Trippple Checks - Page 17

Madpoo · 2015-04-29, 22:15

Quote:

Originally Posted by petrw1

By morning.

Awesome... it's getting whittled away for sure.

Here's an updated list.

I reserved the 29M block and also extended the large exponent block I'm doing down to 50M and up.

I have a few machines tied up at the moment... It may seem like overkill but I'm rechecking every exponent > 4194304 done by a v17 client that has only been double-checked... call me nuts.

There are 1394 of those. But it should be done sometime over the weekend I think. Those systems chew through small exponents at a pretty good clip.

LaurV · 2015-04-30, 03:28

I am taking the first gap (from 24M to 26M [edit except Mark's 3 expos])
It will start on about 15 hours.

cuBerBruce · 2015-04-30, 03:43

My last one is now complete. All 10 of those I did resulted in matching the previously confirmed residue.

As I said before, I will not be taking on any more right now.

Madpoo · 2015-04-30, 04:04

Quote:

Originally Posted by cuBerBruce

My last one is now complete. All 10 of those I did resulted in matching the previously confirmed residue.

As I said before, I will not be taking on any more right now.

Nice... thanks for helping.

I see petrw finished his last one too, and thanks to LaurV for signing up for some triple checks, I've added you in there.

I'll attach an updated list tomorrow.

NBtarheel_33 · 2015-04-30, 13:45

Check out 28488127. This was on a borged machine of mine back in 2008; somehow (perhaps via an image or restarts) my system ran the LL test and *seventeen* double (17-tuple?) checks! The weird thing is that the P-1 for this exponent also was run a total of *twenty-one* times (same B1 and B2!) with as many as three runs of P-1 being completed in between the repeated LL tests.

In other words, this exponent is a "hot mess" and certainly would benefit from an independent double-check just to let it rest in peace.

NBtarheel_33 · 2015-04-30, 13:51

Quote:

Originally Posted by NBtarheel_33

Check out 28488127. This was on a borged machine of mine back in 2008; somehow (perhaps via an image or restarts) my system ran the LL test and *seventeen* double (17-tuple?) checks! The weird thing is that the P-1 for this exponent also was run a total of *twenty-one* times (same B1 and B2!) with as many as three runs of P-1 being completed in between the repeated LL tests.

In other words, this exponent is a "hot mess" and certainly would benefit from an independent double-check just to let it rest in peace.

Here is another exponent from that same system "Steve": 28488913. It had nine LL runs, one first-time and eight octuple-checks, but no independent double-check. Thankfully someone else must have done the P-1, so that didn't get the OCD treatment.

That "Steve" system must have been really messed up. At least all of the residues did match, with zero error code, which might be promising.

Madpoo · 2015-04-30, 14:04

Quote:

Originally Posted by NBtarheel_33

Check out 28488127. This was on a borged machine of mine back in 2008; somehow (perhaps via an image or restarts) my system ran the LL test and *seventeen* double (17-tuple?) checks! The weird thing is that the P-1 for this exponent also was run a total of *twenty-one* times (same B1 and B2!) with as many as three runs of P-1 being completed in between the repeated LL tests.

In other words, this exponent is a "hot mess" and certainly would benefit from an independent double-check just to let it rest in peace.

Funny... wonder how that happens? Maybe you had the worktodo.txt set to read-only, so every time it finished a run it'd start over on the same work each time Prime95 restarted. LOL

Yeah, once this current set of double-checks is all done I planned to expand the search and look for self-triple+ checks. I have to modify my query to find those.

This is one where there are results that bled over from v4 to v5 of the server... it's so cool now to have the old logs added in...we can see where those old entries came from now...I may have to plumb it's depths and see if there are more v4/v5 self verified results for my related triple-checks).

These 2 will be good examples to make sure whatever query I whip up actually works. I have a hunch there won't be that many. The only self-triple checks are likely to be accidents like yours.

chalsall · 2015-04-30, 14:52

Quote:

Originally Posted by NBtarheel_33

In other words, this exponent is a "hot mess" and certainly would benefit from an independent double-check just to let it rest in peace.

I'll take on both of those.

LaurV · 2015-05-01, 02:37

Did ye get credit for all that?

Madpoo · 2015-05-01, 04:40

Quote:

Originally Posted by Madpoo

I'll attach an updated list tomorrow.

