Possible extention to the "GPU to 72 Tool" project? - Page 4

Dubslow · 2011-11-09, 17:41

Our discussion had nothing to do with TF bit levels, just that him getting P-1 assignments this way would remarkably reduce his P-1 throughput.

As for what we're doing, it is at the wavefront, or rather, what the wavefront has missed. We are only TF'ing exponents that have had no LL tests.

We are already doing everything we can for TF. I personally don't go to 72 in one go because I can get many more exponents through if I don't go as high. (Though as it stands chalsall is holding assignments until they get to 71, keeping in mind that we're 5-10M lower than what you're talking about.)

Also, those TF's to 72 bits are only available through manual testing, so only we will get them. As soon as chalsall's spider is unable to saturate us, then we will move higher up the chain.

chalsall · 2011-11-09, 18:48

Quote:

Originally Posted by davieddy

I may frequently act the buffoon, but I can be serious.

Then please be so here.

Quote:

Originally Posted by davieddy

The progress of the wavefront relies on tens of thousands of cores LL testing to completion (even if slowly in some cases). Why should us few "reliable and enthusiastic" folk hog all the low exponents?

Because we can. In an organized, coordinated, cooperative, automated and "legal" way.

And because it is actually good for GIMPS.

I would argue you cannot look at the Workers Progress report and say we're not having a positive effect. Already 24 factors found; approximately 54 LL/DC tests saved at the back end of the "LL Wavefront".

Quote:

Originally Posted by davieddy

We should be selflessly assisting the vast majority of participants by TFing the assignments to the max.

That is what this sub-project is all about.

davieddy · 2011-11-09, 22:37

Quote:

Originally Posted by Dubslow

Our discussion had nothing to do with TF bit levels, just that him getting P-1 assignments this way would remarkably reduce his P-1 throughput.

I apologise for quoting your sensible discussion.
It just triggered my serious rant.
You are right. P-1 stands a ~5% chance of finding
a factor, while one bit of TF stands ~1%.

The subtle reason for doing P-1 before
the "last bit or two" of TF has flown out of the
window since GPUs came on the scene.

We CPUs can do a decent P-1.
We shouldn't get out of bed to do the "last bit" of TF.

David

davieddy · 2011-11-09, 23:04

Quote:

Originally Posted by chalsall

Then please be so here.

Lets get serious
David

PS I love your tool.

Dubslow · 2011-11-10, 00:47

Quote:

Originally Posted by chalsall

I would argue you cannot look at the Workers Progress report and say we're not having a positive effect. Already 24 factors found; approximately 54 LL/DC tests saved at the back end of the "LL Wavefront".

Also add in the two factors that I've found in the last couple of days, clearing out the massive dump Mr. P-1 gave me before this project went up.

Chuck · 2011-11-10, 01:26

Now that this is up and running without (as far as I can see) any hitches, is there any interest in incorporating ckdo's TF before DC into the process? Do others see this project as worthwhile?

I still have about 770 exponents in the 28.9M range I am taking from 68 to 69.

Chuck

diamonddave · 2011-11-10, 02:46

Quote:

Originally Posted by Chuck

Now that this is up and running without (as far as I can see) any hitches, is there any interest in incorporating ckdo's TF before DC into the process? Do others see this project as worthwhile?

Personally, I'm dividing my GPU time between DC-TF (roughly 500 exp to 69) and GPU to 72, so count me in as someone interested.

KyleAskine · 2011-11-10, 03:48

Quote:

Originally Posted by diamonddave

Personally, I'm dividing my GPU time between DC-TF (roughly 500 exp to 69) and GPU to 72, so count me in as someone interested.

Works for me too. Whatever the GPU to 72 system gives me, I am quite happy with. Whether it is factoring for a first time test, or a DC does not matter to me!

Ethan (EO) · 2011-11-10, 04:57

Quote:

Originally Posted by Chuck

Now that this is up and running without (as far as I can see) any hitches, is there any interest in incorporating ckdo's TF before DC into the process? Do others see this project as worthwhile?

I still have about 770 exponents in the 28.9M range I am taking from 68 to 69.

Chuck

I'd love to have DC factoring included in GPU to 72; it may only save one LL test, and it slightly reduces our insight into the reliability of LL tests, but for me personally it is very appealing.

Ethan (EO)

davieddy · 2011-11-10, 05:49

Quote:

Originally Posted by Ethan (EO)

I'd love to have DC factoring included in GPU to 72; it may only save one LL test, and it slightly reduces our insight into the reliability of LL tests, but for me personally it is very appealing.

Ethan (EO)

Penny wise, pound foolish.

DCs for the foreseeable future are already TFed to 68.
LLs should accordingly be TFed to 72.
They ain't.

David

PS note the correct usage of "already" you
Newspeak guys.

Dubslow · 2011-11-10, 07:29

I an currently running around half and half, but as soon as my current (massive) batch of TF-DC is done, I'm gonna stop that. About a month ago the assignments were half as long, but this batch I have at least is pushing what I see as the "worth-it" boundary. (That is to say, I'm eliminating no more than the same amount of factors for DC as for LL, and the theoretical breakeven point is when I find twice as many DC factors as LL factors; that point is long gone.)

2011-11-10, 01:26	#39
Chuck May 2011 Orange Park, FL 3×5×59 Posts	TFing before double check? Now that this is up and running without (as far as I can see) any hitches, is there any interest in incorporating ckdo's TF before DC into the process? Do others see this project as worthwhile? I still have about 770 exponents in the 28.9M range I am taking from 68 to 69. Chuck

2011-11-10, 07:29	#44
Dubslow Basketry That Evening! "Bunslow the Bold" Jun 2011 40<A<43 -89<O<-88 1110000110101₂ Posts	I an currently running around half and half, but as soon as my current (massive) batch of TF-DC is done, I'm gonna stop that. About a month ago the assignments were half as long, but this batch I have at least is pushing what I see as the "worth-it" boundary. (That is to say, I'm eliminating no more than the same amount of factors for DC as for LL, and the theoretical breakeven point is when I find twice as many DC factors as LL factors; that point is long gone.) Last fiddled with by Dubslow on 2011-11-10 at 07:30

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Stockfish game: "Move 8 poll", not "move 3.14159 discussion"	MooMoo2	Other Chess Games	5	2016-10-22 01:55
Aouessare-El Haddouchi-Essaaidi "test": "if Mp has no factor, it is prime!"	wildrabbitt	Miscellaneous Math	11	2015-03-06 08:17
Collaborative mathematics: the "polymath" project	Dougy	Math	11	2009-10-21 10:04
Would Minimizing "iterations between results file" may reveal "is not prime" earlier?	nitai1999	Software	7	2004-08-26 18:12

2011-11-09, 17:41	#34
Dubslow Basketry That Evening! "Bunslow the Bold" Jun 2011 40<A<43 -89<O<-88 1C35₁₆ Posts	Our discussion had nothing to do with TF bit levels, just that him getting P-1 assignments this way would remarkably reduce his P-1 throughput. As for what we're doing, it is at the wavefront, or rather, what the wavefront has missed. We are only TF'ing exponents that have had no LL tests. We are already doing everything we can for TF. I personally don't go to 72 in one go because I can get many more exponents through if I don't go as high. (Though as it stands chalsall is holding assignments until they get to 71, keeping in mind that we're 5-10M lower than what you're talking about.) Also, those TF's to 72 bits are only available through manual testing, so only we will get them. As soon as chalsall's spider is unable to saturate us, then we will move higher up the chain.