Stable under Prime, WHEA errors at lower loads

darkfalz

Member
Jul 29, 2007
181
0
76
I got a new HSF for my 2 year old 3570K (Cooler Master D92) because I wasn't happy with my summer temps @ 4.0 under load.

Obviously this runs at least 25 degrees cooler than my previous stock HSF under heavy load so I decided to try pushing my overclocking a bit more (which was previously heat limited).

My settings for 4.0 were offset -0.020v and nothing else changed.

I have been having some trouble getting a stable 4.2 though. I have tried various combinations of -0.01v with medium LLC, +0.010 to +0.030 with no LCC and now +0.020v with medium LLC.

My problem is that it ran for 12 hours stable Prime95, but I get an occasional WHEA error running a lighter CPU task, like Unigine Valley.

Maybe it's not being supplied enough voltage when not under full load?

What should I increase? LLC or Offset or go to a set voltage?

Is there a good tester (OCCU with idle periods?) that tests stability at various loads?
 

Ketchup

Elite Member
Sep 1, 2002
14,559
248
106
What are your memory speeds like?

So, it runs perfectly fine at 4.0?
 

darkfalz

Member
Jul 29, 2007
181
0
76
Memory speed is at stock XMP 1600 same speeds have been running for years. 4.0 stable for some time even with stock HSF, I just didn't like the temps it was hitting during summer.
 

darkfalz

Member
Jul 29, 2007
181
0
76
Well, seems I don't get the WHEA under Valley at +0.05v offset. LLC didn't seem to help, just increased vcore/heat under heavy load.

I'm going back to 4.0 I think. 5% performance increase is not worth the 0.07v and 10 degrees extra, I think my chip is a bit of a dudd even with the new cooler. But Prime95 peaking at 60C is pretty good (was peaking at 90 before!) at 4.0 is good enough. Most games GPU limited anyway.
 

darkfalz

Member
Jul 29, 2007
181
0
76
I think I figured out why my temps were higher than last summer. I did a BIOS update a few months ago. It seems "CPU Phase Control" changed from its default of either Standard or Optimized to "Extreme" - and this seems to generate more heat. I have changed it back to Standard. Probably didn't need the HSF in that case but I am liking the new temps.
 

darkfalz

Member
Jul 29, 2007
181
0
76
Okay, got it 4.2 stable with +0.035v and respectable temps.

Just for fun I've tested it @ 4.5. Requires a painful 1.36v to be stable (limited testing, might need a a tad more). Peaks just over 80 in Prime95 SmallFFT (low to mid 70s for Blend).

Ouch... but I guess at least I know my chip can do 4.5.

4.4 stable at +0.130v.

Unigine Valley seems better at detecting errors than Prime. I guess the wildly varying CPU load gives it and the voltage regulators a bit of an extra workout. I do have a budget board after all.
 
Last edited:

darkfalz

Member
Jul 29, 2007
181
0
76
I found a good FFT for my chip (In Place FFT 128K 39000 and 46000 particularly) - causes WHEA errors reliably if the vcore isn't high enough, despite it running for hours on Blend fine or passing IBT. This has been useful in revising my vcore for different overclocks. Making up for that 3 years ago now when I didn't buy an aftermarket cooler, hehe.

I've also discovered that a cooler CPU can handle lower vcore at the same frequency. Obviously there's a delicate dance going on here between temperature and the laws of threshold voltage and still supplying enough juice to keep the chip stable. It would make a very interesting graph. I'd never previously thought this was the case and that the only factor was thermals ie. minimum stable vcore is minimum stable vcore at any temperature as long as it is under the point of overheating. Appears not... I can undervolt all the way to -0.060v (or more? but I want to stay out of the "red" so knocked it back to -0.050v) and be stable at 4 GHz whereas previously it was -0.020v. I guess it makes sense too when you've got worse cooling and you try raising voltage to hit a higher overclock you hit a wall quickly because you run into the double whammy of requiring a higher voltage because of higher temperatures, and having higher temperatures.

4.0 -0.050v ~61 C peak
4.2 +0.040v ~67 C peak
4.4 +0.140v ~76 C peak
4.5 +0.210v ~85 C peak

Looks like 4.2 is the sweet spot for me (4.0 is also nice and very cool, but since I could get that with the stock HSF it's like I wasted my money on the D92).
 
Last edited: