Intel I7 4770K Overclock Limit

[RoboX]

I'm trying to overclock an I7 4770K mounted on an MSI Z87-G41 motherboard with 2X4GB DDR3 1886. (Latest Bios, Water Cooling)

To find the maximum CPU frequency I set the RAM to 1600 with 1.5v, and I set the ring ratio to 30 (3.000 Mhz) with VCCIN 1.9v, I also disabled LLC EIST and C-state.

By doing stability tests with LinX, the CPU Ratio at 45 does not hold up (crash) even with 1.4v of vcore. The voltages are set in override mode.

At the moment, I'm finding stability only with a CPU ratio of 42 (and I'm trying to find a lower voltage, such as 1.32v).

From what I read on these CPUs and on the results achieved (or in terms of maximum frequency, or in any case of lower frequencies but with decidedly lower voltages) I believe that in my case there is something wrong.

Can you tell me if it's a normal situation?

Also, how can I act to try to achieve stability at higher frequencies or, if I have to stop at a frequency of 4200, at least with a lower voltage?

I enclose some screenshots of the bios of how I set in this multiple (very bad) result:



upload image



upload image



upload image

 

[GorbazTheDragon]

What kind of temperatures are you getting under load?

4.3@1.375v/95c with the relatively weak VRM of that board wouldn't surprise me one bit... It's definitely on the lower end of 4770ks but not unheard of.

 

[Urbklr]

Sounds pretty similar to my 4770K experience, and a couple buddies I know too. My chip would do 4.3-4.4 on safe voltages, but took a huge voltage increase to hit 4.5.

 

[FreedomEclipse]

That's quite a trash motherboard for overclocking. Just saying

 

[natr0n]

Usually 4.2 is the safe spot for chips around that time.

Looking at that board you're using I would personally put a fan over that tiny vrm area. If you want some form of stability.

 

[DR4G00N]

It could be getting too hot with LinX, or it's just not a great chip. Try playing a game to test it instead of running a stress test.

The mobo could be another reason why, bottom of the barrel boards tend to oc fairly poorly.

 

[Jism]

Disable 2 cores, and try to see if it's stable. If it does your against a current limit and not a chip's limit. Current = your VRM, so make sure you have all the settings turned on that the CPU is allowed to use more current then specified by Intel for example.

Was a typical problem with FX bulldozer too. Most people top out at 4.2 ~ 4.4ghz no matter what voltage. Once you disable a few cores, the maximum current goes down, and relatively easy to test if that's the culprit or not.

 

[king of swag187]

Honestly, could just be the VRM giving out on you

 

[Jism]

Honestly, could just be the VRM giving out on you

You have to override if your board is capable enough, any of the limit imposed by Intel or the motherboard. My older Crosshair Formula Z had that option to disable that feature so it would be allowed to use more current then AMD's own 25A implementation. (25A / 1.3V = 260W).

But if your board does'nt have that option, you could do only 3 things:

- Settle in on the maximum clockspeed
- Disable one or two cores and try to push harder
- Upgrade motherboard

 

[eidairaman1]

Vrms need cooling too.

 

[RoboX]

I found the established at 4200 with a vcore of 1.2v (20 cycle of linx with good temperature).

If I increase the frequency, the necessary vcore increases excessively, minimum 1.4

Is this situation normal?

Furthermore, if I delid the cpu and do the tests with 10-15 degrees less than the 100 (actual at 4500 under linx), could it change something in the stability at 4500@1.4v?

What kind of temperatures are you getting under load?

4.3@1.375v/95c with the relatively weak VRM of that board wouldn't surprise me one bit... It's definitely on the lower end of 4770ks but not unheard of.

NOw at 1.2v very good...but 4500@1.4-1.42 90-100

Usually 4.2 is the safe spot for chips around that time.

Looking at that board you're using I would personally put a fan over that tiny vrm area. If you want some form of stability.

Yes, I've already put a fan on the VRM heatsink

It could be getting too hot with LinX, or it's just not a great chip. Try playing a game to test it instead of running a stress test.

The mobo could be another reason why, bottom of the barrel boards tend to oc fairly poorly.

I prefer to test with linx to be sure of stability in every situation. Perhaps a game is too uncertain as a test. It would take something in between.

Disable 2 cores, and try to see if it's stable. If it does your against a current limit and not a chip's limit. Current = your VRM, so make sure you have all the settings turned on that the CPU is allowed to use more current then specified by Intel for example.

Was a typical problem with FX bulldozer too. Most people top out at 4.2 ~ 4.4ghz no matter what voltage. Once you disable a few cores, the maximum current goes down, and relatively easy to test if that's the culprit or not.

It seems like a good solution. Do you say that this way it works like a CPU limit check? Does the CPU limit make no difference if they are only working 2 cores? (I ask this because the voltage required and the stability of the CPU changes significantly, for example with HT activated or deactivated)
Also, aren't there any other variables?

 

[GorbazTheDragon]

For Haswell I wouldn't recommend going much over 1.3v, they quickly get uncoolable even with a delid. The degradation is also reported to start somewhere around 1.35v/80c...

4.2@1.2v sounds pretty normal to me, definitely not a good chip but not absurdly bad either. Just for reference my fastest G3258 does barely over 4.6GHz at 1.36v for cinebench, I would expect that's only 4.5 daily stable. These early design Haswell chips generally had pretty inconsistent clocks, some 4770ks clock like 4790ks into 4.7-4.8 territory, but others barely above 4.2.

 

[Jism]

It seems like a good solution. Do you say that this way it works like a CPU limit check? Does the CPU limit make no difference if they are only working 2 cores? (I ask this because the voltage required and the stability of the CPU changes significantly, for example with HT activated or deactivated)

No dude, if you think that you ran into a clock limit, i.e caused by VRM's or a specification put up by intel, the best way to test is to disable 2 cores, and see if the thing wants to clock higher. If it does, then your running into a current limit which was my initial point. There's 2 ways to bypass that: get a better board and cooling, or just settle in with what you can get.

 

[Keullo-e]

For Haswell I wouldn't recommend going much over 1.3v, they quickly get uncoolable even with a delid. The degradation is also reported to start somewhere around 1.35v/80c...

4.2@1.2v sounds pretty normal to me, definitely not a good chip but not absurdly bad either. Just for reference my fastest G3258 does barely over 4.6GHz at 1.36v for cinebench, I would expect that's only 4.5 daily stable. These early design Haswell chips generally had pretty inconsistent clocks, some 4770ks clock like 4790ks into 4.7-4.8 territory, but others barely above 4.2.

I remember my G3258 running @ 4.7GHz 24/7, benchmarks ran fine @ 4.8GHz but it needed too much voltage for comfortable 24/7 use. Delidded it also just as an experiment (my first delid, delidded with razor) and as expected, temps dropped only few degrees since it didn't run hot before delid. IIRC I had a Thermalright Macho Direct as a cooler.

Got another as a donate, I should grab a motherboard and some DDR3 for it.

 

[GorbazTheDragon]

Yup, mine are all down on the lower end of the ladder, hoping to pick a few more up in the next few months and maybe get a few ones to run with DICE at some point. I'll probably end up using the worse samples as degradation test subjects...

Quite a fun learning platform for OCing IMO, quite simple, getting a lot cheaper now, and there's still plenty of boards and fast DDR3 to go around.

 

[Keullo-e]

Would be interesting to know how many Intel produced those, since it was a "limited edition" and still feels like those are everywhere and you can grab one with very cheap price.

In fact I also ran that one of mine with the copper-core Intel stock cooler, still have it so I guess I'll be using that also later on.