Cause this don't belong in the milestone thread

[Prime95]

rcv: Exponents above a certain point are currently exempt from the expiration rules. I'd have to look up what the limit is, but certainly all 100M digit numbers are yours until the wave catches up.

There is one small advantage to lax expiration rules. It seems there are almost always poachers active. They find the smallest exponent and test it regardless of the assignment progress. If the smallest exponent is 400 days overdue, then at least the poacher didn't step on any toes.

Originally Posted by davieddy

What I consider to be the most disappointing aspect of GIMPS is that ~80% of LL tests are not completed despite being given the clear warning of expected completion date and the need for patience.

I think the majority of exponents that are expired are from torture test users. Despite the welcome screen that tells them not to join GIMPS: they click join, get an exponent, run the torture test, never run prime95 again.

[bcp19]

Originally Posted by davieddy

You just resent me PERIOD.
What would you expect a 2Ghz Core2 to produce per year?
Mine is one of 10,000 other "core2s" trying to find a prime rather than pissing about. Our syndicate's chance of finding a prime in the next year is ~1/6.
I complete an LL test every 50 days or so.
What I consider to be the most disappointing aspect of GIMPS is that ~80% of LL tests are not completed despite being given the clear warning of expected completion date and the need for patience.

D

Actually, you are wrong, I do not resent you, though I do often resent your attitude. I personally think you have some very good ideas, after all GPU72 was formed due to one of them, but you tend to harp on minutae and your facts are often skewed by your perception and lack of willingness to accept other points of view, even when true. Most of the blame you heaped on GPU72 and chalsall was actually the fault of the GIMPS system's way of doing things, yet you persisted on ignoring those facts.

FYI, I meant no ridicule in my previous statement, I was trying to take the points of view expressed by others and see if there could be some sort of common ground available or even a workable compromise. To wit: "plum" assignments. From the comments made thus far, I deduce that these assignments have "value" to certain people. Now, if they have value, why not make them available to those who feel the value is worth paying for? While it would require some code changes to implement, this could even be used to deter poaching in the trailing edge of the LL wave. Imagine a setup where all new "plums" are only given out to people who either 1) prepay for them or 2) have a monthly bill sent to them. Ideally, if a poacher turns in a result on one of these, they could either get 1) a notification that they have been billed for the exponent or 2) get a warning that their submission was not authorized and no credit will be given. If they are billed and ignore the billing, then their account could be prevented from receiving any further credit, even on 'valid' contributions. If the poached exponent turns out to be a prime and someone has paid for the exponent, then they should be the one to get credit for the discovery and the poacher would then become a double check on it when the original person turns theirs in. With the above system in place, would you be willing to pay $6-7 a year to only receive "plum" assignments? It would cost me ~$10-12 a month if I chose such an option, which is less than the cost of the subscription fees on the games I play. In addition to being a small price to pay, there is the obvious benefit to the system, allowing for more growth and potentially a greater payout when primes are found.

Now for the information you will likely ignore: when you PM me a link, I just hit delete, as I do not care to examine some(to me) random garbage to try and figure out what obsure reference you are trying to point out. Stop wasting your time.

[davieddy]

Originally Posted by Prime95

I think the majority of exponents that are expired are from torture test users. Despite the welcome screen that tells them not to join GIMPS: they click join, get an exponent, run the torture test, never run prime95 again.

Yep, and as you, I and presumably many others do, immediately return an assigned exponent if we deem its "plummy" coefficient not up to scratch!
I have long ago stopped moaning about folk doing what they fancy in most of the billion range of exponents, but it would be helpful if "time-wasters" avoided the 50-70M range.

David

[Prime95]

Originally Posted by davieddy

I have long ago stopped moaning about folk doing what they fancy in most of the billion range of exponents, but it would be helpful if "time-wasters" avoided the 50-70M range.

We could assign first-time users bigger exponents. Unfortunately, I think this sends the message: "Welcome to GIMPS, here is a real crappy exponent just for you".

@bcp: I don't think paying for exponents is workable or desirable. From a workability standpoint, we'd need to eliminate reports that let users see which exponents haven't been tested and the progress being made (eliminating these reports would eliminate poaching, too.) I'm of the opinion that more information is better for users than less. As to the desirability viewpoint, GIMPS has always been free and I'd rather keep it that way. But, thanks for thinking "outside the box".

[chalsall]

Originally Posted by Prime95

We could assign first-time users bigger exponents. Unfortunately, I think this sends the message: "Welcome to GIMPS, here is a real crappy exponent just for you".

I don't think it sends that message at all.

Later in the day (UTC), everyone currently usually gets candidates up in the 58M range.

I see no reason why someone new (usually a Torture Tester, as you've said) would even notice if they were assigned a high candidate. And if they don't even bother to register on Primenet ("Anonymous"), or use the Manual Reservation system without logging in, then I think they should definitely be assigned higher work.

