Originally Posted by Gridsquares
What can you tell me?
Well, the first cold dose of reality is that you're going to need to read up on OC Haswell, because until you do too much will remain a mystery.
Automatic settings are notoriously incorrect, most of the time. You were lucky on your previous example, though I admit I was able to run my system at 4.4 using the automatic settings, I couldn't push further using them.
Automatic doesn't mean intelligent. The system isn't choosing values for your chip, it's choosing averaged settings from some charts. Because of this, some people get better results after upgrading the UEFI, but in some cases that makes it worse.
The only way to really get a satisfactory, stable result is to gain control.
The basics, which you probably know most of:
- There are multiple voltages:
Inside the 4770K, there are separate voltage controls for cache, CPU, IGP and system agent (I may be forgetting one). You'll concentrate on one at a time, the CPU primarily.
- There are TWO multipliers to consider:
Cache has it's own multiplier, and when you raise the CPU, the cache can (and should) follow behind. Example, CPU @ 4.4, cache at 4.0 or 4.2, CPU @ 4.6, cache at 4.2 or 4.4.
- The motherboard's own Vrin is "outside", the others are "inside".
The VRM's on your motherboard should be providing about 1.9v, or at least that's the common suggestion. Intel's statement is that it should be .4 to .6 higher than the highest of your internal voltages, like the CPU (which is usually the highest).
- Fixed is better.
This is a matter of opinion, but adaptive voltages and changing clock speeds complicate everything. Some get it to work even when they're pushing some overclock, but it's tough to control. Fixed clocks and voltages are much smoother to figure out and stabilize. I disabled all C states, set all cores to keep the same speed, set fixed voltages on all options. Idle temps will be a tad higher, but you'll have better control over the load temp targets as you experiment.
It seems to me, too, that this is the way to understand your chip first. After you have figures to work from, you may be in a better position to configure adaptive settings if that's your goal.
- Test with determination:
No one want's BSOD's, but you need them. You want them during your examination of the CPU. This is freaky on Windows - turn off your disk write caching! That's not usually in most guides, but you CAN (not will, but can) crash your Windows installation with directory corruption with OS lockups.
You learn less when you can boot into a higher clock than when you can't boot. It's simple logic. You can't prove a negative. You can't prove the CPU WON'T fail. You can only prove when it does.
Here's how I approach it.....
Target a speed...say, 4.2 or 3.7, any target.
Focus on CPU Vcore, set the Vrin (your motherboard's supply) to 1.9 or 1.8 (it's not critical yet) - all options fixed.
Set to 1.1 or 1.2v, something you know works.
Boot, run short tests (60 second OCCT is fine). It probably passes.
Now, set to 1.0v. It probably crashes (if it didn't, try 0.9 until it does).
Now, choose half way between the crash and non crash voltage....that is, good a 1.1, crash at 1.0, next try 1.05.
If it passes at 1.05, repeat.....1.0 failed already, 1.05 passed, try 1.025.
Fails...now, between 1.025 and 1.05, switch to 1.0375 (which usually must round to 1.037 or 1.038).
Doing this finds that one spot, within about 2 hundredths of a volt, that you KNOW fails.
The one's that pass are NOT CERTAIN...you have little information available on what DOES work, but now you know what does not.
Move up a notch in speed. If you ran that at, say, 3.7, move to 3.8 and do it again, knowing that at the higher speed it may not work at the same voltage you found at 3.7 (sometimes it does), but that's the point.
You're going to end up with a chart showing the line of failure/pass at each speed going upwards. You'll learn your chip's curve, where it needs more voltage at particular speeds.
The lowest pass voltage is not your target voltage for your setup, it's your absolute minimum to boot. From there you can judge a margin of safety upwards.
Typically you can say that the notch above the failure point of any speed is close to the operating point of the next lower speed (not always, some speed "rows" are about the same).
For example, let's say you learn that your chip fails at 1.16 at 4.2 Ghz, but passes at 1.18 at 4.2Ghz.
If you end up running at 4.0 Ghz, you know it will also pass at 1.18. If your chart shows it failed at 1.14 but passed at 1.16, 1.18 is a STARTING point for testing long term stable operation.
Now you're working on information. You can consider an engineered margin above the absolute minimum the chip requires.
Likewise, if that chip failed at 1.19 at 4.4, but passed at 1.2 at 4.4, it's likely that at 4.0 1.18v or 1.19v is going to be stable. It's not certain, but likely.
At least you know you don't have to keep pushing upwards to 1.25 targeting 4.0 on a chip with that curve.
Typically, 3 to 6 hundredths (.03 to .06) voltage above the lowest passing test is about right.
All that, assuming you ignored cache.
You target some speed using this kind of research, then bump cache a nudge in voltage if you have long term stability problems (or maybe it's RAM, system agent....etc).
Read up. Find a stable, near stock speed you can rely on for daily use, and plan on frequent trials with charts and notes. It will take a few days, maybe a couple of weeks - and you'll want to revisit this "series" of tests later, since the chip changes a little as it ages.
Once you learn your chip's personality, you'll know what it doesn't like, what it needs, and where it can go.Edited by JVene - 1/14/15 at 12:32pm