-   Hardware (
-   -   Hardware errors help (

Chelle 2020-10-20 16:25

Hardware errors help
Anybody have any ideas?
Getting following hardware errors prime95.
As shown in the window:

LL value corrupt could not run jacobi error check
Errno 34 result too large
DOSerror :2

Is likely the cause bad ram it’s ecc though?
Have swapped some ram but don't have enough to change all.

I’ve restarted it for the 3rd time from the beginning happened at 80% 40% and 70% complete aprox. Just gets stuck and repeats the last 1% or 2%.

moebius 2020-10-20 17:04

Which prime95 version do you use?
Could it be due to the RAM-settings in the BIOS (timings, wrong profile,OC,.....)?
Why don't you use PRP + proofs [prime95 v30.3 build 6] (which leads to correct results at many hardware errors, and you should also use this because it is preffered nowadays)?

paulunderwood 2020-10-20 17:26

Try running memtest86+. You might have one or more bad sticks of RAM.

Also install a temperature monitoring software.

kriesel 2020-10-20 17:30

Welcome to the GIMPS Mersenne forum.

Current version of prime95 is v30.3b6. [URL][/URL]
Switch from LL to PRP with Gerbicz Error Check and proof, with power 8 or so enabled by setting Options, resource limits, Temporary disk space limit to 6GB or more of on-disk storage allowed per worker, if you have the space. With the same effort as your past 3 failed attempts at one LL, you could have 2 PRP's completed or close to it, with proof generated, despite recurrent errors, and someone would run brief Cert assignments eliminating the work of conventional double checking or triple checking. So PRP is effectively, from a project-wide view, twice as fast as LL when both are working reliably, and PRP is even better relatively, when not entirely reliable.
In an extreme case, such as on my old unreliable Athlon system that since went to recycling, PRP/GEC may make no progress because of high error rate, while LL with Jacobi may miss errors and make iteration progress incorrectly producing eventual wrong results.

If determined to persist with LL anyway, make use of prime95's backup files.

Much more background/reference info available at [URL][/URL]

Chelle 2020-10-20 19:41

Have just now updated to the latest version, Was trying to see if I could continue the work without starting again. Tried using bu4 file (deleting others) but soon got ‘error writing intermediate file’ ‘result too large’

Yes I could start from beginning and change to PRP, maybe the only way, pity as nearly 70% done.
The main cause is the hardware but it takes like a month to show up each time.

Runtime Error 2020-10-20 21:27

[QUOTE=Chelle;560450]Yes I could start from beginning and change to PRP, maybe the only way, pity as nearly 70% done.[/QUOTE]

Definitely a shame, but PRP will give the right answer even with an occasional error. Also, the new PRP in 30.3 generates a proof file that eliminates the need for a double check! So although restarting will cost 100% of the test, continuing will either cost 130% total (including double check) or 230% total (including triple check) if your residual was wrong.

Uncwilly 2020-10-20 21:49

What is the size of your exponent? Just give us the millions range.

And what is the CPU type that you are using? How much RAM do you have? And are you trying to write your files to a USB drive?

phillipsjk 2020-10-21 02:24

[QUOTE=Uncwilly;560457]... And are you trying to write your files to a USB drive?[/QUOTE]

tangent: my "white elephant" machine is using an 8GB USB drive, but I use BTRFS, which includes a check-sum. It is VERY slow, so am planing to buy a 64GB drive this week to replace it. More space puts off re-writes for a longer period of time (plus it can accommodate the new muti-gigabyte PRP proof files).

Plus I hope the new drive fits vertically in the 1U case (so I can use the drive bays for my backups when not factoring).

Xyzzy 2020-10-21 13:18

Are you using "dup" or "single" for your data?

All times are UTC. The time now is 21:31.

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2021, Jelsoft Enterprises Ltd.