Any tips for overclocking Haswell?

R

Rizen

[H]F Junkie
Joined
Jul 16, 2000
Messages
9,487
Hi guys, I finally upgraded from my old 930 system and I am working on my Haswell overclock right now.

System is in signature. Cooling is a Black Ice Extreme 360 radiator with three 1800rpm Scythe Typhoons, an EK Supremacy Clean CSQ full nickel block, the chip is delidded and I am direct-die cooling with EK Naked Ivy Standoffs.

So far my chip seems pretty crappy. I haven't been able to get it reasonably stable at anything higher than 4.4, and to get it stable at 4.4 requires ~1.37v.

I am running 1.37v vcore, 1.2v cache, 44x multi, 40x uncore multi, DRAM is set to 2133MHz (lowering does not seem to help), system agent and digital / analog IO offsets are set to +0.20, CPU Power in is set to 1.95v.

With these voltages and cooling I am maxing out in Intel Burn Test at 79C (Intel Burn Test is set to Maximum Stress).

Anything else I can do to increase stability? These voltages seem high, but my temperatures seem pretty good for IBT at these voltages. Should I try to push them any higher? I know these chips run hot, it just seems that most good chips will do 4.5+ at <1.3v.
 
Silicone lottery, and you're pretty close to where I am on my recent 4770k build. I'm at 4.5GHz with 1.325 vcore, 1.9 power in, 1.2v ring w/42 multi, +.255 SA offset. Using a Kraken x60 and two Silverstone AP141's @ 60% on an NZXT Sentry 2, my Prime95 (ver 28.1) small FFT gives me mid 80's with spikes to 92c. I also opted for the delid, but I'm reusing the IHS. My only suggestion is to keep trying, it's taken me two days so far to get an 8hr Prime stable build, and I'm still dialing it in.
 
Try these settings and see if it helps with stability. 1.37 cpu voltage seems high for 4.4ghz. I'm running 4.7 at 1.285v.

CPU VRIN Loadline Calibration - Extreme (Advanced Power Settings)
CPU VRIN External Override - 1.9
CPU Ring - 1.2

Let me know if that works.
 
Nirad9er said:
Try these settings and see if it helps with stability. 1.37 cpu voltage seems high for 4.4ghz. I'm running 4.7 at 1.285v.

CPU VRIN Loadline Calibration - Extreme (Advanced Power Settings)
CPU VRIN External Override - 1.9
CPU Ring - 1.2

Let me know if that works.
Click to expand...
Yeah, I am running CPU Ring @ 1.2v and CPU IN @ 1.9 already. ASRock just calls them different things. If I drop the vcore below 1.37 it is not stable long term in IBT or Prime95. I can boot into Windows at around 1.3v or lower but trying to run Prime95 at those settings results in a near immediate hard lock or BSOD.

I read that lowering the uncore multiplier and running the RAM at slower speeds (e.g. 1600 instead of 2133) can help, but it doesn't seem to make a difference for me. Dropping RAM from 2133 to 1600 hasn't allowed me to run lower voltages. It might make a slight temperature difference but my temperatures are okay so I doubt that will make any noticeable difference.

4.7 at 1.285v seems like you have a very good chip! I know ASUS tested something like ~500 Haswells and found that only 10% can run 4.7. I was hoping to help my chances with the delid+direct die cooling, but it seems that I just get better temperatures without a huge boost to stability.

Any other settings worth testing? Does shutting off the various C States and Power Saving Technology actually increase stability or is it just to rule out conflicts while testing? Are there any adjustments to FIVR I should make? There are a lot of settings in the ASRock BIOS that I am uncertain what they do.
 
Last edited:
1. I'm running with full C-states/EIST/etc. enabled. If you have a stable overclock, it shouldn't make a difference :p

2. VRIN/Input LLC - can you confirm you have this at the max setting?

3. Have you been tracking your BSOD codes? Your VRIN should be closer to 2.00 or 2.05 at that high of a vcore. (E.g. I'm using 1.95VRIN with my 1.284 of Vcore)

4. Are you running 2x8 or 4x4? I think there is a bit more difficulty with running four populated DIMMs - stock or overclocked.
 
