take a number of seconds, but it shouldn't take minutes.
Ah, that actually makes sense! :)
From a usability perspective: is it possible to add a notice along the lines of "Caught ^C. Program will terminate after the current block of candidates has finished testing /-\|-\-/|"?
____________
PrimeGrid Challenge Overall standings --- Last update: From Pi to Paddy (2016)
|
|
|
|
|
|
So, I did
fsievecl64.exe -p4500495823751 -P4500746583419 -n10 -N1e6 -t1 -b1 -s10000
(b1 after b100 didn't seem to have an effect)
Hit ^c once and two minutes in GPU load is still >90%.
I hit ^c once more and it instantly quit.
I tried a couple of times more but apparently if you don't hit the sweet spot the second time your driver still goes down quicker than Felix Baumgartner.
____________
PrimeGrid Challenge Overall standings --- Last update: From Pi to Paddy (2016)
|
|
|
rogueVolunteer developer
Send message
Joined: 8 Sep 07
Posts: 1256
ID: 12001
Credit: 18,565,548
RAC: 0
|
So, I did
fsievecl64.exe -p4500495823751 -P4500746583419 -n10 -N1e6 -t1 -b1 -s10000
(b1 after b100 didn't seem to have an effect)
Hit ^c once and two minutes in GPU load is still >90%.
I hit ^c once more and it instantly quit.
I tried a couple of times more but apparently if you don't hit the sweet spot the second time your driver still goes down quicker than Felix Baumgartner.
The software is designed to instantly quit after ^C is hit a second time. I can't speak for how the driver behaves when that happens.
Due to the nature of this sieve, I'll modify the code to output some information that will help you determine how close it is to completion.
This sieve is not well-suited for slower GPUs. |
|
|
valtercVolunteer tester
Send message
Joined: 30 May 07
Posts: 121
ID: 8810
Credit: 20,305,932,457
RAC: 5,733,240
|
|
just tried and got this (W7Ux64, HD6950, Catalyst 12.1, openCL 1.1).
Driver too old?
>fsievecl64.exe -p4500495823751 -P4500746583419 -n
10 -N1e6 -t1 -b4000 -s2000
fsievecl v1.0.2, a GPU program to find factors numbers of the form n!+/-1
OpenCL Error: Program build failure
in call to clBuildProgram
C:\Users\valter\AppData\Local\Temp\OCLE0F0.tmp.cl(1): error: can't enable
all
OpenCL extensions or unrecognized OpenCL extension
#pragma OPENCL EXTENSION cl_khr_int32_base_atomics : enable
^
...OCLE0F0.tmp.cl(53): error: vector operation
requires identical element types
xa_rem = xa_rem * n_mul;
^
rnal error: clc compiler invocation failed. |
|
|
|
|
just tried and got this (W7Ux64, HD6950, Catalyst 12.1, openCL 1.1).
Driver too old?
>fsievecl64.exe -p4500495823751 -P4500746583419 -n
10 -N1e6 -t1 -b4000 -s2000
fsievecl v1.0.2, a GPU program to find factors numbers of the form n!+/-1
OpenCL Error: Program build failure
in call to clBuildProgram
C:\Users\valter\AppData\Local\Temp\OCLE0F0.tmp.cl(1): error: can't enable
all
OpenCL extensions or unrecognized OpenCL extension
#pragma OPENCL EXTENSION cl_khr_int32_base_atomics : enable
^
...OCLE0F0.tmp.cl(53): error: vector operation
requires identical element types
xa_rem = xa_rem * n_mul;
^
rnal error: clc compiler invocation failed.
I dont know if you need .net4 support but if you can try 12.10 do it. |
|
|
HonzaVolunteer moderator
Volunteer tester
Project scientist
Send message
Joined: 15 Aug 05
Posts: 1957
ID: 352
Credit: 6,139,270,864
RAC: 2,271,114
|
|
It is a tinny bit better with 1.02 - only 2 fatal errors instead of 3
Win 2008R2 x64, HD7950, latest Calatyst, latest AMD APP SDK.
Also .NET Framework 4 Multi-targeting pack (WTF is that?), even .NET 4.5 thanks to VS 2012. Visual C++ 2005, 2008, 2010, 2012 both x64 and x86.
