Looking for help with first CPU overclock (FX-4170).

[trents]

I looked on NewEgg for stock settings for the 4170. The stock multiplier would be 21x since the stock frequency is 4200 (21x200 fsb).

 

[trents]

Oh, I forgot to tell you! Go into bios and disable Turbo Boost or whatever it's called there. I think that's what's causing things to jump around. Sorry.

 

[Obsidus]

Oh, I forgot to tell you! Go into bios and disable Turbo Boost or whatever it's called there. I think that's what's causing things to jump around. Sorry.

Yeah, turbo boost was already off. So should I just ignore the core temps wigging out and continue on until it becomes unstable?

 

[trents]

Turbo Boost (a misnomer, really) accelerates some of the cores but not all of them at the same time. It rotates among them. This technology saves power and cuts down on heat but sabotages overclocking when you get past a certain point.

 

[Obsidus]

Interesting.......It is saying the multiplier is 23.0, but it's saying my core speed is 4984.90??

 

[trents]

If Turb is off then it sounds like some other downthrottling technology is still in effect. Let me look at your owner's manual for some bios setting we may have overlooked. These FX CPUs have sophisticated downthrottling technology and this problem is a common one. It's a safety device and responds to temps. If your core temps are exceeding some threshold this is what may be happening.

Do you have the windows CPU meter gadget installed on your desktop? If not please add it so we can see what's happening with the cores when this jumpiness is occurring.

 

[trents]

Look at your core voltage in CPU-z! It's way up to 1.536. Too high! Take that off of auto in bios and manually set it say 1.425. Seems like "Auto" is allowing it to skyrocket.

 

[trents]

Also, look at the CPU-z bus speed. It's at 216 when stock is 200. That's why the CPU core speed is higher than you expect. CPU speed is the product of the bus speed and the multiplier. 216.7x23=4984. Set the bus speed manually in bios (may be called CPU frequency or HT Reference or something similar) to 200.

Don't get discouraged. We're getting a handle on this.

 

[Obsidus]

Trying to take some bios pics for reference...

 

[Obsidus]

Ok....

 

[Obsidus]

Ok I fixed the clock issue by changing the bus to 200 from 216, but the damn voltage is still all saying auto even when it's set to manual, so I cant lower it from 1.524.

 

[Obsidus]

Here.

 

[Obsidus]

*sigh* When I restarted the computer it froze at the desktop, so I went back into the bios and it didn't save me changing the CPU bus to 200; it was still at 216.

 

[trents]

You changed the main DRAM control header to Manual but you left the sub heading voltage control on Auto. You need to change it to Manual as well. Probably "+" or "-" key again.

 

[Obsidus]

You changed the main DRAM control header to Manual but you left the sub heading voltage control on Auto. You need to change it to Manual as well. Probably "+" or "-" key again.

Ok fixed it. I got confused and thought that adjusting the value was how much you wanted to add to the voltage; but it was what you want it to be. So I switched it to 1.425000 and it saved, and the bus is staying at 200 as well.

 

[trents]

Okay, good. Lessons learned in how to manipulate bios values. Everybody goes through this at one point or another. Don't feel bad.

 

[Obsidus]

New CPU-Z and HWmonitor screenie post changes.

 

[trents]

Yep, that's better. Old uncle "Auto" was messing with us.

 

[Obsidus]

Ok I tried running another blend test, and for about 5 seconds things seemed fine, the temperatures remained stable and low, but then about 7 seconds in worker number 4 had a fatal error and ended the stress test. All it told me was:


[Sat Aug 18 20:18:07 2012]
FATAL ERROR: Resulting sum was 4.121876375092226e+016, expected: 4.459881063758814e+016
Hardware failure detected, consult stress.txt file.


Second attempt at blend test, lasted a little longer but now worker 2 failed:


FATAL ERROR: Rounding was 0.5, expected less than 0.4
Hardware failure detected, consult stress.txt file.

 

[trents]

Yep, just add a tad more vcore. That event just tells you the core was not getting enough juice.