3+ table

[R.D. Silverman]

Originally Posted by bdodson

And reserved 3, 563+ c239 which will leave the 3+ table shorter
than the line on the "first five holes" page, updated yesterday. Actually,
the current set also includes 3,563- c199, which will drop that table to
length three. (At the moment, these are on Sam's "who's" list, but not
yet on the NFS@Home "Status".) The c199 has finished ecm pretesting,
while the c239 has another t55 left to go. A very nice pair of Cunningham
numbers; NFS@Home could use a few more dedicated contributors,
or even a few more with just a passing interest.

-Bruce

A note of theoretical interest: If one actually sieved both numbers
together, one could save 25% of the sieving time by sieving both
numbers simultaneously. i.e. the rational side has the same polynomial
for both.

[R.D. Silverman]

Originally Posted by bdodson

NFS@Home could use a few more dedicated contributors,
or even a few more with just a passing interest.

-Bruce

Absolutely! Or, even more than a "few more"

2,1798L seems to be taking an unusually long time to sieve.
Might this be because it requires the 16e siever and many
contributors do not have enough memory? If so, this will
be a big problem moving forward as NFS@Home pushes over
270 digits.

[Batalov]

Originally Posted by R.D. Silverman

A note of theoretical interest: If one actually sieved both numbers
together, one could save 25% of the sieving time by sieving both
numbers simultaneously. i.e. the rational side has the same polynomial
for both.

Same goes for the 2,1191+-, 10,387+-, and 11,287+- reserved pairs!
Makes one think...

[frmky]

Originally Posted by R.D. Silverman

A note of theoretical interest: If one actually sieved both numbers
together, one could save 25% of the sieving time by sieving both
numbers simultaneously. i.e. the rational side has the same polynomial
for both.

All a^(13n)+1's share the same algebraic polynomial, and likewise for all a^(13n)-1's, a^(11n)+1's, etc.

Originally Posted by R.D. Silverman

2,1798L seems to be taking an unusually long time to sieve.
Might this be because it requires the 16e siever and many
contributors do not have enough memory?

Memory is definitely an issue for many contributors, but see that as improving over time. New computers today regularly come standard with 3 or 4 GB of memory. 2,1798L is filtering now, with a total of about 510M relations found. I'll update the website soon with its status and the new reservations.

[R.D. Silverman]

Originally Posted by Batalov

Same goes for the 2,1191+-, 10,387+-, and 11,287+- reserved pairs!
Makes one think...

I said that it is of theoretical interest because generally speaking:

(1) Code to sieve two numbers simultaneously requires a lot more memory
(approx 50% + epsilon more)

(2) There is a lot of overhead and extra data management. This reduces
the theoretical improvement.

I actually put together some crude code a long time ago that did it.
It was not very effective.

[R.D. Silverman]

Originally Posted by frmky

All a^(13n)+1's share the same algebraic polynomial, and likewise for all a^(13n)-1's, a^(11n)+1's, etc.


Memory is definitely an issue for many contributors, but see that as improving over time. New computers today regularly come standard with 3 or 4 GB of memory. 2,1798L is filtering now, with a total of about 510M relations found. I'll update the website soon with its status and the new reservations.

Yep. But 4GB would not be enough to run 4 threads of such code on (say)
an i7. This limits the effectiveness of multiprocessors.

I have an i7 and run 8 threads of SNFS on it all the time. The machine has
8Gb of memory. I would not have enough memory for 8 threads worth
of a 2-numbers-at-once siever.

[henryzz]

Originally Posted by R.D. Silverman

Yep. But 4GB would not be enough to run 4 threads of such code on (say)
an i7. This limits the effectiveness of multiprocessors.

I have an i7 and run 8 threads of SNFS on it all the time. The machine has
8Gb of memory. I would not have enough memory for 8 threads worth
of a 2-numbers-at-once siever.

Just turn off hyperthreading. Not much throughput will be lost.

[axn]

Originally Posted by henryzz

Just turn off hyperthreading. Not much throughput will be lost.

I don't think that is a true statement re: ggnfs sievers. Prime95, yes.

[FactorEyes]

Originally Posted by axn

I don't think that is a true statement re: ggnfs sievers. Prime95, yes.

Uh-huh.

Hyperthreading is a great enhancement for sieving. As fivemack has noted, each i7 core can two sieving threads at 70% throughput, turning it into 1.4x its former self.

These i7 processors will sieve like a hummingbird on meth.

[R.D. Silverman]

Originally Posted by FactorEyes

Uh-huh.

Hyperthreading is a great enhancement for sieving. As fivemack has noted, each i7 core can two sieving threads at 70% throughput, turning it into 1.4x its former self.

These i7 processors will sieve like a hummingbird on meth.

Mine does.

[frmky]

NFS@Home has finished 3,563+. The log is attached.

prp117 factor: 121656494904519877502603905124362379497504761645941205875831210198221747342109726269588052956292905428218771513865307
prp123 factor: 134937859494643008258097068796015791834210619609460109943664994747188230937155152634097561189792109809240657159104064420503