Finicky voltage for a 4670k overclock w/ ASUS Maximum VI Hero

This is a dedicated thread broken off the discussion I have been having with Johan in http://www.overclockers.com/forums/showthread.php?t=732646&page=12

My system:

Core i5 4670k
ASUS Maximus VI Hero
8GB Corsair Vengeance 1866mhz
Corsair H100i with 2x Corsair SP 120 Quiet fans, exhausting out of top of case (Fractal Design R4)

This is the current status of my overclocking stress testing, in attempt to nail down a rock solid multiplier/voltage:

All default values set in UEFI bios of the motherboard, memory speed changed to 1600mhz to keep things low, tweaking set to manual to get a multiplier of 42x across all cores (sync all cores). Voltage set to auto.

I have been running AIDA64 with ONLY the FPU option selected in stress test, and this has been running for 8+ hours now without crashing. The hottest one of my four cores have reached is 83c to 84c give or take.

Voltage is fluctuating of course because it's on auto, and I can see it going to around 1.275 on average.

Running P95 Blend, temps stay quite a bit lower (around high end of 60c), and there have been no stability issues. With this said, should I continue benching with FPU in AIDA64, or drop everything and just test with P95 Blend? If sticking with P95 Blend, should I hop to 43x and keep voltages on auto? Try a manual voltage setting instead?


Much thanks.

I would set the voltage at 1.275v since your cooling seems to handle that OK. I personally prefer P95 blend test for stability when doing real work since you intend to mine with it I would stick with that. Bump the multiplier up to 43 and test for about 20-30 minutes. If it's stable and the temps are still in line then try for 44 and so on. Once you find a multi that craps out on you back down one and test for a couple hours with P95 blend test. You shouldn't have any trouble mining after that.
It's all a balancing act with your cooling and there are a few tests in P95 blend that will really kick that CPU in the shorts. So a long test is important. I ran this test for four hours last night and I'm fairly confident in it's stability. Now I realize that you're using an AIO as was I but not all CPUs produce the same heat at the same volts so just because I can run 1.39v through mine doesn't mean yours will. But keeping your temps below90 c is where you want to draw your line. So even at 1.275v make sure you keep an eye on your temps. That's why I recommend 20 minutes, you'll hit some of the "harder" tests in that time frame.

Thanks Johan, I will do that tonight.

This was some hard testing via Aida64 with FPU setting selected last night, you can see the temps across all the cores etc. Not sure if this is useful or not: https://www.dropbox.com/s/oxm8hrj2q5fafca/R14PRpI.png (Note OCCT's stress is not running, I simply have it open to watch line charts of temps min/max. Aida64 CPUID is open to watch voltages (can see the 1.276v active).

Those images won't work while I'm here, don't tell anyone but I'm supposed to be working. Ha ha Try using HWMonitor free it lists all the important stuff plus will show max and mins and current volts/temps. You can see how it's laid out in that link I posted. I only opened CPU-z for verification on CPU, memory and cache speeds.

Afternoon everyone,

Got a batch of new OC results I wanted to share and get some opinions on.

Thoughts of my results:

- Should I keep stepping down the voltage bit by bit? Until I no longer have stability on 30 minute tests?
- Prime95 Blend is getting things toasty... a bit too toasty for my liking. The above point to lower volts might lower this max temp, should I do that?
- Autovolts suck for higher multipliers, I guess it's the lower auto voltages that kick in occasionally that lead to instability?

44x @ auto volts, XMP profile for ram. Aida64 cpu/fpu/cache/memory test.
No LLC set.
Failed. 7 mins in.
Hitting max 1.328v.

44x @ manual 1.328v. (Seen as 1.344v). XMP profile.
Aida64 cpu/fpu/cache/memory stress test.
No LLC set.
Passed. 30 mins.
Max temp 78c.

44x @ manual 1.328v. (Seen as 1.344v).
XMP profile.
P95 Blend stress test.
No LLC set.
Passed 30 min. Failed in temperature.
Occasionally hitting 93c starting at the 20min mark. Temps fine up until that point.


44x @ manual 1.319v/1.320v. (Seen as 1.328v).
XMP profile.
Aida64 cpu/fpu/cache/memory stress test.
No LLC set.
Passed 30min.
Max temp 77c.

44x @ manual 1.319v/1.320v. (Seen as 1.328v).
XMP profile.
P95 Blend stress test.
No LLC set.
Passed 30min. Failed in temperature.
Hitting 93c often.

Personally , if you're looking for a good solid OC that's not going to error on you while mining, Drop the multi to 43 and see if you can lower the voltage enough that 2 hours of P95 blend stays within your comfy temperature zone. When I stabilize my machine for hard work such as F@H that's going to be running for hours on end, even days. If I pass 2 hours of P95 blend I'm golden. Also , I don't know how hard your mining software is on the CPU but I know that F@H isn't as hot as P95. So if my temps are hitting the low 90s in P95, that's fine for me since F@H will stay in the low 80s. I would test the software at your last setting you listed, leave a temp monitoring program open and see how things work out.

MULTIPLIER: 44x
LLC: 8
VRIN: 2.0v
VCORE: 1.32v (1.344v at load)
VCIN/CACHE: 34x

Running MinerD CPU Crypto Mining for 8+ hours now! Previously, before setting VRIN to 2.0v (had it at 1.7v or so), and not touching the VCIN/CACHE, I had max 1 hour of stability before a BSOD. Max temp on any one core is 81c, not too shabby comparing to your F@H which I believe is similar. Unfortunately I do not know how exactly MinerD is stressing my CPU!

If you needed to drop the cache and up the volts I'd say it's working the IMC pretty good. That's why I suggested P95 blend for testing since it works on the same part of the cpu. Glad it's working for ya. You might be able to get the cache speed up if you give it some added voltage, like 1.23-1.25v

44x @ manual 1.319v/1.320v. (Seen as 1.328v).
XMP profile for timings (mhz dropped to 1600mhz for sake of testing).
P95 Blend stress test.
No LLC set. VCIN at 34x (stock instead of auto). VRIN at 2.0v.
Passed 30mins worth of testing. Failed in temperature.
Hitting 93c often...

So how have you made out, are you mining yet??

Mining works, on the following settings.

44x multiplier, 1.32v vcore, 34x uncore, 2.0 vrin, LLC 8.

Thoughts: I am not sure if I need the uncore or the LLC set, uncore is set so it just isn't on auto, since it starts at 34x anyways on this chip.

I literally have almost the same exact setup as you and found that 4.5/4.6 is the upper limit on the 4670k. I had it up to 1.352v at 4.5 but decided to step it down to 4.2 at 1.195v. When I was running it at 4.5 using P95 I never saw my temps get over 57c so, 83c seems really high to me. When I get home I will look at my notes and see what settings I had at 4.5 but each chip is different so im not sure if it will help a whole lot.

peanutbudder said:

I literally have almost the same exact setup as you and found that 4.5/4.6 is the upper limit on the 4670k. I had it up to 1.352v at 4.5 but decided to step it down to 4.2 at 1.195v. When I was running it at 4.5 using P95 I never saw my temps get over 57c so, 83c seems really high to me. When I get home I will look at my notes and see what settings I had at 4.5 but each chip is different so im not sure if it will help a whole lot.

Click to expand...

Hey any little bit helps.

I can't prevent a bsod error 101 running x264 at 45x. Tried all the way from 1.34v to 1.38v, no luck