20090113, 21:48  #12  
May 2007
Kansas; USA
2×5^{2}×7×29 Posts 
Quote:
Great. Thanks for letting us know. Feel free to coordinate sieving with various folks here if you think someone might be interested. 

20090207, 23:46  #13  
I quite division it
"Chris"
Feb 2005
England
31×67 Posts 
Quote:
http://mersenneforum.org/showpost.ph...7&postcount=12 

20090208, 04:19  #14 
Jun 2006
Chertsey Surrey UK
101100110_{2} Posts 
I am working on 2931

20090209, 18:26  #15 
Jun 2006
Chertsey Surrey UK
358_{10} Posts 
Got a few K's reserved.
2927 through 2953. What is the fastest way to sieve, i used NewPGen in the past. going up to 2T would take forever with that program though. Do we have to sieve to 2T ? 
20090209, 18:31  #16  
A Sunny Moo
Aug 2007
USA (GMT5)
1869_{16} Posts 
Quote:
srsieve and sr2sieve can be downloaded at http://www.geocities.com/g_w_reynolds/. Once you download and extract them you can run them each with the h option to see the various command line flags. Feel free to drop me or Gary a line if you need help getting started with srsieve/sr2sieve. Max P.S.: If you're doing just one or two k's, then the most efficient way to sieve them is to start with srsieve as before, but then switch to sr1sieve instead of sr2sieve. sr1sieve can only run one k at a time, but is much faster than sr2sieve for one k. Thus, if you're doing <=2 k's, your best bet is to sieve them separately with sr1sieve, but if you're doing more than 2 then you should use sr2sieve. The process for running sr1sieve is slightly different than that for sr2sieve; again, feel free to drop Gary or me a line if you need any help. Last fiddled with by mdettweiler on 20090209 at 18:32 

20090209, 20:03  #17 
May 2007
Kansas; USA
10150_{10} Posts 
Brian,
On another related note, starting in ~3 days, I'm going to need some help sieving k=10032000 for n=500K1M; a total of 499 k's. I'll have sieved to P=1T and we need it sieved to P=~30T. If you have a quad or two that could help with that, that would be very helpful. I may put the sieving up for public reservations but it will be somewhat low priority at this point because we'd like to push the 5th/6th/7th drives higher after the 8th drive has finished to n=500K. The smaller k=4001001 LLR's faster than k=14002000 at the same nrange. I mention this because we'll be using sr2sieve and there should be a lot of helpful tips in there on running it. Gary 
20090209, 20:33  #18 
Jun 2006
Chertsey Surrey UK
101100110_{2} Posts 
Gary
I should be able to help out with the sieving. BTW Sr2sieve isn't exactly user friendly :) 
20090210, 01:38  #19  
May 2007
Kansas; USA
2×5^{2}×7×29 Posts 
Quote:
This is true but it is more user friendly then it used to be. I have time to provide some instructions for sieving your range now. I'll assume that you want to sieve n=350K1M. 1. Create a folder and copy the most recent versions of 64bit srsieve, sr2sieve, and srfile into it. 2. In the folder in #1, create a file that has all of your forms in it. For upcoming instructions, I'll call the file "forms29272953.txt". It would just be a notepad/wordpad file with the following lines: 2927*2^n1 2929*2^n1 2931*2^n1 2933*2^n1 (etc.) 3. Do an initial sieve using srsieve. Go to the command prompt, navigate to your folder, and run srsieve with the following command: srsieve a n 350e3 N 1e6 m 1e9 P 500e6 forms29272953.txt [If running Linux, put a "./" before the program name.] a tells it to write to an "ABCD" file (the kind that sr2sieve can read) n is the lower nlimit, i.e. n=350K N is the upper nlimit, i.e. n=1M m tells it to not display factors below 1G (1 billion) on the screen. (the display will slow down the program) P is how far you want it to initially sieve using srsieve. Anywhere between P=200e6 and 1e9 is good. 4. #3 will write out a file with some cryptic name. (It will tell you what it is.) Rename that to something meaningful. For reference in the upcoming instructions, I'll call it "sieve29272953.txt". 5. Run sr2sieve using sieve29272953.txt. At the command prompt, run the following command: sr2sieve P 5e12 i sieve29272953.txt P is the max sieve depth, i.e. P=5T. (If you wanted to begin sieving at a different limit than what is shown on the 1st line of the file, i.e. if you were sieving on multiple cores, you would need to use the p along with the P commands.) i tells sr2sieve the input file name. The sieving in #5 will not remove k/n pairs from the file. It writes out factors to a file called factors.txt instead. 6. Run srfile to remove the factors in #5 and write it to a file that can be understood by LLR. At the command prompt, run the following command: srfile G k factors.txt sieve29272953.txt. G tells it to write out a file that is readable by LLR in nvalue sequence. k means that the next file contains "known factors". To understand what these known factors are, take a look in that file. It should be pretty clear what they are. 7. #6 will write out a file with some cryptic name. Once again, it will tell you what it is. Rename it to something more meaningful. 8. (Optional) Change the sieve limit on line #1 in the file from #7 to your true sieve limit (in this case 5T or 5000000000000). Unfortunately neither sr2sieve nor srfile will change it even though you sieved to P=5T. It will still show 500000000 (500M), which was your sieve limit from when you ran srsieve in #3. #8 is not technically necessary but I do it because you may want to sieve the file more in the future and you will likely find that you will forget where you stopped sieving. Now...that is the easy instructions. I didn't get into computing optimum sieve depth. P=5T is not likely nearly high enough for a large range of n=350K1M but is probably enough for n=350K500K or 350K600K. Also, I didn't get into splitting up the sr2sieve part amongst several cores, which you'll likely want to do, lest it take you many weeks to sieve. Nor did I get into sieving, breaking off the lower nrange, LLRing, then sieving the higher nrange some more, etc., which you'll want to do if your max nlimit is more than twice your min nlimit. Once you've done the above a couple of times, it's super simple, I promise. At that point, the trickier part about optimum sieve depth and all that mumbojumbo won't seem too bad. I too thought it was hideously difficult and nonuser friendly when I first ran sr2sieve, only to find out that it was quite easy once I did it a couple of times. All of the programs come with a helpful README file. You can also get help on the various switches (commands) by typing srsieve h (or sr2sieve h or srfile h) at the command prompt. One final thing: Never use NewPGen again! (lol) It is WAY too slow for any kind of sieving job except for exotic forms such as twins, quadruplets, Sophie Germain's, etc. The sr(x)sieve suite of programs are many times faster. Good luck! Gary Last fiddled with by gd_barnes on 20090210 at 01:47 