To share my experience with GPU72, I introduced such a policy fairly early on when some low-throughput individuals always came by early in the morning to claim low candidates -- sometimes years worth of work. Although there were a few complaints, it only took a little while to find a balance for the heuristics between how many GHz Days / Day were historically rendered to where the cut-off should be.

As a side effect of this, anyone new weren't assigned low candidates, and so we didn't have to wait a month before the assignments were auto-expired.

[Batalov]

Originally Posted by Prime95

We could assign first-time users bigger exponents. Unfortunately, I think this sends the message: "Welcome to GIMPS, here is a real crappy exponent just for you".

"Welcome to GIMPS, here is an excellent exponent just for you; if Mp is prime, it will take decades to beat your record!"

[NBtarheel_33]

Originally Posted by davieddy

With multiple cores, and AVX processors on stream, one could reasonably have hoped for a dramatic increase in the rate of LL completions.
This has not occurred.

David

But what in fact is the ratio of these new processors to the older ones, e.g. Pentium 4 or even the first-generation Cores? GIMPS receives a large percentage of its contributions from businesses and universities, both of which in many cases are (1) shy of switching from XP to Vista or 7 (or 8), due to incompatibility or unfamiliarity (or, yes, cost!); and (2) able to run what they need to run just fine on systems that are a few years old (P4s are more than enough for basic office work, Web browsing, etc.). I would wager that a large percentage of curtisc's systems, for instance, are single-core P4's with perhaps a smattering of earlier multi-core Core systems thrown in. (Wait, I have a test account that is a member of his team, let me check:

Now, of course, there are the extreme computing hobbyists who purposefully build systems and push them to the max just for DC projects, but they are a small contingent. You might have a few stress testers that get hooked on the project, or feel the need to give something back in exchange for George's work on Prime95. And you always have those who lie in the intersection of those savvy (and wealthy) enough to buy the best available consumer machines, and those who have an interest in GIMPS. But I think that, by and large, the average GIMPS user is going to be running on technology that is somewhat below the "latest and greatest".

Some math: Say the newest CPUs give GIMPS a 50% speed boost (even this is nebulous; 50% vs. what?). If only 5% of GIMPS' cores can take advantage of this boost, then GIMPS throughput goes from X to .95X + (.05)(1.50)X = 1.025X, which is an increase in overall throughput of 2.5%! Keep in mind that this is also assuming that these new, late, great cores are running 24x7 AND are running 24x7 solely for GIMPS!

The lesson to take away from this is that the latest, greatest technology is certainly a great boon for GIMPS, but do not expect to see major advances in throughput overnight. It will be a long time before the majority of GIMPSters own an Ivy Bridge (or even a Sandy Bridge, or even a Nehalem), let alone a GPU capable of making a serious contribution. Read the Dell forums and discussion groups - there is a whole legion of folks out there who are happy with (and actively seeking spares of) their Northwoods and Prescotts.

[Dubslow]

Originally Posted by NBtarheel_33

Now, of course, there are the extreme computing hobbyists who purposefully build systems and push them to the max just for DC projects, but they are a small contingent.

I wouldn't call it small, per se, but the large majority of them do the other DC projects that are scientific with "legitimate use". They typically participate in GIMPS only to gain standing in sites like this one. (This is among the reasons I suggested porting Prime95 to BOINC ().)

Edit: Some might remember when phoenicis was challenging nucleon (he's now #20); he's from the first site I linked to. (I had a friend from high school who's a member there ask me how to use mfaktc (there's more impetus for updating the PDF guide properly -- that guide is written by phoenicis, actually).)

[kladner]

Originally Posted by Dubslow

I wouldn't call it small, per se, but the large majority of them do the other DC projects that are scientific with "legitimate use". They typically participate in GIMPS only to gain standing in sites like this one. (This is among the reasons I suggested porting Prime95 to GIMPS.)

How come Team GPU to 72 isn't listed?

EDIT: Don't you mean "porting Prime95 to BOINC" or some other? I mean, P95 is GIMPS.

[davieddy]

Originally Posted by Prime95

We could assign first-time users bigger exponents. Unfortunately, I think this sends the message: "Welcome to GIMPS, here is a real crappy exponent just for you".

THX George. This has kept me laughing all day!
ATM my expo should complete by the 2nd Sept.
~50M, and already had a "suspect" test. I would have returned it were it not for the the info that my friend Dubslow had TFed it from 70 to 72.
Should it turn out to be prime, I shall certainly give him some marks for good behaviour!

D

[Prime95]

I just changed the server to update the double-check and first-LL preferred assignment thresholds daily. The lowest 3000 exponents will be considered preferred.

Let me know if it doesn't update properly over the next few days.