Massive new mover on Top Producers list (glitch!) - patched

[piforbreakfast]

WOW! Username "Dmytro of Dnipro" has jumped 624 spots in one day to claim the top spot on the top producers list doing nothing but P-1 tests. Total GHz days are actually hidden. I've never seen anyone move up the overall producers list that fast. Could someone have done this renting out hundreds or thousands of cores on AWS?

[ATH]

No, it is probably because of version 30.8 faster P-1 stage 2. You can do super large B2 values now, and then calculation of GHz-days have not been altered, so you get a ridiculous amount of credit.
I myself got 4000-9000 GHz-days for 1 curve at B2=10¹⁴, and 46,882 Ghz-days for B2=5*10¹⁴ I stopped doing them now.

This guy probably did like B2=10¹⁷ or 10¹⁸ or several of them, and got an insane amount of credit.

[Batalov]

The credit function should be adapted to reflect not the computational cost, but the computational benefit for the whole project as a baseline (then people will apply effort where the project needs it). And then tuned as needed (to attract resources).

It is very easy for someone to spend a lot of GHz days while moving the probability of the elimination of any given candidate by a percent of a percent. For example, - is the higher B2 limit changing the probability of a factor found from 6.5% to 6.7% (this is computable/modellable)? Then that's how much higher it should be "paid", and not "paid" 10^5 times more than with default B2.

People should not be encouraged to do senseless work by inflating its payoff in "tokens" *. But currently they are.
___
* not saying that it could not be a definite tuning mechanism, which of course it can.
PrimeGrid does that periodically: increasing pay for some project e.g. by 10%.
But if one suddenly increases some side channel's project cost 10^5 times, then surely everyone and their mothers will start doing just that.
People are easily manipulated.

[ATH]

Here are some results, but the limits does not look crazy here (see screenshot), but when you look at the exponents it is insane:
108017587 B1= 4 697 591 239 862 648 832 B2= 4 731 979 646 332 043 264
108017531 B1= 4 697 591 239 862 648 832 B2= 4 731 979 646 332 043 264

Those B1 values are fake stage 1 did not get faster, and B2 are fake as well probably, so maybe it is just fake results instead of using the new 30.8 version.


Edit: Maybe it is just a bug on the server? Because in the history of those 2 exponents the limits are fine.

[petrw1]

Originally Posted by ATH

Here are some results, but the limits does not look crazy here (see screenshot), but when you look at the exponents it is insane:
108017587 B1= 4 697 591 239 862 648 832 B2= 4 731 979 646 332 043 264
108017531 B1= 4 697 591 239 862 648 832 B2= 4 731 979 646 332 043 264

Those B1 values are fake stage 1 did not get faster, and B2 are fake as well probably, so maybe it is just fake results instead of using the new 30.8 version.


Edit: Maybe it is just a bug on the server? Because in the history of those 2 exponents the limits are fine.

Definitely faked. B1\B2 on top is much higher than values in individual events

[charybdis]

Looks more likely garbled than faked... here's are the binary expansions of the correct and incorrect values, aligned for convenience:

B1:
          100010011001001011000
100000100110001001100100101100000000000000000000000000000000000

B2:
          1101101011110011010101101000
100000110101011010111100110101011010000000000000000000000000000

[S485122]

There no evidence of fakery. A look at the Recent Results page shows the following results reported by that user :

User : Dmytro of Dnipro, Computer : AERO_15X_V8
Exponent   WorkT.  Date              Age   GHzdays  Result
108017587  NF-PM1  2021-12-09 16:32  5.3 *********  B1=1127000, B2=229586280
108017531  NF-PM1  2021-12-09 16:32  5.3 *********  B1=1127000, B2=229586280
123003787  NF-PM1  2021-12-09 16:32  2.8  123.3153  B1=1351000, B2=256825170
108042839  NF-PM1  2021-12-09 16:32  3.0   96.5328  B1=1127000, B2=229586280
108042829  NF-PM1  2021-12-09 16:32  3.0   96.5328  B1=1127000, B2=229586280

I'd vote to have that user keep its "server glitch" score :-)

[chalsall]

Originally Posted by S485122

I'd vote to have that user keep its "server glitch" score :-)

Seconded. Agile development almost by definition breaks stuff.

That's why it is so very interesting, and welcomed.

[Viliam Furik]

Originally Posted by S485122

I'd vote to have that user keep its "server glitch" score :-)

I agree. But I'd like to at least know the glitch number of GHz-D, which is unfortunately so high that the server can only display it by asterisks. Or perhaps it's an uncomputable number...

[petrw1]

Originally Posted by ATH

Here are some results, but the limits does not look crazy here (see screenshot), but when you look at the exponents it is insane:
108017587 B1= 4 697 591 239 862 648 832 B2= 4 731 979 646 332 043 264
108017531 B1= 4 697 591 239 862 648 832 B2= 4 731 979 646 332 043 264

Those B1 values are fake stage 1 did not get faster, and B2 are fake as well probably, so maybe it is just fake results instead of using the new 30.8 version.


Edit: Maybe it is just a bug on the server? Because in the history of those 2 exponents the limits are fine.

The crazy B1/B2 have been corrected.

[Batalov]

Originally Posted by petrw1

The crazy B1/B2 have been corrected.

...and because the score is derived from B1 - the score vaporized.