Mav451 said:
1. I'm running with full C-states/EIST/etc. enabled. If you have a stable overclock, it shouldn't make a difference :p

2. VRIN/Input LLC - can you confirm you have this at the max setting?

3. Have you been tracking your BSOD codes? Your VRIN should be closer to 2.00 or 2.05 at that high of a vcore. (E.g. I'm using 1.95VRIN with my 1.284 of Vcore)

4. Are you running 2x8 or 4x4? I think there is a bit more difficulty with running four populated DIMMs - stock or overclocked.
Click to expand...
1) Okay, that was my understanding as well I just wanted to make sure it hadn't changed.

2) I don't see this in my ASRock's BIOS. Do you mean CPU Load Line Calibration? Here's the ASRock BIOS settings from HardOCP's review: http://hardocp.com/article/2014/01/29/asrock_z87_killer_fatal1ty_motherboard_review/3#.UvfrqrTX95I

3) I've been getting the same BSOD codes but I forgot to write them down. Will do so from now on. I changed VRIN to 1.97 - this board adds about .03-.05 on top of what I manually set, so 1.97 translates to ~2.02 under load in Windows.

4) 2x8GB 2133MHz Corsair DIMMs.

Thanks :)
 
Yeah seems "Enabled" is functioning as intended if you are getting above your BIOS settings for VRIN. Just didn't want to see it drooping.

As for consistency - I'm curious how it will respond to additional VRIN. There was once a narrative that 101 codes meant add more VRIN, 124/9C meant vcore, but that was not consistent across other mobos to really be an iron-clad rule. *However, it is what I observed in my testing. I suppose that if you were seeing the same thing and the code at least changes with more VRIN, we are making progress :p

I generally increment VRIN 0.05 at a time actually, but let's see how it responds.
 
Thanks. I'm testing some more things now. I raised the VRIN and I dropped the uncore back to stock (35x multi) since it seems to make very little difference in performance but can have adverse affects on core clock stability.

I'm running IBT on Maximum Stress for 10 loops since it's faster than Prime. Right now I am testing 4.5GHz at 1.365v core, 2.00 VRIN, 1.2v ring/cache, +0.25 offset on digital/analog (IMC), and 2133MHz RAM. It's passed one loop of IBT so far, and temps haven't gone over 80C.
 
Thanks. Good information.

I've rolled the memory and uncore back to stock settings and I am just going to focus on the core until I get that worked out. I've almost got 4.5GHz stable and at less voltage than I was running at 4.4 before, but still kinda high (1.36v). I'll post updates in a little while.

Running the memory @ 1333 instead of 2133 drops load temps about 5C, I wasn't expecting it to be that drastic. Peaking around 75C at 4.5 right now.
 
Yeah there was a great deal of misinformation early on (particularly with guides enforcing a 1:1 Core/Unclock as the ideal). Safe to say, this is where the unrealistically low overclocking results came from :p

I would argue that the empiric results, seen in the OCN Haswell Statistics thread better represents real-world results :p Yes, there is always the argument of what is 'stable' (24/7 vs. 3D stable, certain bench stable, etc.), but you will see that 4.5 @ 1.3vcore is actually the average. This actually jives fairly well with what is disseminated in the Asus PDF. (As a reference, my 4.5Ghz overclock requires 1.188vcore at load)

I also found that 2133 was not limiting my core overclocking at all. Of course, there is mobo to mobo variance, but I believe you should be able to eventually set your RAM back to 2133.

Out of curiosity, do you remember what your vcore defaulted to as 'Auto' when you first booted?
And curious if you got a Malay or Costa Rica chip too.
 
I think the early Haswell chips were less refined somehow and they had a lower avg OC as I started binning chips they got better and I saw an avg of 4.4.
 
So I really am leaning towards this cheap just being on the low end.

I can pass Intel Burn Test @ 4.4 with around 1.33v, but if I try to run Prime95 blend it hard locks, BSOD, or just shuts off within 20 seconds every time.

Right now I am running 4.3 @ 1.32v and it seems to be running okay - Prime has been running for nearly an hour without a problem. I tried 4.3 @ 1.3v and it BSOD'd after ~30 minutes of Prime95.

I was always under the impression that IBT was a better stress test than Prime95, or at least comparable. Has anyone seen this problem before? I could almost pass IBT @ 4.5 with only 1.335v, but Prime95 is way more taxing.
 
I can give you a haswell overclocking tip.



You should have bought a SandyBridge. :D
 
Rizen said:
I was always under the impression that IBT was a better stress test than Prime95
Click to expand...

Generally every chip responds differently. I had one that Windows media player visualizations was its overclocking kryptonite.
 
xorbe said:
Generally every chip responds differently. I had one that Windows media player visualizations was its overclocking kryptonite.
Click to expand...
That's hilarious, haha.

