| Author |
Message |
|---|
JimB Honorary cruncher
Send message
Joined: 4 Aug 11
Posts: 918
ID: 107307
Credit: 977,945,376
RAC: 0
                     
|
|
There's a possibility that we'll get Primorial and Factorial primality testing moved from PRPNet to PrimeGrid some time in the next few years. If so, we're going to need a lot more sieving on those. So we've reopened sieving on both of those projects and they'll stay open for the foreseeable future. |
|
|
 Dave
Send message
Joined: 13 Feb 12
Posts: 3171
ID: 130544
Credit: 2,233,021,669
RAC: 614,280
                        
|
|
Nice. |
|
|
HonzaVolunteer moderator
Volunteer tester
Project scientist
Send message
Joined: 15 Aug 05
Posts: 1949
ID: 352
Credit: 6,010,546,983
RAC: 1,514,199
                               
|
|
Jim, do we have - at least preliminary - estimate of how deep we are going to sieve this time?
____________
My stats |
|
|
GDBSend message
Joined: 15 Nov 11
Posts: 284
ID: 119185
Credit: 3,928,395,725
RAC: 2,134,632
         
|
|
I'm having problems downloading the Factorial sieve text file. The file downloaded by my chrome browser doesn't look like the text file that gets saved. The sieve file displayed in browser is readable, but sieve file saved doesn't have CRLFs. When I try to run PFSIEVE, I get "Cannot open input file ". Any ideas on how to resolve this? |
|
|
Michael GoetzVolunteer moderator
Project administrator
Send message
Joined: 21 Jan 10
Posts: 13954
ID: 53948
Credit: 392,369,445
RAC: 169,843
             
|
I'm having problems downloading the Factorial sieve text file. The file downloaded by my chrome browser doesn't look like the text file that gets saved. The sieve file displayed in browser is readable, but sieve file saved doesn't have CRLFs. When I try to run PFSIEVE, I get "Cannot open input file ". Any ideas on how to resolve this?
The file almost certainly has Unix-style "newline" line endings rather than Windows-ish "CRLF" line terminations. That's fine; the sieve can read it just fine. If you try to view the file in Notepad, you won't see line breaks. If you use more intelligent editors, you will.
The problem with the sieve not opening the file almost certainly is caused by either an incorrect command line, the sieve file not being in the correct directory, or the sieve file not being named correctly. There's also an outside chance that file permission are the culprit. The most likely cause of the problem is the command line.
Want to post both a DIR listing and the command line you're using?
____________
My lucky number is 75898524288+1 |
|
|
JimBHonorary cruncher
Send message
Joined: 4 Aug 11
Posts: 918
ID: 107307
Credit: 977,945,376
RAC: 0
|
|
Honza wrote: Jim, do we have - at least preliminary - estimate of how deep we are going to sieve this time?
Not at this time. The old numbers were based on how much work was being done on PRPNet. Once this moves to BOINC with a new badge for each project, we're going to be testing candidates orders of magnitude faster than PRPNet. Real Life is rather busy right now, but I'll look into those numbers soon. |
|
|
|
|
Once this moves to BOINC with a new badge for each project, we're going to be testing candidates orders of magnitude faster than PRPNet.
Nah, nobody is going to prefer that form of numbers. After all, it is just a silly math proof they misunderstood in school. It is not as amazing as k*2^n ± 1. /JeppeSN |
|
|
HonzaVolunteer moderator
Volunteer tester
Project scientist
Send message
Joined: 15 Aug 05
Posts: 1949
ID: 352
Credit: 6,010,546,983
RAC: 1,514,199
|
Once this moves to BOINC with a new badge for each project, we're going to be testing candidates orders of magnitude faster than PRPNet.
Both Factorial and Primorial will become most complete and reliable search, thanks to double-checking. Looking forward to it.
Will there be full double-check from the beginning?
Also wondering: according to fsieve_7T.txt, there are ~771k candidates left out of 2M (for n<1M both +1 and -1 form). With leading edge around 246k, there are ~600k candidates left in uncharted territory.
Would that last a year once started on BOINC? Yes, tasks are becoming longer and longer...
Would additional sieving shrink candidate's number to 500k?
Once Factorial and Primorial seaches are running on BOINC, will there be 1M<n<2M or <5M sieving?
No need to answer all, just really looking forward to this :-)
Older servers with older CPUs are good for this sieving where no AVX/AVX2 is of any benefit and they are struggling return LLR task first.
____________
My stats |
|
|
Dave
Send message
Joined: 13 Feb 12
Posts: 3171
ID: 130544
Credit: 2,233,021,669
RAC: 614,280
|
|
Sounds a good alternative to GCW sv for slow machines therefore. Maybe individuals will look towards parking GCW sv at a badge level & building these new projects up when they happen. |
|
|
GDBSend message
Joined: 15 Nov 11
Posts: 284
ID: 119185
Credit: 3,928,395,725
RAC: 2,134,632
|
I'm having problems downloading the Factorial sieve text file. The file downloaded by my chrome browser doesn't look like the text file that gets saved. The sieve file displayed in browser is readable, but sieve file saved doesn't have CRLFs. When I try to run PFSIEVE, I get "Cannot open input file ". Any ideas on how to resolve this?
The file almost certainly has Unix-style "newline" line endings rather than Windows-ish "CRLF" line terminations. That's fine; the sieve can read it just fine. If you try to view the file in Notepad, you won't see line breaks. If you use more intelligent editors, you will.
The problem with the sieve not opening the file almost certainly is caused by either an incorrect command line, the sieve file not being in the correct directory, or the sieve file not being named correctly. There's also an outside chance that file permission are the culprit. The most likely cause of the problem is the command line.
Want to post both a DIR listing and the command line you're using?
I normally use Windows Powershell for manual sieving. After the problems I had above, I tried CMD. Works OK with CMD.
Thanks!
|
|
|
JimBHonorary cruncher
Send message
Joined: 4 Aug 11
Posts: 918
ID: 107307
Credit: 977,945,376
RAC: 0
|
|
I've just temporarily shut off new reservations for Primorial. The early sieving was done so long ago, that I've never seen it. I'm going to do some test sieving on early ranges. If any candidates are removed it means that sieving had problems. Every other kind of sieving I've taken over here also had problems, just so you know. I need up to a week to figure out what's going on. Sorry, everyone. Had I known it was that screwed up I wouldn't have opened sieving or at least I'd have done it differently.
JimB |
|
|
dh1sajVolunteer tester
Send message
Joined: 13 Jul 08
Posts: 49
ID: 25532
Credit: 3,632,409,085
RAC: 53,259
                 
