nan detected!

[Misfit138]

One of my bartons is having a hard time with the 600 pointers. it gets part of the way through until a fatal error occurs, 'nan detected' is what it says. i have loosened ram timings and throttled back my oc a few times but it still wont pass those 600 pointers.
Can anyone tell me if 'nan detected' necessarily indicates either RAM or CPU as the culprit? Or could it be either?
CPU is mobile barton 2400, 1.825 vcore. error occurs at 11.5x 2xx , 11x, 10.5x and 10x. I am currently at 9.5x200, and i have turned advmethods on to get some different WU's, as it seems to be only the 600 pointers which cause it to fail.

RAM is 2x512mb OCZ platinum rev. 2, which tests good all the way up to ddr600 on my nForce4 939. i get the error even at 3-3-3-11 ddr400 on the barton nforce2.
I suspect it may be a defect in the CPU, or a poor overclocking chip. can anyone confirm ?

[tenchi86]

I am not positive, but from google it looks like a floating point type error. So I would guess it's something to do with the CPU. Sorry I know that dosn't help much but maybe with some more googling you can find some more info.

[ChasR]

It could be either ram or cpu. Memory is easiest to rule out. Try running 1 x 512 and swap them and see if the problem persists. Prime 95 ought to catch a cpu problem. Stress CPU ought to catch either. memtest86 usually catches the ram problems. Most of the similar problems on nForce 2 boards turn out to be memory related, frequently ever so slightly mismatched chips on dual channel. You may have to dl another 600 pointer to experiment with. Copy the work folder and queue.dat so you can reload the WU.

[muddocktor]

Yeah, make a copy of the next 600 pointer you get and then unplug the lan cable from the machine. That way, if it does error out, you haven't wasted a good wu. Those 600 point wu's are some of the most severe tests I've seen for testing an overclock's stability, even better than Prime95 IMO.

[Misfit138]

Thank you for the replies; is the work unit contained within C:\Program Files\FAH\work ?
which file is it exactly? or is it the whole folder?

[Captain Newbie]

NaN = Not a Number = physical impossibility committed. A sanity check of sorts.

It's either
1) a bug in the core, or
2) a hardware problem.

[Careface]

most likely a CPU problem, wouldnt it be? like prime not rounding correctly, or trying to divide something by 0? just trying to help :P i know that when i was writing javascript ddr dance point calculators (ages ago) to make the script not completely **** itself when someone ran impossible equations, i had to make it display NaN, instead of crashing

wait on, why the heck am i telling you this? lol.. ill leave :p

Careface*

[ChasR]

Download a 600 pointer, stop the client,make a copy of the ...FAH\work directory and ...\FAH\queue.dat. Unplug from the internet, start the service and test away. If you have a dual install (you shouldn't) on the the Barton it's going to be in the FAH1 and FAH2 directory, you can pick which one to test.

[Misfit138]

will do! as soon as i get another 600 pointer.
thanks again

[Captain Newbie]

most likely a CPU problem, wouldnt it be? like prime not rounding correctly, or trying to divide something by 0? just trying to help :P i know that when i was writing javascript ddr dance point calculators (ages ago) to make the script not completely **** itself when someone ran impossible equations, i had to make it display NaN, instead of crashing

wait on, why the heck am i telling you this? lol.. ill leave :p

Careface*
NaN's occur in two types on the x86--signalling and quiet. This was a signalling NaN (almost certainly), since it caused a problem.

Divison by zero is not necessarily a NaN (although trying to (f)(i)div(s)(p) anything by zero on the x86 will almost certainly trigger a divide fault and invoke an exception rather than trying a mathematical impossibility)--in fact, things that are divided by zero can be characterized as approaching infinity, or approaching a certain finite value. (For instance, the limit of (sin(x) / x) as x->0 is finite, whereas the limit of (1/x^2) as x->0 does not exist but may be characterised as an infinite discontinuity.) NaNs are really more of a result of where a certain function doesn't exist (for instance, log(-1), sqrt(-1)) on the real number line.

If you ever get those in a physical calculation, something weird happened.