processor problems

[Keroberts1]

My cient has been getting a sumout error very repetitivly. Say every few blocks or so. Does this mean that the test is likely to yeild an icorrect result. The test is close yto being finished. Shouldf i finish it.

[Keroberts1]

please someone comment.. I don't want to submit a residue if it is likely an incorrect one.
here are some of my logs.
[Wed Feb 04 16:24:48 2004] n.high = 4528445 . 70 blocks left in test
[Wed Feb 04 16:32:19 2004] internal computation error [mismatched sums]! check your memory/processor. test will restart in 5 minutes.
[Wed Feb 04 16:37:19 2004] AMD Athlon(tm) XP 2000+ detected. Enabling cpu specific optimizations.
[Wed Feb 04 16:37:19 2004] restarting proth test from cache (k=27653, n=5174313) [87.5%]
[Wed Feb 04 16:38:00 2004] resolving hostname
[Wed Feb 04 16:38:00 2004] opening connection
[Wed Feb 04 16:38:03 2004] receiving from server
[Wed Feb 04 16:38:04 2004] logging into server
[Wed Feb 04 16:38:04 2004] login successful
[Wed Feb 04 16:38:04 2004] n.high = 4528445 . 70 blocks left in test
[Wed Feb 04 16:46:41 2004] internal computation error [mismatched sums]! check your memory/processor. test will restart in 5 minutes.
[Thu Feb 05 01:57:34 2004] got k and n from cache
[Thu Feb 05 01:57:34 2004] AMD Athlon(tm) XP 2000+ detected. Enabling cpu specific optimizations.
[Thu Feb 05 01:57:39 2004] restarting proth test from cache (k=27653, n=5174313) [87.5%]

however if someone else completes the test first i will gladly submit the results aftrewards and we can check if the residues matched. Or if anyone tells me that the test is still reliable then i will sumbit it now.

[endless mike]

I was getting the same errors with one of my XP 2000 boxes, it was a result of the heatsink being clogged with dust. I cleaned out the heatsink, and the errors stopped. I finished the in progress test. If it turns out to be bad, it should be found out when we eventually get around to doublechecking it. The double check will have a non-matching residue, and the test should get flagged for a triple check, if SoB implements doublechecking similar to the way that GIMPS does it.

[hc_grove]

That is the way non-matching residues are handled.

The restarts should be from presumed good states, so these messages shouldn't mean anything, but as they are a sign of bad hardware (such as a clogged up heatsink) you can't know for sure.

I would let the test complete and submit the residue, but I'm not the authority on this.

[Keroberts1]

the exact problem I've been having. I was just wondering if having had lots of mistakes found by the program is it likely that some were not caught in time adn the incorrect state was saved into the cache therefore invalidating the whole test. I'm just going ot put this one on hold until i hear from one of the coders on this.

Although if it was wrong it would be tested again eventually if it is prime then it will be a long time before it is retested adn that would result in many un necessary tests being done.

[Keroberts1]

ok well that ones finished and submitted.

[Xaverius]

I had this problem after I OC'ed my system.

On a Athlon XP my rates were 200 * 11 but I don't have a poblem at 190 * 11 .
This is the speed the system runs at now.

[MikeH]

ok well that ones finished and submitted.

If you run through a few days worth of 'supersecret' or 'secret' tests, then ask Louie nicely and he should be able to give you feed back as to whether it makes sense using this PC for new PRP tests.

[Keroberts1]

I'll just keep it in sieving for now

[kugano]

I'd say the first thing you should do is figure out what's wrong with the machine. Like endless mike said, it's usually a sign of something really bad going on that needs to be fixed or at least diagnosed. What kind of hardware is it?

As far as finishing the test, well, that's up to you. I'd say the chances of the results being incorrect are very high if you're getting those errors, but they'll be caught by the second pass (yep, mike, we're handling it the same way Gimps does, it's just not entirely automated yet). The chances of such an error not being discovered on the second pass are ridiculously low. So if you want to keep the box running for stats, well, okay, but I really think you should figure out what's wrong with the machine. It's a big warning sign or, if you prefer, "a desperate cry for help."