A Desperate appeal! (by Richard K. Guy)... deadline is September 30, 2016

[ET_]

Originally Posted by axn

I think it might be ever so slightly more efficient to use a B2/B1 multiplier of 200, since that gives roughly equal times for stage 1 & 2 (which should be optimal according to RDS).
Anyone care to compute the expected curves for both configuration and see whether it makes any difference?

I suppose the GIMPS manual submission page should do the math for you, crediting the appropriate number of curves for a definite B1... Anyway, I will gladly wait for some more tests.

[ET_]

Originally Posted by pinhodecarlos

Share with me those parameters and I will try to bring some support from SETI.USA team and also from myself.

When I set up the estimates for F12-F29, Yoyo answered that such workunits were much too heavy for them.

But if we limit our search on F12 nd F13 (no more than half GB of RAM and 6 hours per WU) I think we may receive some help from other teams.

[GP2]

Originally Posted by ET_

I suppose the GIMPS manual submission page should do the math for you, crediting the appropriate number of curves for a definite B1... Anyway, I will gladly wait for some more tests.

There is no need to use the manual submission page, Fermat results are reported to PrimeNet in the usual way.

[ET_]

Originally Posted by GP2

There is no need to use the manual submission page, Fermat results are reported to PrimeNet in the usual way.

Correct.
I used manual submission to pinpoint the differences when I worked on different B values :redfaace:

[Joe O]

So we currently have
F12: 57,363 of 360,000 for 65 digits
F13: 82,193 of 112,000 for 60 digits

Now we probably can take F13 to the next level, and start on 65 digits. But what happens when we get one or both to the end of the 65 digits level without finding a factor?

[henryzz]

Originally Posted by Joe O

So we currently have
F12: 57,363 of 360,000 for 65 digits
F13: 82,193 of 112,000 for 60 digits

Now we probably can take F13 to the next level, and start on 65 digits. But what happens when we get one or both to the end of the 65 digits level without finding a factor?

We start on the 70 digits level. These numbers need the >100 digit level before NFS we don't stand a chance of progressing beyond ECM any time soon without a new algorithm.

[GP2]

Originally Posted by henryzz

We start on the 70 digits level. These numbers need the >100 digit level before NFS we don't stand a chance of progressing beyond ECM any time soon without a new algorithm.

But what is the "official" B1 value for the 70 digits level? the ECM report page doesn't show it.

Also, are we certain that the mprime program is able to go that high?

[ET_]

Originally Posted by GP2

But what is the "official" B1 value for the 70 digits level? the ECM report page doesn't show it.

Also, are we certain that the mprime program is able to go that high?

The ECM report page used to say "B1=110,000,000" some years ago.
Then it was modified to "B1 = 260,000,000".
Now it is at "B1 = 800,000,000" (BTW, I never understood why it's B1 = 800M instead of 850M as proposed by GMP-ECM gurus).

So I expect George (or someone else for him) will add the new range as soon as it becomes available

[henryzz]

Originally Posted by ET_

The ECM report page used to say "B1=110,000,000" some years ago.
Then it was modified to "B1 = 260,000,000".
Now it is at "B1 = 800,000,000" (BTW, I never understood why it's B1 = 800M instead of 850M as proposed by GMP-ECM gurus).

So I expect George (or someone else for him) will add the new range as soon as it becomes available

The bounds are potentially different due to the b2=100*b1 which is much smaller than gmp-ecm would use.

[ET_]

Originally Posted by henryzz

The bounds are potentially different due to the b2=100*b1 which is much smaller than gmp-ecm would use.

That's why I expected a bigger B1, not a smaller one...

[xilman]

Originally Posted by ET_

That's why I expected a bigger B1, not a smaller one...

The work curve has a very flat minimum. It doesn't matter very much whether you run more curves at a lower B1 or fewer at a higher one, as long as you don't push it too far. To an excellent approximation, the product B1*#curves is the figure of merit.