|
|
Just trying to run Factorial on nVidia GPU hosted on Win10 machine with fsievecl64.exe
Running "-l" returns:
List of available platforms and devices
Platform 0 is a Intel(R) Corporation Intel(R) OpenCL, version OpenCL 1.2
Device 0 is a Intel(R) Corporation Intel(R) HD Graphics 4600
Platform 1 is a NVIDIA Corporation NVIDIA CUDA, version OpenCL 1.2 CUDA 9.1.75
Device 0 is a NVIDIA Corporation GeForce GTX 980
OK so far.
But then I use "-f1" or "-f1 -d0" to start a sieve
but whatever I try, once started a sieve, all I get is:
Platform 0 is an Intel(R) Corporation Intel(R) OpenCL, version OpenCL 1.2
Device 0 is an Intel(R) Corporation Intel(R) HD Graphics 4600
The sieve seems running well, but just on the HD4600 (Platform==0).
So it obviously does not like my GTX980 (Platform==1) but uses the onboard GPU instead.
What am I doing wrong?
Any hint how to get it run on the nVidia?
Jochen
|
|
|
|
|
|
Its just a guess but maybe try -d1 instead of -d0?
Edit:
Ya never mind, it doesn't work. I've tried different combinations -f0 -d1, -f1 -d1, etc. it just doesn't seem to find the Nvidia device on my computer either. The only other solution I can think of is to disable the integrated graphics, but then what's the point of having a dedicated GPU for BOINC tasks? |
|
|
JimBHonorary cruncher
Send message
Joined: 4 Aug 11
Posts: 918
ID: 107307
Credit: 977,945,376
RAC: 0
|
|
JimB wrote: I've just temporarily shut off new reservations for Primorial. The early sieving was done so long ago, that I've never seen it. I'm going to do some test sieving on early ranges. If any candidates are removed it means that sieving had problems. Every other kind of sieving I've taken over here also had problems, just so you know. I need up to a week to figure out what's going on. Sorry, everyone. Had I known it was that screwed up I wouldn't have opened sieving or at least I'd have done it differently.
Upon resieving for p=0-1P I found 30 candidates that should have been removed the first time through. That should not have happened. I'm now resieving p=1P-10P which will finish tomorrow (Sunday). If this early sieving (which was all done by Lennart) can't be trusted, there might be reason to redo primorial sieving from p=10G onwards. Also, primorial was only sieved to n=10M while PFGW is capable of testing up to n=20M. So if sieving is redone it will be with a larger n range. In other words half of the sieving will be new. I'll let everyone know when we decide what has to be done.
|
|
|
|
|
|
After some trial and error and loading of MS Visual C++ 2012 (I'd forgotten) I finally have 1G Factorial sieving. I'm running it on a gtx 460 but wanted to know what all the different switches were.
-n10
-N1M
-t1 - think this is the threads
-b512 - block size?
-s1000
Any info would be helpful. I did make adjustments to both the -t1 and -b512
Thanks Rick
edit: so I hope I did this correctly. There's a text file that needs to be in the working directory I believe. Not sure if it's my browser or what but instead of downloading it opened it. So I created the file and copied the contents. |
|
|
dukebgVolunteer tester
Send message
Joined: 21 Nov 17
Posts: 242
ID: 950482
Credit: 23,670,125
RAC: 0
           
