Primenet eyesores

[davieddy]

Originally Posted by Primeinator

Is this to say that exponents are being eliminated prior to DC even if they have had a first time LL?

This is one of the "eyesores".
Obviously TF/P-1 should be done before the first LL, and
the DC shouldn't need it.
Those "available" P-1s and "assigned" TFs in the 40-50M range are such (LLed once).
No urgency whatsoever to do anything with them.
George said to me:
"I goofed but no harm done".
True, but please erase them.
Or someone knock them on the head.

David

[Primeinator]

Originally Posted by davieddy

This is one of the "eyesores".
Obviously TF/P-1 should be done before the first LL, and
the DC shouldn't need it.
Those "available" P-1s and "assigned" TFs in the 40-50M range are such (LLed once).
No urgency whatsoever to do anything with them.
George said to me:
"I goofed but no harm done".
True, but please erase them.
Or someone knock them on the head.

David

Thank you for clearing this issue up!

[Christenson]

Originally Posted by Primeinator

Is this to say that exponents are being eliminated prior to DC even if they have had a first time LL? Sorry...I've had a couple margaritas and now some sherry...so I may have misinterpreted the implications of the above posts!

Ckdo and some helpers, including me, did some additional TF on exponents definitely in the DC range to 69 bits, and certainly turned in a few factors, thus completing several proofs of compositeness with only a single LL and eliminating the LL-DC. Primenet, of course, DCs factors when they are turned in. In the very low range, below 10M, I believe Petrw1 has been running P-1 when it wasn't run well, and finding factors, too. There's nothing to stop you from asking GPU to 72 for more of these 25M-30M exponents, and no reason to believe you won't find more factors if you do. But the optimum, in terms of minimising total effort to find M48 and prove it is indeed M48 and not M49, is to first try to elimnate 2 LL tests with TF'ing below 72 bits at approximately 50M, before first LL tests, until exponents at that level get hard to find with TF below 72 bits. We do try not to do TF on in-progress LL tests.

This is because there is a multiple bitlevel difference in optimal TF between a 25M exponent with one completed LL and a 50M exponent with no completed LL on it. First, the LL tests are 4 times as hard to do at 50M (2x as many iterations, 2x FFT size), and if a single LL test is eliminated on the DC range, the savings are only half. Second, 70 bits TF at 50M is approximately only half as much work as 70 bits at 25M because there are only half as many factor candidates. Finally, the probability of finding a factor by TF between 69 and 70 bits is approximately 1 in 70, regardless of whether we TF a 25M or 50M exponent.

So yes, we have TF'ed exponents that have had single LL tests, and will do so again in the future, as soon as the yield of work saved to GIMPS on exponents with no LL test drops far enough. Some of us also run DCs on our GPUs, since there is a very limited yield (10%-20% fewer exponents to LL test) to increasing the amount of TF work that can be done.

[Primeinator]

Originally Posted by Christenson

Ckdo and some helpers, including me, did some additional TF on exponents definitely in the DC range to 69 bits, and certainly turned in a few factors, thus completing several proofs of compositeness with only a single LL and eliminating the LL-DC.

....

So yes, we have TF'ed exponents that have had single LL tests, and will do so again in the future, as soon as the yield of work saved to GIMPS on exponents with no LL test drops far enough. Some of us also run DCs on our GPUs, since there is a very limited yield (10%-20% fewer exponents to LL test) to increasing the amount of TF work that can be done.

Thank you. However, looking at the stats page it seems like there is a huge need for DCs at the moment. There are still over 300,000 exponents remaining before M47. Are there any rough estimates on how many years THAT is going to take?

[Dubslow]

I personally recently set myself the goal of keeping at least as many DC's completed as LL. That way, around 80% of my total CPU time goes to LL, while at the same time I do my fair share of DC. I also maintain a similar policy for P-1 work relative to the others.

[petrw1]

Originally Posted by Dubslow

I personally recently set myself the goal of keeping at least as many DC's completed as LL. That way, around 80% of my total CPU time goes to LL, while at the same time I do my fair share of DC. I also maintain a similar policy for P-1 work relative to the others.

For what it's worth; lifetime GIMPS I show 500 LL and 540 DC.
Prior to v5 server DC was not counted seperately so my LL is a bit overstated and DC a bit understated--- by 20 or so.
Over this past year I focused primarily on P-1 with a few diversions. I should be close to 2,000 for the year by year end.
My "tentative" goal for 2012 is 1,000 DC's.

[Primeinator]

Originally Posted by petrw1

For what it's worth; lifetime GIMPS I show 500 LL and 540 DC.
Prior to v5 server DC was not counted seperately so my LL is a bit overstated and DC a bit understated--- by 20 or so.
Over this past year I focused primarily on P-1 with a few diversions. I should be close to 2,000 for the year by year end.
My "tentative" goal for 2012 is 1,000 DC's.

One thousand? How many computers do you have??

[Christenson]

I think that's probably 2-3 high-end GPUs...especially if PetrW means "total 1000DCs by the end of 2012" and not "1000DCs completed during 2012".

And by the way, we'll be very upset (NOT!) if you help with the DC effort!!!
...whether by DC on P95, DC on GPU with CUDALucas, P-1, or TF...

[Primeinator]

Originally Posted by Christenson

I think that's probably 2-3 high-end GPUs...especially if PetrW means "total 1000DCs by the end of 2012" and not "1000DCs completed during 2012".

And by the way, we'll be very upset (NOT!) if you help with the DC effort!!!
...whether by DC on P95, DC on GPU with CUDALucas, P-1, or TF...

If I can get a hold of an older machine I will for sure! I devote my quad core to first time LLs of record size due to its superior number crunching ability. I have an older laptop and an ancient desktop that if I can get them up and running again will dedicate them to the DC effort. I may also get my hands on a somewhat newer Mac desktop of some kind. Depending on its specs I will either dedicate it to DC or LL.

[Dubslow]

I'll need a laptop sometime soon, and when I get it, it's going on P-1 duty. (Gonna have 1600MHz native memory, hopefully!)

[Christenson]

Originally Posted by Primeinator

If I can get a hold of an older machine I will for sure! I devote my quad core to first time LLs of record size due to its superior number crunching ability.

Here's a wacky thought: The next mersenne prime to be found will be in the DCs...it had an erroneous non-zero residue turned in by someone.....it's not too wildly unlikely...especially with three different residues now on the last exponent to be DC'd below 24M....time to think about a GPU and seeing what it can do with CudaLucas...