Team Spirit

[davieddy]

A friend has TFed my last two or three first time LLs
from 68 up to 75 bits on his GPU.

~1000 first time LL tests get assigned each day,
typcally TFed to 70 bits or less.
~200 of these get completed.
As a "joint exercise", can't GIMPS get its act together
and up the level to say 72?

P-1 (if needed) can be done by the CPU doing the LL.
No point in doing further TF without a GPU though.

David

[ckdo]

Originally Posted by davieddy

As a "joint exercise", can't GIMPS get its act together and up the level to say 72?

Well, your friend could have taken some 8.5 expos from 68 to 72 per one taken from 68 to 75. You certainly had your reasons to tell them otherwise.

[Christenson]

Oh shush...everybody that cares to knows I am that friend...and truthfully, I could have taken 20 or 30 27M or 28M exponents from ckdo to 68 bits in the same time, and found 0.2 or 0.3 of a factor. That amount of TF is well beyond the mathematical optimum, given the overall state of TF, and distracting me from upgrading mfaktc, but somebody told me to have some fun...

Be patient, in a project this large, a global optimum is almost impossible, and more TF is at best a 10-20% optimisation. Xyzzy and them with 10-100x as much TF power as me work above 50M, right now ckdo has me finding factors at 27-30M, and I know this is *all* good work.

My CPUs are actually doing majority P-1. I'm a programmer, and it is probably globally optimum for GIMPS (and possibly also my own career) for me to spend my time helping out with programming, rather than any actual running of computers and disproving of primality. It's unlikely that my studying up on the number theory will have much effect on anyone besides RDS.

[Mr. P-1]

Originally Posted by davieddy

P-1 (if needed) can be done by the CPU doing the LL.
No point in doing further TF without a GPU though.

The CPU doing the LL can't always do stage 2 of P-1.

Not that I'm disagreeing with you. It would certainly be desirable for those with GPU's to factor LL wavefront assignments up to a high level. However, I don't have one, so can't participate in that effort. I can, however, do a P-1 with a respectable stage 2.

[davieddy]

Originally Posted by Christenson

Oh shush...everybody that cares to knows I am that friend

You've blown our cover

David
x

[davieddy]

Originally Posted by ckdo

Well, your friend could have taken some 8.5 expos from 68 to 72 per one taken from 68 to 75. You certainly had your reasons to tell them otherwise.

I didn't twist his arm. I just said "a few" more bits.
I'll check out the 8.5 at my leisure, but it is certainly in the
right ball park.

David

One more bit

[Dubslow]

Originally Posted by davieddy

A friend has TFed my last two or three first time LLs
from 68 up to 75 bits on his GPU.

~1000 first time LL tests get assigned each day,
typcally TFed to 70 bits or less.
~200 of these get completed.
As a "joint exercise", can't GIMPS get its act together
and up the level to say 72?

P-1 (if needed) can be done by the CPU doing the LL.
No point in doing further TF without a GPU though.

David

I believe in various places it's been mentioned that George's goal is two more bits than the published bounds on PrimeNet and in the 26.6 source. That would be 71 bits for exponents up to 58,520,000.

[davieddy]

Originally Posted by Dubslow

I believe in various places it's been mentioned that George's goal is two more bits than the published bounds on PrimeNet and in the 26.6 source. That would be 71 bits for exponents up to 58,520,000.

Don't you mean 58,521,000?
Let's get things right round here