20090210, 04:15  #20 
Jun 2006
Chertsey Surrey UK
358_{10} Posts 
Thanks for that it has made things much clearer.
Yes i will want to split the sieving work over multiple cores. Looks like i can vary the values of n and N in step 3 to split the work up. srsieve a n 20e3 N 350e3 m 1e9 P 500e6 forms29272953.txt (Not sure if i'll bother with this, do we need the data ?) srsieve a n 350e3 N 500e3 m 1e9 P 500e6 forms29272953.txt srsieve a n 500e3 N 600e3 m 1e9 P 500e6 forms29272953.txt srsieve a n 600e3 N 700e3 m 1e9 P 500e6 forms29272953.txt srsieve a n 700e3 N 800e3 m 1e9 P 500e6 forms29272953.txt srsieve a n 800e3 N 900e3 m 1e9 P 500e6 forms29272953.txt srsieve a n 900e3 N 1e6 m 1e9 P 500e6 forms29272953.txt Does the above look right ? Those pieces can then be sieved to different depths and LLR'd. Is there a big speed advantage to 64bit ? I don't mind going 64bit on the sieving boxes if there is a good speed gain. Is it worth bothering with older AMD's, i have an X2 4800 doing nothing ? Guessing the core2s sieve much faster so it wouldn't be any good these days. Last fiddled with by PCZ on 20090210 at 04:31 
20090210, 04:32  #21  
May 2007
Kansas; USA
2·5^{2}·7·29 Posts 
Quote:
No, don't split the work up by doing different nranges on different cores for the same kvalues. That would be very inefficient. Efficiency is gained by sieving the entire k and nrange at once. Instead of splitting up the k or nvalues, split up the Pvalues. For instance: 1. Use one core to sieve the entire k and nrange to P=500M using srsieve. That shouldn't take too long...probably < 1 hour. 2. Use multiple cores to sieve the entire k and nrange by doing P=500M100G on one core, 100G200G on core 2, 200G300G on core 3, etc. using sr2sieve. (You can finetune this so that they complete at close to the same time but even if you don't, they'll still likely finish within a few hours to a day of each other. The lower Pranges go slower because there are more possible prime factors.) 3. Assuming that you are using 4 cores, at this point, remove all of your factors from #2. Then (assuming you've removed all factors to P=400G) use sr2sieve and sieve P=400G800G on core 1, sieve 800G1200G (0.8T1.2T) on core 2, etc. I recommend this method because you'll get a speedup in sieving by removing the smaller factors and then starting again with a smaller sieve file across all cores. You'll get much deeper sieve depths in the same amount of time this way instead of splitting it up by kvalue or nrange. What nrange are you sieving and how many cores will you be using? If I know that, I can recommend (close to) the best sieving method. Most AMD's sieve well. Go ahead and use your old ones if you want to mess with them. Run a test on the older and newer machines for a few mins. to see how fast they sieve. On your newer machines, make sure you have a 64bit O.S. You'll get the fastest sieving that way. Gary Last fiddled with by gd_barnes on 20090210 at 04:33 

20090210, 04:42  #22  
May 2007
Kansas; USA
10011110100110_{2} Posts 
Quote:
I see you edited your post. Just a couple of more questions to answer that I didn't answer already: What you suggested here would be correct if you split them up by nrange but no, I don't recommend doing it that way. See my prior post. Unlike most projects, NPLB tries to leave no prime behind, not even the small ones. So yes, it is preferrable that you also run n=20K350K, i.e. nontop5000 ranges. Of course that is your choice. Karsten will reflect all primes, small and large at www.rieselprime.org. Now that I know you will be sieving n=20K1M, shoot me your number of cores, and I can give a more exact recommended sieving method. Since you're doing such a wide nrange, I can point you directly to an answer that I gave to MrOzzy on the CRUS project when he asked about the best way to sieve a different base for n=50K250K. You'll definitely want to sieve the entire thing to a certain depth, break off the lower nrange, LLR it, sieve the rest to a certain depth, break off and LLR a lower nrange, and then do the same at least 2 more times for the remaining nranges. Gary Last fiddled with by gd_barnes on 20090210 at 04:44 

Thread Tools  
Similar Threads  
Thread  Thread Starter  Forum  Replies  Last Post 
Bases 33100 reservations/statuses/primes  Siemelink  Conjectures 'R Us  1663  20200706 20:15 
Bases 251500 reservations/statuses/primes  gd_barnes  Conjectures 'R Us  2111  20200706 12:06 
Report top5000 primes for k=10033000  gd_barnes  No Prime Left Behind  688  20200424 07:31 
misc. k<300 statuses and primes  gd_barnes  No Prime Left Behind  37  20090914 15:09 
Sieving drive for k=10032000 n=500K1M  gd_barnes  No Prime Left Behind  160  20090510 00:50 