Let's finish PRP-CF DC's.

[pinhodecarlos]

Let’s put the date target where we would like both to join, shall we?

[Uncwilly]

Originally Posted by pinhodecarlos

Let’s put the date target where we would like both to join, shall we?

By March 15 would be a nice goal.

[storm5510]

Originally Posted by Uncwilly

...Have the server only hand out DC assignments to Prime95 v29 and lower. Let them do the DC work. Reserve PRP-CF first times for V30. In the long run this will save us cycles on more DC's that the old version creates. Any v30 that asks for PRP-CF-DC should get it.

Caveat: People wanting to run DC's must specifically select them from the list in Prime95, or do a manual reserve if they are using one of the GPU programs. Most do not want to run them. A possible solution would be to reclassify them as PRP's. Yes, this would add the extra burden of running the certifications. There may not be any other way to clean up the backlog. Something needs to change!

[Uncwilly]

Originally Posted by storm5510

Most do not want to run them. A possible solution would be to reclassify them as PRP's.

This is about the PRP'ing of the CoFactors of numbers with a known factor. Most people are not doing that task, much less the double checking of that. In 4 months or less it should be caught up with the front line. At that point any that get done by a machine that can't generate the certs should be few enough to not worry about. And the certs in this range are very short.

[pinhodecarlos]

I’m not a Linux guru, anyone can help me out creating a Linux step by step guide to get PRP-CF-D running on Linux environment? Let’s say I want to do this throughout the command line, with an account not yet available, set workers and threads, type of work. Can all of this be done by command line?

[storm5510]

Originally Posted by Uncwilly

This is about the PRP'ing of the CoFactors of numbers with a known factor. Most people are not doing that task, much less the double checking of that. In 4 months or less it should be caught up with the front line. At that point any that get done by a machine that can't generate the certs should be few enough to not worry about. And the certs in this range are very short.

I decided to run one of these, DC, with Prime95. 76 minutes to complete, it estimates. The last time I tried one of these was on my Xeon system. Over 7 hours to complete. That was a major no-sale for me. I think it was a first-time test.

I learned, after-the-fact, that CF's cannot be done with gpuOwl. It goes through the motions perfectly and generates a proof. At that time, I had ran over 20, submitted the results, and sent the proof's. I never heard of any problem with them.

[ATH]

Originally Posted by storm5510

I learned, after-the-fact, that CF's cannot be done with gpuOwl. It goes through the motions perfectly and generates a proof. At that time, I had ran over 20, submitted the results, and sent the proof's. I never heard of any problem with them.

Gpuowl probably ran the PRP test on the actual Mersenne number 2^p-1, not on the number divided by the known factors as we want: 2^p-1 / knownfactors

[storm5510]

Originally Posted by ATH

Gpuowl probably ran the PRP test on the actual Mersenne number 2^p-1, not on the number divided by the known factors as we want: 2^p-1 / knownfactors

This is about as simple an explanation as I have seen yet.

I am running CF DC's on my Xeon system with Prime95 v30. It takes about four hours each. That machine is not my primary so I do not pay much attention to it other than to check if it is still running properly.

I tried an experiment. I went to my Primenet account and copied the Xeon's current assignment line. It has one factor. I put the line is gpuOwl's "worktodo" file to do just a very short test, about 30 seconds. I wanted to see if any mention was made about the factor in the output. I saw this. "Validating proof residues for power 8." I removed the factor portion and started it again. The same line appears. So, I am not sure what it is actually doing.

No matter, it is running fine on my Xeon with Prime95 and I will leave it go at that.

[ATH]

It is not something that you would just expect to work.
I do not know for sure, but I think it is a fair amount of extra code in Prime95/mprime making it able to run PRP tests on 2^p-1 / knownfactors.
I wish gpuowl could do it, so we could finish them quickly.

The "bug" imo is that gpuowl accepts the worktodo line containing known factors and then just ignores them, instead it should fail so you know it does not work.

[LaurV]

Originally Posted by ATH

I do not know for sure, but I think it is a fair amount of extra code in Prime95/mprime making it able to run PRP tests on 2^p-1 / knownfactors.
I wish gpuowl could do it, so we could finish them quickly.

We wish too, especially when the "fair amount of extra code" boils down to just some division (see Robert's posts about).

And yes, we agree to the fact that gpuOwl, if it can't do the proper work, should give an error, instead of wasting precious time re-doing "quintuple" checks for already-done full PRP tests.

Mihai?

[kruoli]

Currently, in the range from 9M to 11M, there are 1,461 PRP-CF-DCs assigned. In total 36,156 exponents to go.

That means, we have a progress of around 453 exponents per day, so ETA with our current speed is around 80 days. Speed is increasing!