Strategic Double Clicking

[Mark Rose]

Originally Posted by Madpoo

Anyone up for doing another triple-check?

There's an undisplayed false positive in there and we need to verify one of the previous tests (mine, I hope) to confirm the false positive.

M43548073

It may get picked up by someone doing cat 4 so hopefully one of y'all gets it first.

You win!

[Madpoo]

Originally Posted by Mark Rose

You win!

Of course.

The user keeps submitting more false positives from the same Xeon systems... we're a little stumped. I'm not sure what kind of motherboard is being used... he's supplied some diagnostics and I don't remember if it says anything about the manufacturer, chipset (is there more than one for the Xeon E5 v3's?) or cpu revision #...

But so far my own E5 v3 keeps spitting out matching DCs so I'm guessing it's something specific to that system... perhaps something in whatever hypervisor the Linux machines are running under and interesting things happening during some context switching? I don't really know.

It certainly is concerning enough that software that runs great otherwise would be tossing out these false positives on this particular setup. Hmm... hopefully George has a flash of insight that points to where the problem could lie.

[LaurV]

Can't you do something like "this result is not needed due to the fact that this [system/user]sent out too many [false positives/bad results/suspicious/whatever]?

And give no credit.

Then he will pay attention.

[Mark Rose]

I have a few low exponents that need a triple, if anyone is interested:

DoubleCheck=40171801,72,1
DoubleCheck=48405233,72,1
DoubleCheck=48603307,72,1

[bgbeuning]

Originally Posted by Mark Rose

I have a few low exponents that need a triple, if anyone is interested:

DoubleCheck=40171801,72,1
DoubleCheck=48405233,72,1
DoubleCheck=48603307,72,1

Locked and loaded, err added to worktodo and reserved with primenet.

[Mark Rose]

Originally Posted by bgbeuning

Locked and loaded, err added to worktodo and reserved with primenet.

Thanks!

[airsquirrels]

I just switched a big group of GPUs over from TF work to LLDC, so I have a handful of mismatches as I weed out the underperformers. Anyone for some triple checks

DoubleCheck=58980557,72,1
DoubleCheck=58980457,72,1
DoubleCheck=58978061,72,1

Looks like Madpoo is already on these two
59083813
57989489

[GP2]

Originally Posted by airsquirrels

Anyone for some triple checks

DoubleCheck=58978061,72,1

Those are quadruple checks actually. I'll give them a shot.

Edit: should be 73, not 72

[airsquirrels]

Originally Posted by GP2

Those are quadruple checks actually. I'll give them a shot.

Edit: should be 73, not 72

Thanks! I am pretty sure my answer will be the incorrect one. I actually went through the logs and found a quirk in FFT size selection on clLucas/CUDALucas for a range of exponents I was testing right around 58-60M, the code tries to use 3136K, finds an error too large, backs up and continues with 3200K and then goes back to 3136K when the error is low again. Many of those tests came back with mismatches after hundreds of good tests.

[Mark Rose]

I would like to thank

AirSquirrels
ATH
endless_mike
MadPoo
riccardo uberti
rudimeier

for triple checks of my double checks in the last couple of months. Especially ATH, who seems to follow up with a TC in days.

[airsquirrels]

Originally Posted by Mark Rose

I would like to thank

AirSquirrels
ATH
endless_mike
MadPoo
riccardo uberti
rudimeier

for triple checks of my double checks in the last couple of months. Especially ATH, who seems to follow up with a TC in days.

All of these triple checks (and the occasional quads) are mostly thankless, but ideally accelerating us towards more strategic double checks by giving us definitive records for more systems. Madpoo would have to give us updated numbers, but there should be under 2000 unassigned triple checks at this point.