20060111, 16:53  #221 
Jul 2005
2·193 Posts 
Good work Robert!
These files have been mirrored at: http://octoproth.greenbank.org/downloads/oc36.c http://octoproth.greenbank.org/downloads/octo_3_6.exe Last fiddled with by Greenbank on 20060111 at 17:01 
20060111, 16:59  #222  
"Robert Gerbicz"
Oct 2005
Hungary
1,429 Posts 
Quote:
You can download those two files and after it upload them. [EDIT by Greenbank] Thank you! I've fixed the links and uploaded the correct version of octo_3_6.exe. Last fiddled with by Greenbank on 20060111 at 17:03 

20060111, 17:01  #223 
Jun 2003
Oxford, UK
2×7×137 Posts 
Goals
I think a useful goal would be to look at the mathematics behind why some n values produce so many octoproths, whilst others seem to produce less.
Regarding chains of 3 4 etc, it is my bet that such chains do exist, but will prove very elusive. Where n is small, the population of k is also small because of the 2^nk case. I think there is some modular maths that could be applied to perhaps narrow the choice of possible k. We should be looking in the n=30 to 60 range. Of course having the whole possible population of octos for those n vales is essential, as these are the only k that need to be checked! Regards Robert Smith PS congratulations to everyone for making this a "project" and for their hard work in continuing to look for speed ups in software. 
20060111, 17:18  #224  
Jun 2003
1001011101111_{2} Posts 
Quote:
 Kilobit Octo (I think outofreach for single individual but ideal for DC)  Titanic Octo (Pushing it )  Cleaning up all n's below a certain level. Currently it's cleaned up till n=45 (64 sounds tough, but feasible; 80 is too much)  Hopefully the cleanup effort will yield some elusive dodecaproths. In addition, there is always individual glory and bragging rights  most Octo, biggest Octo, etc... Also, somebody could probably try to come up with asymptotic density calculations for Octos and higer forms. EDIT: Ooh.. Almost forgot. Generalized Octos!!! With bases other than 2. For odd base b, k should be even and viceversa. Last fiddled with by axn on 20060111 at 17:23 

