The default k value has been lowered to 2, mostly because the POLYEVAL algorithm is much faster now. This increases memory requirements wrt 5.0.3, increasing k with the -k option helps lower it again. I think with some combinations of input numer size and B2, larger k might even make stage 2 a little faster. If you plan to do a lot of curves with the same parameters, trying different k to see which is fastest would be worthwhile.

Alex

With the new version do we still do the same number of curves at each digit level, or less?

Edit1: Nevermind... [url]http://www.loria.fr/~zimmerma/records/ecm/params.html[/url]
Edit2: Those old values don't look familiar... Hmm...

I'll keep track of the 55 digit effort - it doesn't make sense for this to be the only 2- number tracked by a forum.

We need to decide how best to do this. I could count curves based on
B1=110M, B2 = 4290M - prime95's default.
B1=110M, B2 = 100*B1 - the same b2/b1 ratio assumed in every other column
B1=110M, B2 = (some GMP-ECM value).

I'm kind of leaning to the middle one and have you good folks tell me how many curves are needed, and how many you've done thus far (scaled appropriately).

Also, Bruce Dodson did an extra 3300 curves at P50. This can be translated into a smaller number of P55 equivalent curves.

akruppa made the [url=http://www.mersenneforum.org/showpost.php?p=51205&postcount=22]suggestion[/url] to use fractions of the needed curves, thus a sum of "1" means that all required curves have been run.
Using gmp-ecm6 (for stage2), it is no big deal to calculate the fractions.

The advantage: Everyone can find out the individually optimal bounds (even B1!) and there's is no need for the moderator to do conversions.

The curve count chosen for B2=100*B1 was 46000, which seems to be roughly what gmp-ecm 6.0 reports if you use these bounds with -power 1 polynomial. gmp-ecm 6.0 reports 17899 default curves are required (using 'ecm -v'), and one gmp-ecm 5.0.3 curve is so close to a gmp-ecm 6.0 one that the difference can be ignored. So one gmp-ecm (55-digit) default curve is worth 46000/17899 = 2.57 curves.

For any new curves reported here we can just report the number N of curves done and the number E in the 55 digit column of the ecm-v output, and that will count as N*46000/E curves in George's table.

21 curves using B1=11e7 & B2=680270182898
62 curves using B1=26e7 & B2=2317234062120

Some benchmark numbers for B1=11e7 & M1061 with GMP-ECM 6:
[code]k=10 / Step 1 took 1924304ms / Step 2 took 1533830ms / mem=353
k= 5 / Step 1 took 1921174ms / Step 2 took 1361871ms / mem=529
k= 4 / Step 1 took 1922516ms / Step 2 took 1328514ms / mem=753[/code]

How long do you need for the 26e7 curves?

[QUOTE=Mystwalker]How long do you need for the 26e7 curves?[/QUOTE]
[url]http://www.mersenneforum.org/showpost.php?p=50999&postcount=53[/url]

Thanks.
I think I'll pick 11e7 curves first, as less than 10% is done there.

Btw.:
We can count the 26e7 curves as "very valuable" 11e7 curves as well...

If somebody is interested, I have collected almost 700 curves at stage1=11e7 (this increases by 68 curves per day).
My maximal RAM is 1 GB, so I already have to use higher k and treefile for stage2, which slightly decreases efficiency. I can hand over as many curves as desired...

I currently have access to an Athlon XP 3000+ with 3GB ram - a very useful machine for ECM stage 2! If I may, I'd like to take, say, 200 curves?

Alex