|
Any info would be helpful.
run it with -h to see help |
|
|
|
|
Any info would be helpful.
run it with -h to see help
Thanks! |
|
|
|
|
|
Here's a quick result of a factorial sieve run and the tweaks.
For my first run I only reserved 1G. I reserved 5G for my next run. Will get larger reservations once I get things running like I want them and so I'm not bombarding Jim with reservations.
Running a gtx 460.
First run I had -t1 / -b512 and / -s1000
The gpu ran at a pretty steady 94% usage and the sieve ran at 2.785 p/sec
Second run I have -t1 / -b512 / -s2000
The gpu is running at 97% usage and the sieve is running at 2.902 p/sec
So the -s switch (number of iterations per call to gpu) appears to make a difference. The -t seemed to slow things down. Most likely because I'm running SR5 on all the cores.
Cheers |
|
|
|
|
|
Not sure how the check point is suppose to work and read somewhere there might be an issue using 1 thread. Power outage and when I went to restart the sieve I got an invalid check point so it started from the beginning. Not that big a deal but thought I'd mention it. |
|
|
JimBHonorary cruncher
Send message
Joined: 4 Aug 11
Posts: 918
ID: 107307
Credit: 977,945,376
RAC: 0
|
Not sure how the check point is suppose to work and read somewhere there might be an issue using 1 thread. Power outage and when I went to restart the sieve I got an invalid check point so it started from the beginning. Not that big a deal but thought I'd mention it.
If it happens, at worst you can look at the last factor in the factor file and use that as the argument for -n so that the program will start from there. |
|
|
|
|
|
Figured I might let others know.
Recently I had to restart a factorial sieve multiple times. For some reason I was thinking the log file was re-created each time. Turns out it's not, it just adds to what's there for the current sieve you are working on.
As a result I sent a file containing lots of garbage to Jim who had to weed through to get the valid data. For that I apologize and for everyone else just a note and/or reminder should this happen to you.
Cheers Rick |
|
|
|
|
|
I am running Factorial Sieve on a GTX650ti. It seems quite slow in comparison to GFN Sieving.
Can anybody besides Rick Reynolds (GTX460) post his settings for faster crunching? Has anybody tested the settings on newer GPUs?
I used: -n10 -N1M -t1 -b250 –s5000 –q, but it was not a huge speed-up!
Thanks a lot!
|
|
|
rogueVolunteer developer
Send message
Joined: 8 Sep 07
Posts: 1255
ID: 12001
Credit: 18,565,548
RAC: 0
|
I am running Factorial Sieve on a GTX650ti. It seems quite slow in comparison to GFN Sieving.
Can anybody besides Rick Reynolds (GTX460) post his settings for faster crunching? Has anybody tested the settings on newer GPUs?
I used: -n10 -N1M -t1 -b250 –s5000 –q, but it was not a huge speed-up!
You should consider using -t2 in case the GPU is starving. You should also play around with -b and -s to see if you can find better settings that improve throughput without sacrificing OS responsiveness WRT screen refreshes.
Factorial (and primorial) sieving are iterative. There is no short-cut to determining if some prime number p divides a term. In other words to see if p divides n!+1 we have to compute all values from 1!+1 to n!+1 and test if p can divide any of those values.
For most sievers, this is not the case. ppsieve uses a fixed-n type sieve on numbers of the form k*2^n+1. Given some value for n and p, it can quickly compute which k are divisible by that p. srsieve/sr1sieve/sr2sieve use a fixed-k type sieve (with a discrete log) on numbers of the form k*b^n+1. Given some value for k, b, and p, it can quickly compute which n are divisible by that p.
FYI, I am writing a version of mfsieve (which can replace fsievecl and is used for both factorials and multi-factorials) that can sieve in both the CPU and GPU concurrently. It won't be faster on the GPU, but when combined with the CPU will be faster. |
|
|
|
|
|
Just to come back with some test results (it is one run each):
-p7747G -P7748G -n10 -N1M -t1 -b250 -s1000 -q: 3.148K p/sec, Time 10857.99 seconds
-p7748G -P7749G -n10 -N1M -t1 -b250 -s2000 -q: 3.516K p/sec, Time 9576.95 seconds
-p7759G -P7760G -n10 -N1M -t1 -b256 -s5120 -q: 3.859K p/sec, Time 8657.44 seconds
-p7760G -P7761G -n10 -N1M -t1 -b128 -s10240 -q: 3.606K p/sec, Time 9655.42 seconds
-p7761G -P7762G -n10 -N1M -t1 -b256 -s10240 -q: 3.939K p/sec, Time 8357.68 seconds
All results on a GTX650ti with CPU with climateprediction.net loaded.
|
|
|
