[QUOTE=dragonbud20;410163]Did I get any of that right?[/QUOTE]

Not exactly... :smile:

The reason we try to go to 75 "bits" currently is that is where the "cross over point" is. That is, where it is more efficient to TF on a GPU (based on the statistical probability of finding a factor) than running *two* LL's on the _same_ GPU. This is based on an extensive analysis that James did for us many moons ago.

The reason GPU72 releases some candidates without P-1 done is many do P-1'ing directly through Primenet. Ideally we'd only ever release at 75 bits, but sometimes we're a bit short, so we release at 74 instead. This isn't the end of the world; any such candidates are then recaptured for TF'ing to 75 bits if the P-1 run didn't find a factor. In fact, except for very unusual situations, no candidates are assigned for LL'ing at less than 75 bits nor without P-1 being done.

One last point -- my understanding is it is better for the P-1'ers to work with higher TF'ed candidates because it allows them to search a higher range. Someone with a better understanding of the math would have to explain exactly why.

Edit: Should have been clearer. No candidates are released for LL'ing at less than appropriate TF'ing; below 70M that's 74, above 70M that's 75.

Edit 2: Sorry... After far too busy a day dealing with Humans dealing with Atoms (badly)... My above edit should have been "below 65M that's 74, above 65M that's 75.

This morning, one of my GTX 580s completed a DC assignment, which matched the existing result. I submitted it on the Manual Results page about 4.5 hours ago, and it still has not shown as completed in GPU72. I obtained the assignment via the P95 proxy, and moved it to the CuLu worktodo.txt. PrimeNet shows it as successfully completed.

Have I bollixed things in some way?

[QUOTE=kladner;410346]Have I bollixed things in some way?[/QUOTE]

Please PM me the candidate in question and I'll drill down.

Completed LL/DC work which goes through the proxy will be seen immediately, but there is a secondary spider which checks for such completions for cases where the candidate is checked out via the proxy but submitted manually; that can take a bit longer.

I'd be interested in knowing exactly what is happening in this particular case, to ensure the system is sane.

[QUOTE=chalsall;410347]Please PM me the candidate in question and I'll drill down.

.....[/QUOTE]
Done.

EDIT: And the Eagle has landed! Thanks!

[QUOTE=kladner;410348]Done.[/QUOTE]

OK, thanks.

I reran the secondary script with a larger number of queries, and it found your completion without any further modifications to the script. As in, the code was sane, but it might have taken up to six hours to find this.

As always, if anyone sees anything strange, please bring it to our attention.

[QUOTE=chalsall;410349]OK, thanks.

I reran the secondary script with a larger number of queries, and it found your completion without any further modifications to the script. As in, the code was sane, but it might have taken up to six hours to find this.

As always, if anyone sees anything strange, please bring it to our attention.[/QUOTE]

Thanks for the explanation. Even six hours would not be disturbing if the possibility of the delay is known.

It might not be so bad to run P-1 before TF74-75.
If P-1 (before TF75) finds a factor you 'save' the TF74-75 test.
If you do TF74-75 first and find a factor you 'save' the P-1 test.

The question is: which one would you rather save? A day of CPU core time (P-1) or a couple of hours on a GPS (TF74-75)?

[QUOTE=VictordeHolland;410397]
The question is: which one would you rather save? A day of CPU core time (P-1) or a couple of hours on a GPS (TF74-75)?[/QUOTE]

That's comparing apples and oranges. Better would be to compare GPU TF74-75 to GPU P-1. I don't know if we have benchmarks for GPU P-1 program.

On the other hand, it may be a bit much for GPU72 to manage releasing the exponent for P-1 and reacquiring it for TF75.

[QUOTE=Prime95;410398]On the other hand, it may be a bit much for GPU72 to manage releasing the exponent for P-1 and reacquiring it for TF75.[/QUOTE]

Actually, that isn't a problem at all. In fact, it's already being done -- if "Spidy" sees a candidate about to be assigned for LL'ing at 74 it grabs it to take to 75.

But ideally the sequence would be TF'ing to 75, then P-1, then LL. airsquirrels has agreed move a lot of his firepower over to LLTF'ing for a while, so we should be good.

I was thinking about the predicament we had the past few days with a lack of TF power and it though of something that might be useful for figuring out what type of work to do. Would it be possible to have some sort of indicator on the assignment page or anywhere really on the gpu72 site that indicated when TF or DC needed more work done or we don't have enough buffer for one or the other? I personally do all my assignments manually so I know it could help people who do the same although I'm not sure what could be done for people who get all of their work using misfit or spiders or whatever(I'm not familiar with automating mfactc at all I don't know if those terms are right). I don't know if this has been considered but it would be a nice little feature.

[QUOTE=chalsall;410473]But ideally the sequence would be TF'ing to 75, then P-1, then LL.[/QUOTE]
Actually, I think the ideal sequence for a nvidia gpu would be "TF to 73, P-1, TF to 74, forget about 75" :razz: (well, I didn't switch to new cards like 9xx yet, which seem to be more efficient to TF, but at least this is the case for 580s and Titan, which are very efficent on LL, and almost sure that is the case unde 70M exponents). What you say may be still ok for an amd gpu, which can't do efficient LL.