4790K overclocking + Stressing with Prime95?

AvengerUK

AvengerUK

AvengerUK

Associate
Joined
29 May 2006
Posts
18
Hi all,

Firstly - a little note, I am not too familiar with overclocking since Sandybridge - so there is every chance I've made an error somewhere along the line, hence why I'm posting here!

I recently discovered my 4.7 overclock was unstable (as in, rarely) - which made me revisit it and re-do everything from step 1.

Currently - I'm using the following settings 100% Stable:

Multiplier: 46
Voltage (note: I did this on "Manual" before switching to adaptive): 1.240
Eventual CPU Input: 1.700 (This could probably go lower).

AIDA64 / ITU run fine, temps reach around 65-67 deg max.

Cooling: Predator 240

---------------------------------------------------------------------

Now, as far as I can tell, to get 4.7 - I require an input of around 1.9 (or more), and a core voltage of 1.32 or there abouts.

I'm never sure about the CPU input voltage tbh!

The problem with this, and my current overclock is I'm not sure on the safe voltages for a 4970k - and, I used to use Prime95 for stress testing.

Only, this is now very weird (and toasty).

For example, during 4.7 testing - Prime95 would start off a hit about 72-74 - and be fine. Although after a while, it'd suddenly shoot to dangerous highs of 90-100 (quickly abandoned the test as soon as this happened).

This happens on a "Blend" test everytime, no matter what voltages I'm using.

ITU / AIDA64 is always fine though....?

So yeh, a little confused about that, and also where to go with my 4.7 ambitions.

(Note: currently bios settings are: http://imgur.com/a/ABaaR )

Thanks for any advice :)
 
First off, id reccomend dropping prime95 as a stress tool. Particularly ver 28.5 or later. Due to the use of a non soldered ihs assembly on the cpu, and the use of AVX instructions in theese versions of prime. Temps can get very high on haswell/devils canyon. Asus real bench is a good test.

Max reccomended voltage for 24/7 usage on devils canyon is 1.35. Regarding input voltage, this should be 0.5 higher than your vcore voltage. Ie, in my case 1.300 vcore required 1.800 input voltage for a 4.7ghz oc. No need to set the initial input, just the eventual one. You may also need to play about with the following chipset voltages as you are running 2666mhz ram.

System agent voltage
CPU analog i/o voltage
CPU digital i/o voltage
PCH interfacing voltage

In my case at 4.7 with 2400 mhz ram i needed 1.200 on each. Anything less resulted in memory related blue screen crashes.
 
Thanks both, I'll try Asus real bench

reguarding the:

System agent voltage
CPU analog i/o voltage
CPU digital i/o voltage
PCH interfacing voltage

I'm guessing an easy way to see if fiddling is required is to run the memory without XMP until stable @ 4.7, then turn it back on - if unstable, then go to these voltages?
 
Oooooook.

Having some problems :(

RealBench, with everything stock / XMP (memory) enabled for 2666 - luxmark crashes.

Trying to figure out why before I continue OC'ing - (as originally, thought this was unstable OC...where as it's obviously something else!)

Any ideas?
 
AvengerUK said:
Oooooook.

Having some problems :(

RealBench, with everything stock / XMP (memory) enabled for 2666 - luxmark crashes.

Trying to figure out why before I continue OC'ing - (as originally, thought this was unstable OC...where as it's obviously something else!)

Any ideas?
Click to expand...

The IMC on the CPU not liking 2666MHz and/or the XMP timings is a likely suspect. You should disable XMP, set it to stock 1600MHz, and find out your max CPU overclock first.
 
Well, I've been stress testing all day -

RealBench (the included version of Luxmark) crashes no matter what - so I stopped using that.

With XMP on / 2666 - CPU @ 4.7 - Voltage @ 1.280 - input is at 1.800 -

This passed:

- AIDA64 for 4 hours
- Intel XTU for 4 hours (also did memory for an hour)
- Luxmark 3.1 for 2 hours (ran this because of the previous realbench problems).

Only, upon switching to adaptive 1.280 - it has crashed once (clock watchdog bluescreen) while alt-tabbing out of rocket league!

Is this possibly because of adaptive?

Thanks for all your help so far btw :)
 
Yes because Adaptive adds more voltage in some tests and a little voltage for other things (games etc). Monitor Vcore in realtime to confirm.
 
I've not heard of 'adaptive' before but I would suggest using 'offset' instead. I can't remember the entirety of my configuration but it was something like:

Offset vCore - +0.002 (I think but I'm running negative currently)
Multi set to x46
C states enabled
XMP 2400

If you get a crash while gaming, try increasing the offset and or VCC.
 
Have now done so :)

Now and again it can still crash due to cpu watchdog bluescreen.

Problem is, running AIDA64 / Intel XTU don't crash it! - so I'm not sure if further slight adjustments are helping or not until I crash (or don't) with general usage.
 
Looks like it was the XMP profile after all :/

Not 100% stable yet, (atleast, I'm still stressing).

Set AI to Manaul - input memory timings / voltages / latency manually.

currently @ 1.275v / multi 47 / input 1.8

Using Prime95 27.7 - it's literally the only program that shows up my instability outside of random crashes in gaming etc. Max temp 70.
 
You must log in or register to reply here.
Back
Top Bottom