>fsievecl64.exe -p4500495823751 -P4500746583419 -n10 -N1e6
fsievecl v1.0.2, a GPU program to find factors numbers of the form n!+/-1
OpenCL Error: Program build failure
in call to clBuildProgram
C:\Users\ADMINI~1\AppData\Local\Temp\OCLE7D2.tmp.cl(1): error: can't enable
all OpenCL extensions or unrecognized OpenCL extension
#pragma OPENCL EXTENSION cl_khr_int32_base_atomics : enable
^
C:\Users\ADMINI~1\AppData\Local\Temp\OCLE7D2.tmp.cl(53): error: vector
operation requires identical element types
xa_rem = xa_rem * n_mul;
^
2 errors detected in the compilation of "C:\Users\ADMINI~1\AppData\Local\Temp\OCLE7D2.tmp.cl".
Internal error: clc compiler invocation failed.
____________
My stats |
|
|
rogueVolunteer developer
Send message
Joined: 8 Sep 07
Posts: 1256
ID: 12001
Credit: 18,565,548
RAC: 0
|
C:\Users\ADMINI~1\AppData\Local\Temp\OCLE7D2.tmp.cl(53): error: vector
operation requires identical element types
xa_rem = xa_rem * n_mul;
I don't understand why this fails for you, but nobody else. OpenCL is complaining because n_mul is unsigned and xa_rem is signed. This is most likely triggering the second error.
I'll patch this. |
|
|
Scott BrownVolunteer moderator
Project administrator
Volunteer tester
Project scientist
Send message
Joined: 17 Oct 05
Posts: 2392
ID: 1178
Credit: 18,654,341,482
RAC: 6,969,388
|
|
The issue may be Windows 2008. It is not an officially supported OS for AMD's OpenCL drivers, though it does appear to work for some with some very methodical installation/update processes. See here for example.
|
|
|
HonzaVolunteer moderator
Volunteer tester
Project scientist
Send message
Joined: 15 Aug 05
Posts: 1957
ID: 352
Credit: 6,139,270,864
RAC: 2,271,114
|
The issue may be Windows 2008. It is not an officially supported OS for AMD's OpenCL drivers, though it does appear to work for some with some very methodical installation/update processes. See here for example.
Yeah, could be.
On the other hand, Windows 2008 R2 is very close to Windows 7.
I haven't found anything that works under Win 7 and doesn't on Win 2008 R2 - not counting artificial sw limitation complaing about "need bussiness edition".
Even such hw as Logitech driving wheel works fine.
GPU-Z confirm OpenCL availability: OpenCL 1.1 AMD APP (844.5)
EDIT: Interestingly, Google found my older post about wwwwcl - also having troubles running WallSunSun
wwwwcl v2.1.5, a GPU program to search for Wieferich and WallSunSun primes
Platform 0 is an Advanced Micro Devices, Inc. AMD Accelerated Parallel Processing, version OpenCL 1.2 AMD-APP (923.1)
But running wwwwcl now with latest Catalyst confirms OpenCL 1.1 - same as GPU-Z
EDIT2: Uninstalled Catalyst, uninstalled drivers etc. and fresh install again.
Can't get OpenCL 1.1 with SDK 2.7. And here is why - latest isn't always the best.
As of SDK 2.7, the AMD APP SDK GPU runtime is not supported. To get the APP GPU runtime file, install the Catalyst driver. The SDK 2.7 runtime supports only CPU hardware; it is installed under Program Files.
SDK 2.6 has older OpenCL for GPU, a year old.
Latest Catalyst has OpenCL package from May 2012. It also has .NET 4 included.
It looks like latest Catalyst is all you need to run fsievecl64 or wwwwcl64 - in theory.
Still a bit confused how I managed to get newer OpenCL half a year ago...
____________
My stats |
|
|
valtercVolunteer tester
Send message
Joined: 30 May 07
Posts: 121
ID: 8810
Credit: 20,305,932,457
RAC: 5,733,240
|
C:\Users\ADMINI~1\AppData\Local\Temp\OCLE7D2.tmp.cl(53): error: vector
operation requires identical element types
xa_rem = xa_rem * n_mul;
I don't understand why this fails for you, but nobody else. OpenCL is complaining because n_mul is unsigned and xa_rem is signed. This is most likely triggering the second error.
I'll patch this.
well, I got the same errors with a different OS, see a few posts before... (Message 58782) |
|
|
rogueVolunteer developer
Send message
Joined: 8 Sep 07
Posts: 1256
ID: 12001
Credit: 18,565,548
RAC: 0
|
|
I've posted 1.0.3 to my website. I think I've addressed the most recent issue with running the code on older OSes. I added logic so that if one hits ^C, the threads will report their status periodically so that you can determine how close the program is to completion. |
|
|
|
|
|
1.0.3
32bit
64bit
Changelog:
- Added new state WS_INTERRUPTED so that threads can report on status after ^C is hit. |
|
|
|
|
|
Minor grammar bug in the ^C
CTRL-C accepted. Please wait for threads to completed.