And, attached. Finally finished the last of the 190M exponents. Whew. Still 70 something days for the 383M one. That's the system where I have all 10 cores on one chip cranking at it, and to keep the other chip happy I have to feed it smaller exponents. I haven't figured out where the cutoff point is. I can't do a 50M+ exponent on there without seriously slowing down the other one, but so far it's handled up to the 29M exponents okay.

If I run out of triple-checks I can feed it before it's done, I'll probably toss it a batch of triple checks of other randomly curious things in some smaller ranges.

Also, I unfortunately still see some people checking in new self-verified things. Just got a new 74M one. Sigh. Maybe I can talk George into disallowing auto or manual assignments where the same user already did the first check. It'd accept them if they just did it themselves, with no assignment, but it would hopefully make people think that, yeah, maybe letting someone else verify it is a good idea.

I just explained it to my teacher wife with the analogy "we shouldn't be grading our own papers"

LaurV · 2015-05-01, 07:26

What's the percent of the work DC-ed by the same user for which you did TC and found the verified residue to be wrong?
If it is close to zero, then, (channeling my inner davieddy) ye shut the fun up...

More important would be to include CRC's in cudaLucas reports to make it difficult to be faked by a simple text editor. Many of us do LL and DC in parallel in different GPUs and with different shift bit. By "in parallel" I mean in the same time. This helps catching a mismatch in its infancy, and save the work of finishing one LL just to find out at DC that one residue at the half of the way was wrong in the first test. We would like to get credit for both the LL and DC, and that is why all the struggle was to implement the shifting in cudaLucas. If you don't accept both results, then you lost us as a producer. And it would solve nothing, if someone wants to cheat, he can create anytime two different accounts on PrimeNet, use a anon proxy, etc.

2015-04-30, 03:28	#178
LaurV Romulan Interpreter Jun 2011 Thailand 5×1,931 Posts	I am taking the first gap (from 24M to 26M [edit except Mark's 3 expos]) It will start on about 15 hours. Last fiddled with by LaurV on 2015-04-30 at 03:30

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Double checks	casmith789	PrimeNet	7	2015-05-26 00:53
Help doing some quadrup1e+ checks	Madpoo	Data	28	2015-04-06 17:01
Double checks	Rastus	Data	1	2003-12-19 18:20
How do I get rid of the Triple Checks??	outlnder	Lounge	4	2003-04-07 18:06
Double-checks come in pairs?	BigRed	Software	1	2002-10-20 05:29

2015-04-30, 03:43	#179
cuBerBruce Aug 2012 Mass., USA 13E₁₆ Posts	My last one is now complete. All 10 of those I did resulted in matching the previously confirmed residue. As I said before, I will not be taking on any more right now.

2015-04-30, 13:45	#181
NBtarheel_33 "Nathan" Jul 2008 Maryland, USA 5×223 Posts	Check out 28488127. This was on a borged machine of mine back in 2008; somehow (perhaps via an image or restarts) my system ran the LL test and seventeen double (17-tuple?) checks! The weird thing is that the P-1 for this exponent also was run a total of twenty-one times (same B1 and B2!) with as many as three runs of P-1 being completed in between the repeated LL tests. In other words, this exponent is a "hot mess" and certainly would benefit from an independent double-check just to let it rest in peace.

2015-05-01, 02:37	#185
LaurV Romulan Interpreter Jun 2011 Thailand 10010110110111₂ Posts	Did ye get credit for all that?

2015-05-01, 07:26	#187
LaurV Romulan Interpreter Jun 2011 Thailand 10010110110111₂ Posts	What's the percent of the work DC-ed by the same user for which you did TC and found the verified residue to be wrong? If it is close to zero, then, (channeling my inner davieddy) ye shut the fun up... More important would be to include CRC's in cudaLucas reports to make it difficult to be faked by a simple text editor. Many of us do LL and DC in parallel in different GPUs and with different shift bit. By "in parallel" I mean in the same time. This helps catching a mismatch in its infancy, and save the work of finishing one LL just to find out at DC that one residue at the half of the way was wrong in the first test. We would like to get credit for both the LL and DC, and that is why all the struggle was to implement the shifting in cudaLucas. If you don't accept both results, then you lost us as a producer. And it would solve nothing, if someone wants to cheat, he can create anytime two different accounts on PrimeNet, use a anon proxy, etc.