Who is ruining GIMPS? (The broken record, Side B)

[Prime95]

Some data from the server. LL assignments since Oct. 1.

exponent    bits factored      number
62M           73                      290
62M           74                      819
63M           74                      604
64M           74                      643
65M           74                       87
66M           74                      455
67M           71                        4
67M           72                        1
67M           74                      226

[davieddy]

Originally Posted by chalsall

GPU72 has an automatic "rip-cord". If it sees that any candidates above 63M are about to be assigned which have not been TFed to at least 74 bits (or not P-1ed), it will release a batch in 62M which are TFed to "only" 73 (and P-1ed).

Originally Posted by chalsall

please rest assured that no LLer will ever be given a candidate above above 63M TFed any less than 74.

Originally Posted by davieddy

AND PIGS MIGHT FLY.

Originally Posted by TheMawn

https://www.gpu72.com/reports/estima...tion/primenet/ According to the 30-day moving average of LL and TF firepower, all exponents below 66M will be trial factored to 74 bits in 213 days and it'll take a year on top of that to LL all exponents below 66M. How is this data insufficient? What else is there to say?

Originally Posted by Prime95

Some data from the server. LL assignments since Oct. 1.

exponent    bits factored      number
62M           73                      290
62M           74                      819
63M           74                      604
64M           74                      643
65M           74                       87
66M           74                      455
67M           71                        4
67M           72                        1
67M           74                      226
                                          ____
Total                                   3129

Many thanks for this data, George.

I assume it is for 1-4 Oct, so we would expect VERY roughly that 1200 would be "brand new", the remainder having been returned.

Note that 67M doesn't yet fall within GPUto72's purview, consequently the pigs are still airborne for the time being.

I don't think Primenet should be letting 71 and 72 bits be getting assigned.
Of course 73 bits is Chris and my bone of contention!

Since Chris appears to have pulled his ripcord 290 times, I seem justified in my refusal to grant pigs a certificate of airworthiness

D

[kladner]

Originally Posted by Prime95

Some data from the server. LL assignments since Oct. 1.

George posted some numbers. Isn't it a bit indefinite as to what those numbers reference? There are, after all, other pools than GPU72 from which fish can be caught. It seems conjectural to assign different sets to particular work flows.

Perhaps the more perceptive folk about can enlighten me.

[chalsall]

Originally Posted by kladner

Perhaps the more perceptive folk about can enlighten me.

George has provided some important empirical data from his independant perspective from within PrimeNet. This is data which only he, Scott and James has access to -- thank you for this George.

What this shows is that my claim that "no candidates above 63M will be assigned with less than 74 bits of TFing" is mostly true, but not 100%.

Reading the report, it shows that in the time period covered 3,129 candidates were assigned for LLing above 62M.

1. 290 in 62M were assigned factored to 73 "bits.
2. 819 in 62M were ... 74.
3. 4 in 67M were assigned factored to 71.
3.1. 1 in 67M ... to 72.
3.2. 226 in 67M ... to 74.

Or, in other words, 3,124 out of 3,129 were assigned with the communally agreed TF level. Doing the math, 99.84%.

I don't know for sure how those 5 candidates "slipped through". There are a couple of possibilities... Either they were explicitly requested by the worker, or suddenly a request was made for assignments from PrimeNet in-between the times that Spidy checks the situation, and fires the "safety valve" (which releases several, not just one, candidates) if needed.

At the end of the day I don't really think 0.16% is that much of a mountain...

So everyone knows, this shouldn't be too much of a problem for too much longer -- some additional LL TF firepower has come online....

[kladner]

Thanks for elucidating, Chris. (EDIT: .....or for presenting your take, if that is a more tasteful way of putting it.)

[chalsall]

Originally Posted by davieddy

Those pigs take a lot of firepower to keep them airborne.

It could be argued the same of you.

[chalsall]

Originally Posted by davieddy

You can't keep a good man down.

Agreed.

[garo]

I do not wish to prolong this thread unnecessarily but since my name was mentioned I would like to say that I revised my views on releasing all 63M immediately after I corresponded with chalsall via PM and he explained his automatic release valve. That and looking at George's stats is enough to convince me and I am minded to lock this thread if davieddy carries on any longer.

[firejuggler]

Ignore davieddy, that's the easiest way.

[davieddy]

Originally Posted by davieddy

Being a shit-hot assembly programmmer I promptly wrote my own LL test. With the aid of "The Math" page from GIMPS I also wrote a trial factoring program.

I ran the LL test on a pentium2 350 and a K6 450, and did the TF on a 486 66. Within a few months I had confirmed that M29 was in fact M29.

I actually used "high school" squaring.

Now the appropriate TF bit level rose from tiny to about 54.
How would you determine the appropriate moment to increment it?

This entire "project" was undertaken purely for enjoyment (rather like all the "work" I've ever done in my life). Do you see how it has helped my judgement of when GIMPS should increment TF bit level?

Anyone else's credentials even come close?

David

[chalsall]

Originally Posted by davieddy

Anyone else's credentials even come close?

Please forgive me for this David.

But are you familiar with the concept of a vacuum?