GPU to 72 status...

[TheMawn]

Shouldn't factoring 63M to 74 be the priority then? Or is that the thing about some user having a whole bunch of 63M reserved and not handing in results?

[kracker]

Originally Posted by TheMawn

I really don't understand P-1. Does having more memory speed up the process or increase the range of prime factors that might be found? Having 1,000 GHz-Days of P-1 puts you in the top 100. Having 10,000 GHz-Days puts you in the top 10. Obviously we don't need very much P-1 if P-1 is keeping ahead of TF.

Um, compare GHz days on doing a exponent with P-1 vs TF?

[kladner]

My P-1s are all ~63.5M. They yield ~ 4 GHz-days apiece.

A ~63M TF from 73-74 coughs up ~30 GHz-days. I crank out a lot more of those. Such are the joys of 2 GPUs. The P-1s are something to keep the CPU cores busy in the meantime.

[chalsall]

Originally Posted by Prime95

I do not have a strong opinion. If we are currently assigning 66M LLs that are only TFed to 73, then we should be assigning the 63Ms instead. If the 66M assignments are TFed to 2^74 and Chris believes he has the firepower to both keep ahead of the LL wavefront and get the 63Ms to 2^74, then I'm OK with holding the 63Ms back.

Copy. Thank you.

And trust us....

[TheMawn]

Originally Posted by chalsall

Copy. Thank you.

And trust us....

Besides... If he messes up, you can just turn Makes Sense^TM assign double-checks until we catch up

[Mini-Geek]

Originally Posted by TheMawn

I really don't understand P-1. Does having more memory speed up the process or increase the range of prime factors that might be found?

If you ask that questions for fixed bounds (B1 and B2), then more memory speeds up the stage 2. But Prime95 takes the amount of memory allowed into account when it chooses the bounds. So in short, the answer is "maybe both, maybe just one, and maybe neither".

Having more memory speeds up stage 2. If you have more memory, then, you increase the range of prime factors that might be found, while the time required may go up or down (whatever it takes to find factors most efficiently).

If you have so little memory allowed that Prime95 decides to skip stage 2, it will choose a slightly higher B1 (i.e. a longer stage 1), but the overall test will take less time than with a stage 2.

So in practice, more memory might actually make the P-1 for a single number take longer, but with a higher chance of finding a factor, which means better efficiency for you and GIMPS (more factors cleared per time spent).

(something interesting I just realized: there's a slim chance that for a given exponent, you'd find a factor by assigning the minimum memory and skipping stage 2, and you'd miss that factor if you assign a generous amount of memory and thus choose a lower B1; this would occur if, e.g. the former B1=B2=660 000 and the latter B1=570 000, B2=11 000 000 and the k in the factor has 2 or more factors n with 570 000 < n < 660 000)

[cheesehead]

Originally Posted by TheMawn

I really don't understand P-1. Does having more memory speed up the process or increase the range of prime factors that might be found?

It can be either, depending on circumstances.

Here's one explanation of the two different ways of invoking P-1 and their different relationships between B1/B2 bounds, "available memory", speed, and probability of finding a factor.

Here's another version.

[I still haven't posted a grand amalgamated version at mersennewiki.org :-( ]

[garo]

Looks like Primenet is assigning 67Ms. SPE chalsall?

[chalsall]

Originally Posted by garo

Looks like Primenet is assigning 67Ms. SPE chalsall?

Nope.

Instead, very careful management. Note all those assignments are TFed to at least 74, and P-1'ed.

[kladner]

Originally Posted by garo

Looks like Primenet is assigning 67Ms. SPE chalsall?

SPE? (I'm sure it's obvious, but the meaning eludes me.)

http://en.wikipedia.org/wiki/SPE

Science and technology

• Serum protein electrophoresis
• Solar proton event, in which protons emitted by the Sun become accelerated to very high energies
• Solid phase extraction, a separation process used to isolate compounds from a mixture
• Solid-phase epitaxy, a method for forming crystalline materials from amorphous precursors
• The Sound Pattern of English, in phonology, a book by Noam Chomsky and Morris Halle, and method of notation of phonological rules
• Stanford prison experiment, in psychology, an infamous study of prisoner and prison guard roles
• Subgame perfect equilibrium, in game theory, a refinement of a Nash equilibrium used in dynamic games
• Synchronous payload envelope, in telecommunications, related to synchronous optical networking

Computer science

• Stani's Python Editor, a multi-platform IDE for the Python programming language
• Synergistic Processing Element, part of the Cell microprocessor used in the PlayStation 3 video game console
• Software performance engineering, in performance engineering
• Signal-Processing Engine, for example in the PowerPC e500

Organizations

• Schlumberger Limited, formerly Société de Prospection Electrique, the world's largest oilfield services corporation
• Sigma Phi Epsilon, a fraternity for male college students in the United States
• Societas Privata Europaea, a Europe-wide form of a limited company
• Sony Pictures Entertainment, an American-based entertainment company owned by media corporation Sony
• SPE (company), former name of the Belgian energy company EDF Luminus
• Special Police Establishment, predecessor of Central Bureau of Investigation, India
• Special purpose entity, a legal entity created to fill a narrow or temporary objective

Professional societies

• Society for Photographic Education
• Society of Petroleum Engineers
  • SPE Certified Petroleum Professional, a designation
• Society of Plastics Engineers
• Society of Professional Engineers UK

Other

• Supervisory patent examiner, a patent examiner, usually a civil servant, working at a patent office
• SPE Certified, a food quality assurance certification

[sdbardwick]

In this context SPE is Stupid Programmer Error (^TM chalsall).