Possible extention to the "GPU to 72 Tool" project?

[Dubslow]

Sweet!
Thanks

[petrw1]

Originally Posted by Dubslow

Sweet!
Thanks

I'll take that as a yes.
Call you tell mne how I go about doing it as this is a "GPU to 72" project and I do NOT have a GPU and do NOT intend to do factoring but P-1 instead

[chalsall]

Originally Posted by petrw1

I'll take that as a yes.
Call you tell mne how I go about doing it as this is a "GPU to 72" project and I do NOT have a GPU and do NOT intend to do factoring but P-1 instead

"GPU to 72" was simply a name for this sub-project. It rhymed.

We found that there was a lot more we could do with it, and are requesting feedback from those who might do other work....

[ET_]

Originally Posted by petrw1

If that (doing P-1 only) is an option I could help out there too.

I've got 9 cores available.

I'm working on P-1 too.

With 2GB of available memory, I'd rather take exponents from GPU to 72 than from Primenet

Luigi

[petrw1]

Originally Posted by chalsall

"GPU to 72" was simply a name for this sub-project. It rhymed.

We found that there was a lot more we could do with it, and are requesting feedback from those who might do other work....

Do I take this to mean P-1 only support isn't there yet?

Thinking out loud:

Could I ask for GPU TF assignments and then manually change the Factor= line to PFactor= lines ... with the appropriate parms?
Then I would need some way of unrequesting the TF assignment the system thinks I have.

[chalsall]

Originally Posted by petrw1

Could I ask for GPU TF assignments and then manually change the Factor= line to PFactor= lines ... with the appropriate parms? Then I would need some way of unrequesting the TF assignment the system thinks I have.

You could. But we ask you not to.

The "GPU to 72" system is based on trust. Much like PrimeNet.

P-1 work is next in the queue for the system. You and Mr. P-1, and others, have already expressed interest.

GPU.Mersenne.Info assigns each worker work, and then watches for completion of that work. Once that has been completed (and only then), it might be reassigned to others, or simply released back to PrimeNet.

[petrw1]

I took one assignment Factor=.......69,70
I added the PFactor line in front so it will complete first ....
Then I will still complete the Factor=

And no more until told otherwise

OK?

[Dubslow]

I'd just point out the obvious inefficiency of doing any TF work on the CPU. It will seriously cut into the P-1 you could do otherwise, even ignoring the GPU thing.

[petrw1]

Originally Posted by Dubslow

I'd just point out the obvious inefficiency of doing any TF work on the CPU. It will seriously cut into the P-1 you could do otherwise, even ignoring the GPU thing.

Understood. But probably only a day of TF (equiv to 1/2 of a P-1) for this 1 only assignment.

[Dubslow]

Well that's 1/3 less time for P-1, if you were to continue doing it like that. Your choice though.

*chalsall's choice, really.

[davieddy]

Originally Posted by Dubslow

I'd just point out the obvious inefficiency of doing any TF work on the CPU. It will seriously cut into the P-1 you could do otherwise, even ignoring the GPU thing.

Originally Posted by petrw1

Understood. But probably only a day of TF (equiv to 1/2 of a P-1) for this 1 only assignment.

Originally Posted by Dubslow

Well that's 1/3 less time for P-1, if you were to continue doing it like that. Your choice though.

*chalsall's choice, really.

The only reason for doing extra bits of TF is the advent of GPUs.

It's not Chalsall's choice, it's ultimately George's.
However, he released exponents ahead of the wavefront (then 53.5M)
for further TF (hopefully only by GPU) after finding MY arguments
"very persuasive". I have thought and posted copiously about the
sensible, SIMPLE strategy ever since Oliver announced the lightning fast
mfaktc.
All Chalsall did was to unjustly chastise me for focussing
on the LL wavefront, falsely accusing me of dictating what everyone
in GIMPS should be doing.

The goal is to TF new LL assignments (i.e. at the wavefront, not retiries)
as far as practical (72 ATM) by GPU in one go.
Now work out why this STILL isn't happening.

I may frequently act the buffoon, but I can be serious.

David

PS the progress of the wavefront relies on
tens of thousands of cores LL testing to completion (even if slowly in some cases).
Why should us few "reliable and enthusiastic"
folk hog all the low exponents?
We should be selflessly assisting
the vast majority of participants by TFing the
assignments to the max.

PPS I see Primenet has made 53M to 58M
available for TF.

GPUs should "grab" these NOW (before some
misguided CPU gets hold of them) and
take them straight to 72 bits.