That last d shouldn't be there.
Also, I don't see the threads reporting in. All I see is the regular
p=4500529393583, 6.300K p/sec, 3 factors found 0 secs/factor, 0.96 CPU cores, 13.4% done. ETA 01 Nov 18:05 and I'm assuming it doesn't take ~20 minutes to complete a thread.
____________
PrimeGrid Challenge Overall standings --- Last update: From Pi to Paddy (2016)
|
|
|
|
|
|
For a short test you can use:
fsievecl64.exe -p4500495823751 -P4500499942973 -n10 -N1e6 -t1 -b4000 -s2000
Log file looks ok. |
|
|
rogueVolunteer developer
Send message
Joined: 8 Sep 07
Posts: 1256
ID: 12001
Credit: 18,565,548
RAC: 0
|
Minor grammar bug in the ^C
CTRL-C accepted. Please wait for threads to completed.
That last d shouldn't be there.
Also, I don't see the threads reporting in. All I see is the regular
p=4500529393583, 6.300K p/sec, 3 factors found 0 secs/factor, 0.96 CPU cores, 13.4% done. ETA 01 Nov 18:05 and I'm assuming it doesn't take ~20 minutes to complete a thread.
The ^C logic isn't fully tested. If I have a chance, I'll look at it tonight. It should report based upon the size of -s, .e.g if -s is set to 1000, it should report after each 1000 n is tested.
It could take 20 minutes depending upon the settings of -b, -t, -s, plus the speed of your GPU. |
|
|
HonzaVolunteer moderator
Volunteer tester
Project scientist
Send message
Joined: 15 Aug 05
Posts: 1957
ID: 352
Credit: 6,139,270,864
RAC: 2,271,114
|
|
Yet again a bit better with 1.03 - just 1 fatal error on HD7950
c:\_PG\FPSieve>wwwwcl64.exe -l
List of available platforms and devices
Platform 0 is a Advanced Micro Devices, Inc. AMD Accelerated Parallel Processing, version OpenCL 1.1 AMD-APP (844.5)
Device 0 is a Advanced Micro Devices, Inc. Tahiti
c:\_PG\FPSieve>fsievecl64.exe -p4500495823751 -P4500746583419 -n10 -N1e6
fsievecl v1.0.3, a GPU program to find factors numbers of the form n!+/-1
OpenCL Error: Program build failure
in call to clBuildProgram
C:\Users\ADMINI~1\AppData\Local\Temp\OCL5F67.tmp.cl(1): error: can't enable
all OpenCL extensions or unrecognized OpenCL extension
#pragma OPENCL EXTENSION cl_khr_int32_base_atomics : enable
^
1 error detected in the compilation of "C:\Users\ADMINI~1\AppData\Local\Temp\OCL5F67.tmp.cl".
Internal error: clc compiler invocation failed.
____________
My stats |
|
|
rogueVolunteer developer
Send message
Joined: 8 Sep 07
Posts: 1256
ID: 12001
Credit: 18,565,548
RAC: 0
|
Yet again a bit better with 1.03 - just 1 fatal error on HD7950
c:\_PG\FPSieve>wwwwcl64.exe -l
List of available platforms and devices
Platform 0 is a Advanced Micro Devices, Inc. AMD Accelerated Parallel Processing, version OpenCL 1.1 AMD-APP (844.5)
Device 0 is a Advanced Micro Devices, Inc. Tahiti
c:\_PG\FPSieve>fsievecl64.exe -p4500495823751 -P4500746583419 -n10 -N1e6
fsievecl v1.0.3, a GPU program to find factors numbers of the form n!+/-1
OpenCL Error: Program build failure
in call to clBuildProgram
C:\Users\ADMINI~1\AppData\Local\Temp\OCL5F67.tmp.cl(1): error: can't enable
all OpenCL extensions or unrecognized OpenCL extension
#pragma OPENCL EXTENSION cl_khr_int32_base_atomics : enable
^
1 error detected in the compilation of "C:\Users\ADMINI~1\AppData\Local\Temp\OCL5F67.tmp.cl".
Internal error: clc compiler invocation failed.
