New work types for GPU72?
 

Hey all.

Since we're only a few weeks away from taking everything below 60M to at least 72 bits, I thought now might be a good time to talk about the possibility of adding some new work types coordinated by the system. Some possibilities which come to my mind:

1. DC P-1.

There are some who are cleaning up candidates which have had a LL completed, but no P-1 done. Would anyone here like to have such work available and tracked by the GPU72 system?

2. Great Internet Mersenne Factor Search (GIMFS).

This would coordinate those wishing to take candidates which have already had a LL and a matching DC completed, but are interested in doing additional factoring work. It is understood that this would not help the GIMPS effort at all, but again there are some who enjoy doing this type of work, and having it coordinated through GPU72 would mean there would be less chance of "toes being stepped on".

Note that this would actually be two different sub-work types -- GIMFS TF, and GIMFS P-1 for previous P-1 done "poorly".

Now, to be clear, I created GPU72 to help the GIMPS effort, and GPU72 will always have TF assignments available to process candidates ahead of the two "wave fronts". But at some point the need for the work will be less critical, and some might enjoy doing other work, including George's suggestion of moving the GPUs from TFing to doing LL/DCing.

Thoughts? Any other work types anyone would like to see?

P.S. davieddy, please let us know if you need us to call an ambulance for your coronary.... :wink:

[QUOTE=chalsall;302772]Hey all.
1. DC P-1.

There are some who are cleaning up candidates which have had a LL completed, but no P-1 done. Would anyone here like to have such work available and tracked by the GPU72 system?
[/QUOTE]

I'm currently working on this. doing both the TF to 71 and P-1 afterward.

I would like it very much if the exp that need P-1 were TF to 71. It would mean allot less manual work for me.

As for P-1, I am doing and assuming each factor saves 2.6 LL Test.

Result example:
M44962261 completed P-1, B1=615000, B2=13991250

Most of them were LL test done by curtis and I am assuming the stage1 was done but never reported because it was stuck on stage2 because not enough memory was assigned to stage2 (8MB)

So if gpu272 were to assign P-1 work with the assumption that only 1 LL test will be saved if a factor is found, the work will be done for very little ROI.

Saves 2 LL:
Pminus1=1,2,44962261,-1,495000,12375000,68

Saves 1 LL:
Pminus1=1,2,44962261,-1,235000,4700000,71

Also going from 68 to 71 further reduce the bound, that's why I go to 2.6 test saved, it gives me the amount of P-1 worked that would have been done without further TFing.

Some simple alternatives:

TF to 73 down to let's say 55 Millions
TF to 73 up to 6x Millions
TF to 74 from 6x Millions up to 75 Millions (in a few years)

Probably, TFing will always be in front of the LL wave, therefore some people switching to LL on GPU looks more effective to me for the GIMPS project.

[QUOTE=chalsall;302772]
Note that this would actually be two different sub-work types -- GIMFS TF, and GIMFS P-1 for previous P-1 done "poorly".[/QUOTE]
That might be 3, counting the GIMFS-ECM.
And indeed, it doesn't help GIMPS too much, but some of us would like to have it (me included).
There was a time when I was looking for someone/something to coordinate the work on 100M-digit-TF, this was when I helped uncwilli to take the 332-334 range all up to 69 bits. That DOES help GIMPS, in its "ultimate" goal of taking the 150k EFF prize. And there is nothing which coordinate the TF effort in that area (except PrimeNet whose coordination is poor-to-zero when bitlevel gets over the limit). I think you should start with this: TF for 100M expo.

[QUOTE=Radikalinsky;302781]TF to 73 down to let's say 55 Millions
TF to 73 up to 6x Millions
TF to 74 from 6x Millions up to 75 Millions (in a few years)[/QUOTE]

In order: already planned, already underway and already planned.

[QUOTE=Radikalinsky;302781]Probably, TFing will always be in front of the LL wave, therefore some people switching to LL on GPU looks more effective to me for the GIMPS project.[/QUOTE]

Agreed. But there are some who enjoy the "immediacy" of finding factors.

[QUOTE=LaurV;302782]I think you should start with this: TF for 100M expo.[/QUOTE]

I made a public offer to help there, but it was generally agreed that Uncwilly is handling the 332M range well enough. But if it is desired to revisit that (only with Uncwilly's agreement), it would take me almost no time to facilitate.

With regards to LMH and ECM, I don't really think I could offer anything which PrimeNet itself doesn't already do well enough.

[QUOTE=diamonddave;302777]I would like it very much if the exp that need P-1 were TF to 71. It would mean allot less manual work for me.[/QUOTE]

OK. Assuming there are no objections, consider this on my "todo" list.

I would want to coordinate with you and James to make sure I have the bounds being logically set.

"Something old" type of work to do for you: please separate the LL and DC lines in the "view assignments/prod heuristics for user xxxx" table. The amount of combined LL always confusing me... :smile:

How many candidates are we releasing without P-1? How many of those do not get P-1 from primenet before releasing.

I can ramp up P-1, but that would require me to take some of my graphics cards offline. Since it is 100 degrees today, that would not hurt my feelings.

Let me know how P-1 stands.

[QUOTE=KyleAskine;302789]Let me know how P-1 stands.[/QUOTE]

Since a couple of "big guns" have joined our P-1 efforts, we're now not releasing any candidates without P-1 done below 56M. And we're completing more P-1 than there are LL completions.

In short, because of the daily candidate "churn" (old assignments expiring) and our efforts, PrimeNet is now not handing out any assignments TFed to less than 72, nor without P-1 done.

Right now the most important thing is to stick with exactly what we're doing -- please don't move to P-1 at the detriment of TFing (at least, not yet).

Chalsall, I'm going to follow what you decide. :smile:

Anyway, I'd like to advance the DC wave, so I'm going to use LL-DC more often. That's why I'd be more interested in keeping the DC TF ahead of limits, and eventually rising one bit the DC-TF limits, if some TF work should be chosen.

Luigi.