Join PrimeGrid
Returning Participants
Community
Leader Boards
Results
Other
drummerslowrise

1)
Message boards :
Number crunching :
2019 Tour de Primes
(Message 125763)
Posted 137 days ago by Andy
Going to stick with SoB and forgo any practical chance of the P_2019 badge.
I suppose it's Red or Bust this year. And, more likely than not, bust.

2)
Message boards :
Generalized Fermat Prime Search :
DO YOU FEEL LUCKY?
(Message 125673)
Posted 137 days ago by Andy
Absolutely not. If I were lucky, SOB would have ended before this forsaken thing showed up over there <

3)
Message boards :
Seventeen or Bust :
The SoB Double Check has begun
(Message 125383)
Posted 143 days ago by Andy
Heck if 10% of the tasks left on machines from the challenge return results instead of timing out I'd take it.

4)
Message boards :
Seventeen or Bust :
The SoB Double Check has begun
(Message 125381)
Posted 143 days ago by Andy
No sharp rise in the Waiting% post challenge is encouraging.

5)
Message boards :
General discussion :
Current prices of RTX 2080s vs GTX 1080s
(Message 124632)
Posted 158 days ago by Andy
I believe it is expected that there will soon be GTX 11xx cards equivalent less the ray tracing functionality, which isn't yet supported by a damned thing (and halves your performance to boot when turned on) for a lower price than RTX cards.
This makes the RTX cards sell poorly... and imo provides a plausible cause of the price cuts.

6)
Message boards :
Seventeen or Bust :
The SoB Double Check has begun
(Message 121407)
Posted 234 days ago by Andy
Have we gone below 50% work waiting yet?

7)
Message boards :
Seventeen or Bust :
The SoB Double Check has begun
(Message 117638)
Posted 412 days ago by Andy
n=25m, 181 done, 0 bad
~~If we take an independent model, the accuracy such that 181/0 has a 50/50 chance to occur is about 99.62% (0.38% incorrect results). In reality, there is likely some dependence (a bad PC will likely return more than 1 bad result...) but that's impossible to quantify.
If there truly was a doublecheck in this range that matched residues already, it's a shame it wasn't documented  if this carries on through a significant portion of the remaining residueonly checks, this is a lot of work that we're 100% required to waste, as much as 15% of the DC. ~~
Just noticed, as I was writing, that I have 2 pending inconclusives from (send date) April 22 and April 26, which must be from the n=25m range (since there were 0 waiting n=24m tasks on April 21). Looks like none of the bad results have managed to find 2 fast multithreaded machines, not that all the residues are correct.
In fact... one is on a borderline malicious machine (http://www.primegrid.com/results.php?hostid=899451) grabbing hundreds of tasks, and letting them time out... I doubt the owner realizes what it's doing since their other machines are behaving fine.

8)
Message boards :
Number crunching :
Badges III
(Message 117458)
Posted 419 days ago by Andy

9)
Message boards :
Seventeen or Bust :
The SoB Double Check has begun
(Message 117403)
Posted 422 days ago by Andy
If they did DC, then is there a max N with a bad residue after which all are seemingly correct? Since the odds of a DC confirming an incorrect result are astronomically low...

10)
Message boards :
Seventeen or Bust :
The SoB Double Check has begun
(Message 117215)
Posted 426 days ago by Andy
So then by a similarly flimsy calculation, there was about a 9.7% chance that the completed portion of the doublecheck would have had a missed prime (1(28415/28416)^(2887)), so the total chance of the DC finding at least 1 prime is roughly 1 in 4.
The question then becomes, is this 1 in 4 chance higher or lower than doing leading edge work? Only 1 in 30 actually have a snowball's chance, but the other 29/30 typically need only be run once. Plus the fact that these tasks are all faster, by some amount, than leadingedge tasks, and are more likely to be prime because of prime density.
If we again ignore the growing improbability of finding a prime as n increases, it would have only been needed that more than ~222 LE SoB workunits get returned each month for the leading edge to have been equally efficient, which is almost certainly true, 15 tasks returning per day (since I can see that at least 33 tasks have returned so far today, based on freedc)
If instead we take the prime number theorem to be perfect, then those numbers in the range we've been working in for the last year are about twice as likely to be prime, so it would take twice as many LE workunits to match the efficiency. This would mean we'd need 30 tasks returning per day which may or may not have been happening. I'd have to guess it'd be close to this number on average based on number of tasks in progress & deadlines for the higher Ns.
Long and short: On the back of a napkin, the DC is probably similar in timeefficiency for prime finding to ignoring the single residue tests and attacking the leading edge.

Next 10 posts