I did not expect that. I need to investigate further. I have a Tahiti at home, but it is so slow I haven't used it for anything. I'll try to get code to run on that before I post updated source. |
|
|
rogueVolunteer developer
Send message
Joined: 8 Sep 07
Posts: 1256
ID: 12001
Credit: 18,565,548
RAC: 0
|
|
I don't think that:
#pragma OPENCL EXTENSION cl_khr_int32_base_atomics : enable
is necessary in OpenCL 1.1. Rebirther, can you remove that line from fseivecl.h and rebuild? |
|
|
|
|
I don't think that:
#pragma OPENCL EXTENSION cl_khr_int32_base_atomics : enable
is necessary in OpenCL 1.1. Rebirther, can you remove that line from fseivecl.h and rebuild?
Done, download below. |
|
|
HonzaVolunteer moderator
Volunteer tester
Project scientist
Send message
Joined: 15 Aug 05
Posts: 1957
ID: 352
Credit: 6,139,270,864
RAC: 2,271,114
|
|
Guys, now it is running on AMD, well done.
>fsievecl64.exe -p4500495823751 -P4500746583419 -n10 -N1e6 -t1 -b250 -s1000
fsievecl v1.0.3, a GPU program to find factors numbers of the form n!+/-1
Invalid checkpoint. Starting from beginning of range
Sieve started: (cmdline) 4500495823751 <= p < 4500746583419 with 1999982 terms
4500499942973 | 59736!-1
4500495823751 | 236399!-1
4500505759129 | 286081!+1
p=4500600237827, 4.460K p/sec, 3 factors found 0 secs/factor, 0.02 CPU cores, 41.6% done. ETA 02 Nov 21:05
4500616334017 | 139076!-1
p=4500626338097, 4.463K p/sec, 4 factors found 0 secs/factor, 0.02 CPU cores, 52.0% done. ETA 02 Nov 21:05
4500637557179 | 718880!-1
p=4500704674769, 4.477K p/sec, 5 factors found 0 secs/factor, 0.02 CPU cores, 83.3% done. ETA 02 Nov 21:05
4500730804397 | 641251!-1
4500746583419 | 649090!-1
Sieve complete: 4500495823751 <= p < 4500746583419 8605918 primes tested
Elapsed time: 1933.85 sec. (0.31 init + 1804.29 sieve) at 4450 p/sec.
Processor time: 40.98 sec. (0.31 init + 40.67 sieve).
Seconds spent in CPU and GPU: 0.09 (cpu), 1932.74 (gpu)
Percent of time spent in CPU vs. GPU: 0.00 (cpu), 1.00 (gpu)
CPU/GPU utilization: 0.00 (cores), 0.00 (devices)
____________
My stats |
|
|
HonzaVolunteer moderator
Volunteer tester
Project scientist
Send message
Joined: 15 Aug 05
Posts: 1957
ID: 352
Credit: 6,139,270,864
RAC: 2,271,114
|
|
If those numbers are correct, HD7950 can do 1G in ~2 hours.
>fsievecl64.exe -p4523G -P4524G -n10 -N1e6 -t1 -b250 -s1000
fsievecl v1.0.3, a GPU program to find factors numbers of the form n!+/-1
Sieve started: (cmdline) 4523000000000 <= p < 4524000000000 with 1999982 terms
p=4523026134101, 4.449K p/sec, 0 factors found 0 secs/factor, 0.02 CPU cores, 2.6% done. ETA 03 Nov 00:03
____________
My stats |
|
|
HonzaVolunteer moderator
Volunteer tester
Project scientist
Send message
Joined: 15 Aug 05
Posts: 1957
ID: 352
Credit: 6,139,270,864
RAC: 2,271,114
|
|
OK, I let it run 4523G-4524G and compared with already uplaoded factor using CPU.
4523183953231 | 799897!-1
4523251874171 | 221356!-1
4523280465751 | 169!-1
4523483705159 | 447082!+1
4523690446619 | 321538!-1
4523713207253 | 670369!-1
4523849082821 | 674493!-1
4523872790453 | 435980!-1
4523950780633 | 726428!-1
4523186406983 | 590816!+1
4523183953231 | 799897!-1
4523202438569 | 938028!+1
4523251874171 | 221356!-1
4523280465751 | 169!-1
4523369176033 | 473101!+1
4523425959511 | 933437!-1
4523483705159 | 447082!+1
4523514281899 | 239045!+1
4523686856287 | 231566!+1
4523690446619 | 321538!-1
4523716181599 | 19986!-1
4523713207253 | 670369!-1
4523849082821 | 674493!-1
4523872790453 | 435980!-1
4523950780633 | 726428!-1
____________
My stats |
|
|
HonzaVolunteer moderator
Volunteer tester
Project scientist
Send message
Joined: 15 Aug 05
Posts: 1957
ID: 352
Credit: 6,139,270,864
RAC: 2,271,114
|
