Round off error

[Androx72]

Hello I am new to the forum and greet you all ... I have a problem with my pc .... with Prime95
.. I bought an AMD FX 8350 Gigabyte 990FXA UD7 mounted on a Gigabyte 990 fxa ud7... default no overclocking and standard voltages, memory is memtest + ok temperatures of 50° on core Intel burning test passed without errors x 30.
With Prime95 version 27.9 I always error on cpu # 2

here

Trying for 1000 iterations using 3200K FFT exponent 61224923.
If average roundoff error is above 0.24282, then a larger FFT will be used.
Trying for 1000 iterations using 3200K FFT exponent 61232363.
If average roundoff error is above 0.24282, then a larger FFT will be used.
Trying for 1000 iterations using 3200K FFT exponent 61246249.
If average roundoff error is above 0.24282, then a larger FFT will be used.
Trying for 1000 iterations using 3200K FFT exponent 61246321.
If average roundoff error is above 0.24282, then a larger FFT will be used.
[Fri February 22 21:27:52 2013]
Trying for 1000 iterations using 3200K FFT exponent 61224923.
If average roundoff error is above 0.24282, then a larger FFT will be used.
Trying for 1000 iterations using 3200K FFT exponent 61232363.
If average roundoff error is above 0.24282, then a larger FFT will be used.
Trying for 1000 iterations using 3200K FFT exponent 61246249.
If average roundoff error is above 0.24282, then a larger FFT will be used.
Trying for 1000 iterations using 3200K FFT exponent 61246321.
If average roundoff error is above 0.24282, then a larger FFT will be used.
Trying for 1000 iterations using 3200K FFT exponent 61324849.
If average roundoff error is above 0.24282, then a larger FFT will be used.
Iteration: 212/61361423, ERROR: SUM (INPUTS)! = SUM (OUTPUTS), 1.510314394830373e +081! = 0
Possible hardware failure, consult the readme.txt file.
Continuing from last save file.
[Fri February 22 21:34:12 2013]
Trying for 1000 iterations using 3200K FFT exponent 61224923.
If average roundoff error is above 0.24282, then a larger FFT will be used.
Trying for 1000 iterations using 3200K FFT exponent 61232363.
If average roundoff error is above 0.24282, then a larger FFT will be used.
Trying for 1000 iterations using 3200K FFT exponent 61246249.
If average roundoff error is above 0.24282, then a larger FFT will be used.
Trying for 1000 iterations using 3200K FFT exponent 61246321.
If average roundoff error is above 0.24282, then a larger FFT will be used.
Trying for 1000 iterations using 3200K FFT exponent 61324849.
If average roundoff error is above 0.24282, then a larger FFT will be used.
Final average roundoff error is 0.23782, using 3200K FFT for exponent 61232363.
Final average roundoff error is 0.23696, using 3200K FFT for exponent 61224923.
Final average roundoff error is 0.23961, using 3200K FFT for exponent 61246249.
Final average roundoff error is 0.23916, using 3200K FFT for exponent 61246321.
Final average roundoff error is 0.24826, using 3456K FFT for exponent 61324849.
[Fri February 22 22:57:33 2013]
Iteration: 57916/61224923, ERROR: ROUND OFF (0.5)> 12:40
Continuing from last save file.
Iteration: 57804/61224923, ERROR: ROUND OFF (0.48828125)> 12:40
Continuing from last save file.

It could be a hardware problem?

[Batalov]

You can try to remove unwanted variables (the closeness of the FFT cross-over boundary).

Try this:
1. make a new folder, copy the .exe file there
2. Edit your worktodo.txt file in that directory to include the line
Test=FFT2=3584K,61224923,75,2
3. Let it run for a while. If it will still fail, then it is likely a hardware problem.

[Batalov]

I have now run this exponent locally at both 3200K and 3584K up to iteration 100,000 and don't see any problems. (the CPU is not AMD, though)

If your test run will produce errors again at the larger FFT size, the usual recommendations of:
1. checking all seating problems (the CPU sink, the memory - take out, rearrange, seat them well).
2. cleaning (may not be necessary if you are saying that this is a fresh computer)
3. downclocking (start with memory, or you can lower the base freq if you can; it is hard to believe that the CPU would be at the edge of usability with stock clock) or giving more voltage. Btw, are you sure that you memory does not require a special voltage? Check that and correct if necessary. *For example, one of my older comps has a 2.1V memory (it's DDR2, don't reach for the heart drops just yet), but when BIOS sometimes goes awry and needs a full reset, the standard voltage goes back to 1.8V - and then surely it sort of functions on idle, but starts firing errors under any load.