Is it worth persisting with this machine?

[stebbo]

Hi All,

I have one machine that keeps giving bad results. This machine is in out-of-the-box configuration - no fiddling with anything. I've run any number of tests on it to try to get it to fail over long periods of time, but nothing shows up.

A screenshot of it's recent history is attached. If the occasional bad result is not a drain on resources then I'm happy to keep doing double-checks with it but am just as happy to retire it as well.

Thoughts?

Cheers,
Chris.

[Mark Rose]

Reseat the RAM. Run memtest86. Try swapping RAM and power supplies.

A machine that doesn't perform LL/DC 100% correctly doesn't hurt the project at all, but it does consume more electricity per correct result.

[axn]

Originally Posted by Mark Rose

A machine that doesn't perform LL/DC 100% correctly doesn't hurt the project at all, but it does consume more electricity per correct result.

Worst case, it could delay a prime discovery by a decade (until doublecheck finds it). I would not use this machine for first time LL. DC & ECM should be ok.

[xilman]

Originally Posted by stebbo

A screenshot of it's recent history is attached. If the occasional bad result is not a drain on resources then I'm happy to keep doing double-checks with it but am just as happy to retire it as well.

Thoughts?

Do you have any guess as to how often an error occurs? Hourly, daily, weekly, monthly? If so it would be wise to use your machine to run tasks which take significantly less elapsed time than that.

Even machines with hourly faults, as long as the faults don't crash the machine, can be used with an ECMNET client. Details and assistance on request, but 109.109.168.71:8194 may be all you need if you want to go that route.

[chalsall]

Originally Posted by stebbo

Thoughts?

There's no real downside to the GIMPS project having this machine continue. It probably wouldn't even delay the (*exceedingly* unlikely) missing of a MP by much time because your machine will be given "the evil eye" by Madpoo's "Strategic Double Check" initiative.

However, it is _strongly_ advised you don't use this machine for any mission critical "real" work you might be doing with it.

I've said before and I'll say forever: running GIMPS has saved me more time, money and grief than I can even begin to guess over the years by pointing out machines which are about to die.

[henryzz]

Originally Posted by stebbo

Hi All,

I have one machine that keeps giving bad results. This machine is in out-of-the-box configuration - no fiddling with anything. I've run any number of tests on it to try to get it to fail over long periods of time, but nothing shows up.

A screenshot of it's recent history is attached. If the occasional bad result is not a drain on resources then I'm happy to keep doing double-checks with it but am just as happy to retire it as well.

Thoughts?

Cheers,
Chris.

Is this a skylake machine? If so the BIOS may need updating.

[Madpoo]

Originally Posted by chalsall

There's no real downside to the GIMPS project having this machine continue. It probably wouldn't even delay the (*exceedingly* unlikely) missing of a MP by much time because your machine will be given "the evil eye" by Madpoo's "Strategic Double Check" initiative.

Could be... I was the one who did the triple-check on those two exponents that ended up proving his results wrong.

In those cases at least, it was because they were double-checks, so a mismatch there is definitely going to land the exponent in my crosshairs.

First time checks, well... it all depends on if they've ever done (or had) a doublecheck that matched or not. Someone doing first time checks exclusively isn't likely to get a doublecheck done anytime soon, so bad systems can go undetected for years.

We have looked before at trying to do at least one DC of every computer that's ever turned in a result... there are a lot though. Special attention has been given to those with a lot of results turned in since a mismatch on any of those would be more interesting.

That's also one of the reasons for the change in the assignment code lately to periodically assign a DC as a "sanity check" once a year or whatever, because it really does help the project if we can find bad systems earlier in the process.

EDIT: Oh... for that particular system, here are its stats broken down by year / month. Overall it has more good than bad. That "suspect" result in March 2017 could go either way (78077873):

YYYY-MM	Bad	Good	Sus	Unk
2016-12	1	4	0	0
2017-1	0	0	0	2
2017-2	0	0	0	2
2017-3	1	0	1	1
2017-4	0	2	0	2
2017-5	1	6	0	0