Which 24/7 OC would you use?

[Qwinn]

Thanks for your thoughts everyone.

Ye old 4.4Ghz settings were still going strong at 6 hours OCCT when I stopped it. Max temp spike 75c. Not going to waste the rest of the day running OCCT, it's gaming time. I will try leaving it for a longer test while I'm at work later this week. Whether that's on this modded 1801 or the new 1902 that just came out depends on whether we can find out if the microcode has been updated in 1902. I wish they'd give more detail on what's changed than "Improves Stability".

Qwinn on the other hand has the swift and a pair of 980ti's which at max game settings except AA of course, I bet he's pulling close to 144FPS in most games. I would say he would get anywhere from 5 - 10FPS increase if he overclocks his cpu and gpu's.

You're right that in 2D I could run most anything at 144fps. The thing is, I run pretty much every game in 3D Vision. This effectively halves your framerate. In Witcher 3 at 1440p on all max settings (except vsync, blur effects and AA, don't want or need any of those) in 3D Vision I can get 35-70 FPS. OC'ing does get me another 3-5 FPS in that which is definitely worth it. 3D Firestrike keeps telling me I am CPU bound at this point, hence the desire to OC to 4.5Ghz (which I've finally established to myself I *can't* get entirely stable without a new cooling solution or a new magic BIOS).

Nate, you're probably right. I'm not throwing away the 4.5Ghz OC settings that lasted 4.5 hours, it's a good backup for benchmarks and stuff at least. But for 24/7 I prefer stable-as-stock, that way if and when I do crash I can be pretty certain it's an app conflict or something along those lines, which could be resolved some other way. For a few months I was running at 4.5 Ghz 3.8 Ghz cache at 1.275v in the BIOS. Thought it was stable, though I would still crash in Witcher 3 once in a while. Then I tried OCCT. Failed in ten minutes :/ Since I switched to the 4.4 Ghz settings I don't think I've crashed at all except for overambitious GPU overclocking (which, incidentally, passed 8 hours of Heaven Benchmark in 3D Vision, but can still crash Witcher 3, sigh). I'll be testing a lowered GPU OC more today.

[Qwinn]

Ok, so, if anyone's interested, here's where I'm at now. (Truth to tell, I post these for two reasons: 1) a "hard copy" documentation of my setting for when I run an attempt, and 2) seriously, posting on forums while running OCCT frequently causes me to crash/freeze where I wouldn't have otherwise, it's like the ultimate stress test.)

Having decided to stick with 4.4 Ghz, I'm going for a mild cache overclock to go along with it, and let me tell you, it's a frikkin' ***** to get even the mildest cache overclock stable on my chip. I'm having decent results with the following though (3 hours OCCT + forum posting so far):

44x100 core, 38x100 cache
LLC: 6
Vcore: Adaptive 1.25 (1.28 under OCCT load)
Vcache: Manual 1.1v
Vinput: 1.95v (1.89v under OCCT load)
Memory: XMP settings (2666 15-15-15-35) @ 1.2v
CPU and DRAM Power Phase: Extreme
CPU and DRAM SVID: Disabled
VR Fault Management: Disabled
Internal PLL Overvoltage: Disabled
Max CPU package temp spike after 3 hour OCCT: 81c
Max CPU hottest single core spike after 3 hour OCCT: 76c

If this turns out to be stable, I might try running it again with a little lower vinput, but otherwise I think I'm happy with all that for permanent 24/7 use. Reasonable? I'm most concerned with the vinput, is that setting ok for 24/7? I'm pretty comfortable with the other settings and temps under stress.

[kaese]

Ok, so, if anyone's interested, here's where I'm at now.

Thanks for posting - it is interesting to see the "full" parameter set of other peoples OCs.
Regarding the cache OC I made similar experiences (with two different 5820K, one from Costa Rica, mfg date mid 2014; one from Malaysia mfg end of 2015).
A multiplier of 38x for the cache (and 1.2V) was only nearly stable for both CPUs (actually it was bench-stable but not for heavy real-life & scientific applications).
That is why I decided to stay with a 36x OC what could be achieved easily rock stable, in my case at 1.15 V.

I must add that I want to use adaptive voltage for the cache with additional turbo voltage (I dont wont to put the cache at 1.15 V all day during idle).

[Qwinn]

I must add that I want to use adaptive voltage for the cache with additional turbo voltage (I dont wont to put the cache at 1.15 V all day during idle).

Funny you should say that. I've read in half a dozen places that adaptive supposedly doesn't work at *all* for cache. And yet, I have gotten it mostly stable before. When this run is done I may try it again with changing to Adaptive 1.1 as my only change.

Here's the thing: I came to that 1.1v for 3.8Ghz setting by running cache at 38 multiplier with adaptive 1.2v total cache voltage. When I checked in HWInfo64 during an OCCT run, though, even though I was giving it leeway to go all the way up to 1.2v, it never actually went over something like 1.06v. So I plopped in manual 1.1v and so far it's fine. You may want to try a similar experiment and see what it actually uses under adaptive.

[Pleot]

Im starting to have a grudge on OCCT

I can run 8 hours on AIDA and 8 hours on RealBench without a glitch - where temps wont go above 70

OCCT will rise temps up to 84 on the same conditions and fail after 5 or 6 hours with BSOD. What does OCCT do to my processor and is it really instability? Everything seems fine on real life so...

[Qwinn]

"OCCT will rise temps up to 84 on the same conditions and fail after 5 or 6 hours with BSOD. What does OCCT do to my processor and is it really instability?"

I think it is. I passed all other bench tests too, but would then, rarely, like every few days, crash in Witcher 3. I figured it was just app conflicts, or issues with 3D Vision and the fixes I had installed. When OCCT discovered some instability and I resolved it, the crashes in Witcher 3 went away too. I suppose if you *never* crash in your games or your other apps, then there's no need to worry about OCCT, but if you ever do, OCCT is pretty damn good at establishing, or eliminating, instability as the cause.

I actually think I just discovered why I haven't been able to get an OC on the cache working. I didn't realize it needs a boost to vcore as well (I know, Raja says so in his guide, but I guess I fixated on the system agent part as I was less familiar with it and overlooked pumping vcore). I'm approaching my attempts from the other side now, giving it more than I think it'll need, and if I can get stability then I'll start dropping voltages till I get unstable.

[AdamC]

Max CPU package temp spike after 3 hour OCCT: 81c

My air cooler NH D15 does a better job than your AIO water cooler. I have never seen above 80C, not even upper 70's in stress test with below settings.

44x100 core adaptive +0.250v System Agent +0.030v everything else auto
40x100 uncore
2666 15-17-17-35-2 auto 1.2v

[AdamC]

Im starting to have a grudge on OCCT

I can run 8 hours on AIDA and 8 hours on RealBench without a glitch - where temps wont go above 70

OCCT will rise temps up to 84 on the same conditions and fail after 5 or 6 hours with BSOD. What does OCCT do to my processor and is it really instability? Everything seems fine on real life so...

Didn't know this OCCT will stress CPU more than realbench or prim.

[Qwinn]

OCCT is the only thing I've run that can get my temps that high. Think highest I've seen outside of it is in the high 60s.

[Pleot]

prime is an assassin for haswell-e processors, I run it for abut 5 min and then halt. Anyway if there is serious instability it will fail right away. Tried running it longer but when my temps reached 94 I stopped On the same conditions Realbench wont go beyond 75