This is my first time using IBT. It does get the temps up but seems Prime95 is better for overall stress testing including memory.
 
Well I upgraded too. This are my results at 4.3 GHz.
LHDbLYM.png
 
Okay so it looks like for complete stability at 4.4GHz I am going to need 1.41v. Which is pretty absurd. I am getting a peak temperature of 92C on Blend on all four cores loaded in Prime95 at that speed (usually hovers around 65-75C but spikes to upper 80s/low 90s). I've tried pretty much every BIOS setting, lower uncore, lowering memory, etc. I can easily pass IntelBurnTest at ~1.34v but Prime95 requires much more. I'm not a huge fan of "partial stability" - I want to know the system is rock solid.

How much long term risk is there at this heat / voltage level? Bearing in mind that this is a gaming PC and real world temperatures will probably not exceed 65-70C max.

Edit: I'm just going to dial it down to 4.2GHz. Not worth the crazy high temperatures.
 
Last edited:
These voltages some of you guys are quoting are insane--I can't remember where the info comes from but the rumor is 1.3v is about the safe ceiling. From what I remember it was suspected voltages above 1.3 were a major contributing factor to fast chip degradation. I've had chips that would NOT go past 4.2GHz stable no matter the voltage and this is the only chip I've found (out of 8 4770k's that I was able to test) that can do 4.6 stable. This is at 1.27v manual--which gets bumped by my motherboard to 1.3v (even the UEFI voltage monitor shows 1.3v steady when core is set to 1.27).

I wonder the percentage of people who stress test their Haswell on Auto or Adaptive voltage setting in BIOS/UEFI? Even Valley Benchmark by itself on adaptive voltage setting my chip's IVR calls for 1.36v at 46x/44cache. The entire concept of adaptive voltages for z87 seems bugged to me, or at least it is on the 2 M6 Hero boards I've used. The only voltage setting you should ever use on Haswell is Manual. Then set windows power management to allow your processor % to drop to 5%, and c-states work no matter what.

As for tips on overclocking Haswell, it's the basics, learning about using Manual voltages and other than making sure my System Agent, Digital I/O, and Analog I/O were slightly bumped when I started running into weird problems that prime95 wasn't catching, whoever said silicone lottery nailed it.
 
I run at 1.25v @ 4.25GHz. If I push any faster, Maya3D and After Effects grinds my system to a halt and it becomes very unstable.

My system has been @ 4.25GHz (1.25c) since September of 2013. CPU maintains a constant temp of 55-60c with an H80i CPU cooler. No crashes and excellent stability.
 
Bluesun311 said:
These voltages some of you guys are quoting are insane--I can't remember where the info comes from but the rumor is 1.3v is about the safe ceiling. From what I remember it was suspected voltages above 1.3 were a major contributing factor to fast chip degradation. I've had chips that would NOT go past 4.2GHz stable no matter the voltage and this is the only chip I've found (out of 8 4770k's that I was able to test) that can do 4.6 stable. This is at 1.27v manual--which gets bumped by my motherboard to 1.3v (even the UEFI voltage monitor shows 1.3v steady when core is set to 1.27).

I wonder the percentage of people who stress test their Haswell on Auto or Adaptive voltage setting in BIOS/UEFI? Even Valley Benchmark by itself on adaptive voltage setting my chip's IVR calls for 1.36v at 46x/44cache. The entire concept of adaptive voltages for z87 seems bugged to me, or at least it is on the 2 M6 Hero boards I've used. The only voltage setting you should ever use on Haswell is Manual. Then set windows power management to allow your processor % to drop to 5%, and c-states work no matter what.

As for tips on overclocking Haswell, it's the basics, learning about using Manual voltages and other than making sure my System Agent, Digital I/O, and Analog I/O were slightly bumped when I started running into weird problems that prime95 wasn't catching, whoever said silicone lottery nailed it.
Click to expand...
I've read that 1.5 is the ceiling but that's an insane amount of cooling to keep that from throttling. I don't really feel comfortable @ 1.4v. I did however my i7 930 @ 4.4 using 1.4v for a few years and the only "degradation" I had was that it got slightly unstable and had to be dropped to 4.2. No problems other than that.

Either way I don't usually have CPUs for more than a few years so not too worried about lifespan.
 
You must log in or register to reply here.
Back
Top