Prime95 tests and rounding errors, memory or CPU-related?

[Solema]

I was reading in another post somewhere that rounding errors in Prime95 are usually memory-related. Is this true? Also, which of the three tests, small FFT's, large FFT's, or mixed should I run to purely test CPU stability. Which test is best to test in-Windows memory stability? Are there any other Windows-based programs good for just testing RAM stability? I have heard that Memtest stable doesn't always mean Windows stable.

 

[tomt4535]

You want to run small FFT's to test your cpu stablity. I think the blend test is good for testing memory stablity. Super Pi also does test both cpu and memory stablity.

 

[Solema]

Any idea on what the rounding errors signify?

 

[arcas]

To test your CPU stability: set your CPU to your target speed and lower your RAM multipler way down and run your stress tests. Any errors you get are going to be CPU-related aside from an outright failed DIMM (and memtest86 should catch most of those).

To test your RAM stability: set your CPU multipler way down, set your RAM to your target speed and run your stress tests. Any errors you get will be RAM-related.

 

[crispy2010]

What voltage is your ram running at?

 

[Markbnj]

Any idea on what the rounding errors signify?

The author of the program might be able to shed some light, but as far as I can see it is tough to say. At the highest level I assume it means the program expected to see one result from an operation, and instead saw another. This could indicate FPU errors due to CPU stress. But the program can only get to that data by either reading memory or reading a register. Memory bus errors could account for either of those events, I suppose.

I'm inclined to believe it is FPU though.

 

[Solema]

Thanks for the advice guys. My RAM is running at 2.8v. It's Revision 2 of the Corsair XL TCCD chips on Brainpower PCB's.