Possible hardware error
 

Well, this is disappointing. After a few months of running smoothly, my machine has turned up with this error on Worker6/Core5 of the Phenom II 1090T:

ERROR: SUM(INPUTS) != SUM(OUTPUTS), 6.242418901921988e+080 != 0

I have done some searching and reading, and it seems that this is a single incident. However, I have stepped back from the mild overclock which I had been running:[INDENT]206 x 17, DDR3 RAM @ 1600 instead of 1333
[/INDENT]I thought that I had verified stability via Torture Tests and Memtest86, but such is not the case, obviously. Now the RAM is back running at its SPD rating, and the CPU is doing 200 x 17. Full details will follow when I can get them together.

Here is my question. This exponent has only reached ~4.5%. I have read references to restarting an exponent, but I don't know how to accomplish this. Rather than leave the results of this run in question, I would start this one over. I know that everything is double checked, but I'd rather have a clean run.

What is the means of getting that assignment to restart?

Stop and exit Prime95.
Go to the Prime95 folder and delete the temporary files associated with the exponent: you will typically have 2 files per exponent, something like pW968997 and pW968997.bu (the actual file name depends on the exponent you´re testing, as it replicates the last 6 digits of the exponent). If you have chosen to have 2 backup files per exponent, you will also find pW968997.bu2. Delete all (2 or 3) temporary files related to the exponent.
Restart Prime95. The test will restart from scratch.
If you are only at 4.5%, I think it´s a good idea to restart the test.

As for the actions to take after such an error, I would recommend to up the CPU and RAM voltage slightly and try to keep the overclock (unless you are already running at higher than standard voltages, in which case it´s better to go back to stock speeds, as you have done).

Thanks!
 

Thanks, lycorn.

The deed is done.

I must confess that I have not given up the OC entirely. I put the RAM back to stock, as my reading tended to implicate memory for this type of error. I also put the base clock back to stock. At this point I'm only running a multiplier boost from 16x (stock) to 17x.