Sierpinski base 4

[geoff]

At p=5e12, the sieve speed for a P2/400 running Linux is:

proth_sieve_cmov 0.42: 193 kp/s
sr2sieve-1.4.9-linux32-i686: 340 kp/s

If anyone wants to try out sr2sieve, there are some binaries here.

[Citrix]

Originally Posted by geoff

At p=5e12, the sieve speed for a P2/400 running Linux is:

proth_sieve_cmov 0.42: 193 kp/s
sr2sieve-1.4.9-linux32-i686: 340 kp/s

If anyone wants to try out sr2sieve, there are some binaries here.

That is impressive. Have you tried your program for PSP's dat?

[tcadigan]

Originally Posted by geoff

At p=5e12, the sieve speed for a P2/400 running Linux is:

proth_sieve_cmov 0.42: 193 kp/s
sr2sieve-1.4.9-linux32-i686: 340 kp/s

If anyone wants to try out sr2sieve, there are some binaries here.

ooooo nice!!!
I just ran a quickie....just to see what speed it gives. it's nice big boost^^

at 5500 billion I'm getting 800 kp/s
I don't remember the exact numbers I was getting with jjsieve, but it was something down in the 400-500 range iirc (don't have time to do a full test right now)

this is a 3.2 p4 HT, with a c120 gnfs sieving, and llr for this project picking up the extra cycles. so I would expect that 800 kp/s to jump even higher if I was running just it on my system.

this is awesome because it puts 1000 billion of sieving in reach for 2 weeks +/- a couple days.

great work with your program Geoff!

[S485122]

Originally Posted by geoff

If anyone wants to try out sr2sieve, there are some binaries here.

The files containing the P4 and prescott versions have some problems with extraction.

I do not have these problems with the 1.4.8 versions...

[Xentar]

Originally Posted by geoff

sr2sieve-1.4.9-linux32-i686: 340 kp/s
here.

Is sr2sieve right? Shouldn´t it be sr4sieve, or something like that?

Originally Posted by Xentar

Is sr2sieve right? Shouldn´t it be sr4sieve, or something like that?

a*4^b+1=a*2^(2*b)+1

and thats the form which is sieved i think

[Xentar]

Originally Posted by thommy

a*4^b+1=a*2^(2*b)+1

and thats the form which is sieved i think

ok, thats possible - didnt think about that, sry.
But then, we would need another input file, because of 2*b, right?

sure. but thats taken care of.

[geoff]

Originally Posted by Jacob Visser

The files containing the P4 and prescott versions have some problems with extraction.

I do not have these problems with the 1.4.8 versions...

There is some room for confusion here: note that sr2sieve is for base 2 only, sr5sieve is for base 5 only.

File corruption is possible, I use floppy disks to transfer the files before upload :-(

Is sr2sieve right? Shouldn´t it be sr4sieve, or something like that?

It might be slightly faster to recompile the program to use base 4 instead, but the files are already in base 2 format so it is probably not worth the effort.

That is impressive. Have you tried your program for PSP's dat?

Yes, it is still slower than proth_sieve for PSP. Sr2sieve seems to work better with a small range of n or a large number of k. (But sr2sieve can sieve Sierpinski and Riesel numbers together, so if PSP and Rieselsieve were using the same n range then the difference might be less for a combined sieve).

[Citrix]

I think once you implement the SPH algorithm, your program should be faster than proth_sieve. Any plans to implement SPH algorithm?

[geoff]

Originally Posted by Citrix

I think once you implement the SPH algorithm, your program should be faster than proth_sieve. Any plans to implement SPH algorithm?

Is this post a description of the SPH algorithm?

I have been thinking about how to implement those ideas, but I think it is still quite a lot of work, and I may not have a lot of time to work on the program in 2007.

The source code is there if anyone wants to improve it, or to use it as a base for their own program.