20060111, 21:53  #225 
"Robert Gerbicz"
Oct 2005
Hungary
1429_{10} Posts 
Warning about octo_4_0 bugs!!!
Some time before I've uploaded octo_4_0.exe to my website. I couldn't post the c code because mersenneforum was unavailable. In that time I've tested for some large Range the new program ( before I've tested only for small ranges, for known solutions ) and I've observed that octo_4_0 has got bugs!!! Some solutions has been missed! So delete that file from your computer!
Greenbank if you'll get a results text from version 4.0 ( you can find that for each project in that text file ) then consider that it is very very possible that some solutions has been missed. Ps. I've deleted that file from my webpage and there is a warning message about it! Last fiddled with by R. Gerbicz on 20060111 at 22:04 
20060111, 23:33  #226 
"Robert Gerbicz"
Oct 2005
Hungary
1,429 Posts 
octo_4_3 program
This new octo program is using now exactly the kmin,kmax range in the sieve, previous versions has oversieved to use a special sieve, now I've rewritten that part, so now there is no oversieve. Note that the speedup is between 0%25% depending on your range, it isn't easy to give an average rate ( 25% is a theoretic value), and there is also smaller number of prp tests, because there is no redefinition of range.
See my attachment for c code or download the exe for windows from my webpage:http://www.robertgerbicz.tar.hu/octo_4_3.exe 
20060111, 23:39  #227 
"Robert Gerbicz"
Oct 2005
Hungary
1429_{10} Posts 
octo_4_5 program
It is exactly the same as 4.3 version but it'll try to use one more prime in the sieve reduction step, this is value of x in the program. So the speed of this is exactly the same as the speed of 4.3 if it'll use the same x value, but it can be faster for one more prime. Try it for different ranges and different n values. Note that it can be slower than 4.3 !!!
Now magic_constant=10^5, see previous versions: it was started from 2*10^6 See the attachment for the c source or my webpage for the exe file for windows: http://www.robertgerbicz.tar.hu/octo_4_5.exe Ps: It would be very good if somebody could test for known n values and known ranges the programs! Note that version 4.0 has got a very serious bug! Last fiddled with by R. Gerbicz on 20060111 at 23:44 
20060112, 04:48  #228 
Sep 2004
UVic
2×5×7 Posts 
171
n=171, kmin=1, kmax=10000000000, version=4.5
Starting the sieve... Using the first 6 primes to reduce the size of the sieve array The sieving is complete. Number of Prp tests=156 Time=1 sec. n=171, kmin=10000000000, kmax=1000000000000000, version=4.5 Starting the sieve... Using the first 10 primes to reduce the size of the sieve array 99180515934675 171 855203637332835 171 42579432483105 171 111420166614855 171 The sieving is complete. Number of Prp tests=15652022 Time=12084 sec. 
20060112, 10:12  #229 
Nov 2003
7046_{8} Posts 
Thanks to axn for the feedback and to R.Gerbicz for quick action and many new versions. Four new versions in one day is a little bit too many for me but I never saw ver.4.0 so I tested three versions: 3.6, 4.3, and 4.5. using the same input parameters like before (n=336, k=100111T). All compiled using O2 and march=pentium4. Here are the times:
3.6 ... 179 sec, 133206 PRP tests (the same time like my ver.3.5) 4.3 ... 165 sec, 125962 PRP tests 4.5 ... 171 sec, 125962 PRP tests (the same like 4.3) So I'm now using 4.3. I noticed that the only difference between 4.3 and 4.5 is the size of magic_constant. Maybe 10^5 (now) is not the best value for my test case? Can you change the program so that the constant can be set from the command line. But you will also have to tell us how to use it. Finally, a proposal for a simple new feature: Can you count the number of found OctoProths in one run, and print that at the end. With many numbers on the screen and in the results file it's easy to miss a precious Octo! It's also useful when testing correctness of the program (I use n=5255 and k<1T). Now I have to count the number of Octos in the results file (90 for n=52, 12 for n=53 etc.) Sorry, but today I didn't have time for other tests (O1, O3 and bench.c). 
20060112, 11:20  #230  
"Robert Gerbicz"
Oct 2005
Hungary
1,429 Posts 
Quote:
If you reach another breakpoint ( because magic_constant is smaller ), then x is larger by one, so it is possible that you get an improvement ( this is depending on the Range and on n value ), but it is also possible that it'll be slower. I've gotten the two cases on my PC for different input parameters. It isn't need to modify further magic_constant, because in 4.3 this is 4*10^5, if it would be larger then we get a smaller x value ( or the same ) but this is suboptimal. In version 4.5 this is 10^5, so you'll use the same x value or x+1, if it would be smaller then that isn't optimal, because we are sieving up to 10^5. Also note that using bench.c isn't very practical in this case, because we are testing very special numbers, and for this mpz_powm instruction is faster by a factor of 2. So probably my program isn't very bad for n<512 Last fiddled with by R. Gerbicz on 20060112 at 11:25 

20060112, 15:45  #231  
Jul 2005
386_{10} Posts 
Quote:
Code:
$ gcc m64 mpowerpc64 O2 fomitframepointer oc45.c o oc45 lgmp $ ./oc45 336 100000000000000 111000000000000 You can also find the k n values in results_octo.txt file ( These are 3probable primes ) n=336, kmin=100000000000000, kmax=111000000000000, version=4.5 Starting the sieve... Using the first 8 primes to reduce the size of the sieve array The sieving is complete. Number of Prp tests=125962 Time=82 sec. Last fiddled with by Greenbank on 20060112 at 15:46 

Thread Tools  
Similar Threads  
Thread  Thread Starter  Forum  Replies  Last Post 
Small Primes for Octoproths <= 155  ValerieVonck  Octoproth Search  100  20070216 23:43 
Found Octoproths  Range Archive  ValerieVonck  Octoproth Search  0  20070214 07:24 
Number of octoproths per n  Greenbank  Octoproth Search  15  20060120 16:29 
Need help with NewPGen(octoproths)  jasong  Software  1  20050510 20:08 