Assignment confusion

[Chuck]

I am a little confused about these assignments...I have a number of 60,xxx,xxx TF 73 —> 74 assignments I got through GPUto72, but they show as being assigned for LL testing by PrimeNet.

http://www.mersenne.org/report_expon...0327517&full=1

http://www.mersenne.org/report_expon...0327517&full=1

There are quite a few more, I just picked a couple of examples. Is this some sort of start-up anomaly as we move to the new rules?

[Uncwilly]

Originally Posted by Chuck

I am a little confused about these assignments...I have a number of 60,xxx,xxx TF 73 —> 74 assignments I got through GPUto72, but they show as being assigned for LL testing by PrimeNet.

My understanding is that is how GPU72 spidey snags them, as LL's. That way, until Spidey releases them (after they have been completed), they are held by GPU72. PrimeNet won't let you grab a 'TF' for greater than the current limit. You have to grab it as an LL.

[TheMawn]

Yes. GPU72 also deals in LL's for GPUs (and CPUs but I won't go into that strange topic because I myself don't understand it) so it holds an assignment for TF up until the required level and then an LL.

Because GPUs increase the feasible TF level, GPU72 takes assignments that are "ready for LL" but TF's them some more and internally keeps them until sending them out for LL later. They can also be released back to primenet which detects "equal to or greater than optimal TF level" and reassigns them... as LL assignments, regardless of the optimal GPU level.

[axn]

Originally Posted by Uncwilly

My understanding is that is how GPU72 spidey snags them, as LL's. That way, until Spidey releases them (after they have been completed), they are held by GPU72. PrimeNet won't let you grab a 'TF' for greater than the current limit. You have to grab it as an LL.

I had the same question, and clarified it with Chris (over PM). Just continue with the assignments.

[LaurV]

True what Unc said. PrimeNet is not aware of GPUs. As the CPUs are lousy at factoring, the factoring limits are much lower. If you factor "over the limit" you reserve the exponent as LL, or DC, then you do your factoring, report the result, get the credit, and unreserve the exponent. The reserve/unreserve part is done by GPU72, when it detects that you did the TF bits you pledged for. That is why it says "don't do more bits than you pledged for", the exponent will be un-reserved and someone may duplicate your work in the same time with you, wasting time and resources.

[TheMawn]

Originally Posted by LaurV

That is why it says "don't do more bits than you pledged for", the exponent will be un-reserved and someone may duplicate your work in the same time with you, wasting time and resources.

I never knew why it said that...

[Chuck]

All of the factoring assignments I currently have from GPUto72 show in PrimeNet as assigned to "For Research" LL. Is this the user that GPUto72 is now using to reserve its assignments?

[chalsall]

Originally Posted by TheMawn

Because GPUs increase the feasible TF level, GPU72 takes assignments that are "ready for LL" but TF's them some more and internally keeps them until sending them out for LL later. They can also be released back to primenet which detects "equal to or greater than optimal TF level" and reassigns them... as LL assignments, regardless of the optimal GPU level.

Yes... To explain further, "Spidy" uses a few different techniques to get assignments which need additional TFing, including reserving candidates as LLs.

It used to then transfer them over to the "GPU Factoring" account as TF assignments, but with the new assignment rules this is no longer possible. So if you see anything assigned to you for TF'ing (or P-1'ing) owned by "For Research" don't worry -- that's my personal "trusted" account which I'm now using to fetch low LLs.

This also means that they don't appear as TF assignments on Primenet, but other than this everything's the same. The system will hold them until they're appropriately TFed, and then release them back to Primenet for actual LL assignment.