Strategic Double Clicking

[UBR47K]

Triple checks needed (Round-Off CuLu errors happened at 2048K FFT)

DoubleCheck=38750629,71,1
DoubleCheck=38673287,71,1

[cuBerBruce]

Originally Posted by Madpoo

Here's another batch of 2:1 + bad to good for anyone wanting some:

exponent	Bad	Good	Unk	Sus	Solo	Mis	worktodo
49812869	2	0	2	0	2	0	DoubleCheck=49812869,72,1
50145749	2	0	1	0	1	0	DoubleCheck=50145749,73,1

I've taken these two.

[fivemack]

Taken

51770489	15	4	2	1	2	1	DoubleCheck=51770489,73,1
51940249	15	4	2	1	2	1	DoubleCheck=51940249,73,1
52307537	5	2	6	1	6	1	DoubleCheck=52307537,73,1
53998279	5	2	6	1	6	1	DoubleCheck=53998279,73,1

should be done by start of March

[ATH]

Originally Posted by ET_

38194643 unverified.

Queued this one.

[kladner]

For real, this time, mismatch on 37334837.

[dragonbud20]

Originally Posted by kladner

For real, this time, mismatch on 37334837.

I've got it should be done in a week

[kladner]

Originally Posted by dragonbud20

I've got it should be done in a week

Thanks!

[ATH]

Originally Posted by Madpoo

Turns out he meant this one:
M38904911

It did not match unfortunately and will need a quad check: http://www.mersenne.org/M38904911

It ran on my Titan Black with 2160K FFT. Max error was only 0.10742.

[Mark Rose]

Originally Posted by ATH

It did not match unfortunately and will need a quad check: http://www.mersenne.org/M38904911

It ran on my Titan Black with 2160K FFT. Max error was only 0.10742.

I queued it.

[kladner]

Wow! Thanks to both of you.

[Madpoo]

In my ongoing quest to weed out bad systems (well, bad tests), I'm toying with another revision to my query.

Currently I filter the results based on: user, cpu, year, app-version

I've noticed here and there that within a calendar year, a machine might go through cycles just like it does year-over-year, so I added in filtering by quarter to see how that would shake out.

Here's an example of one machine for particular year (2009) and how it shakes out by quarter:

CpuId	Good	Bad	Sus	Unk	Solo	Mis
2009-1	0	1	0	0	0	0
2009-2	5	0	1	0	0	1
2009-3	3	1	0	3	3	0
2009-4	0	6	0	1	1	0

You can see the ups/downs it had over 2009. 2nd quarter was okay but by the time we get to the 4th quarter, it's not doing so well. It didn't have any 2010 results so it probably gave up the ghost, related to those bad results.

Filtering by quarter means I may not bother filtering by the app version. That was just a way of doing kind of the same thing... some date during the year when they upgraded version as a way to break down the timeline into smaller chunks. There really isn't much difference in error rates from one version to the next.

If I go much beyond quarterly breakdowns, there might not be much data to work with. That's only a 3 month period and many older systems just didn't do a lot in 3 months... it could be harder to spot longer term trends.

But for now I think this will help bring in a few more "easy pickings".