Overclocking i7 4790k

[ProfitMajin]

I didn't even think of using it under the IHS tbh, I would agree that LM seems to be the best for under the IHS, and hopefully, that would do the trick regarding the hotspots.

Mostly I was thinking of using the pad on top of the IHS to even out any potential unevenness that could maybe cause hotspots? Reviews/tests with the product are quite scarce at the moment, the ones I have seen compare it to products such as MX4, so I'd imagine in comparison to Kryonaut it may come up lacking.

M

 

[ProfitMajin]

I've just been watching a review from GN on the Radeon VII where he tested the stock solution which used a type of thermal pad, and then replaced it with Kryonaut.

In the test, Steve concluded that when the die is uneven, the pads will always come out on top, as they're superior for ensuring contact, but on an even surface, the paste should win due to being thinner.

I've seen some other reviews from GN on needing to LAP the later i9 generation Intel CPUs as they're uneven, but I can't seem to find any info on whether or not this was an issue for 4th gen. If they're also known for being uneven, it might be best to opt for the pad.

M

 

[an0nym0us]

I didn't even think of using it under the IHS tbh, I would agree that LM seems to be the best for under the IHS, and hopefully, that would do the trick regarding the hotspots.

Mostly I was thinking of using the pad on top of the IHS to even out any potential unevenness that could maybe cause hotspots? Reviews/tests with the product are quite scarce at the moment, the ones I have seen compare it to products such as MX4, so I'd imagine in comparison to Kryonaut it may come up lacking.

M

I don't think anyone meant for you to use the Carbonaut thermal pad between the CPU die and IHS. If there was uneven contact between the CPU die and IHS, then the the IHS may potentially require lapping (on inside and out, basically wherever it's not flat). However, this is unlikely the main cause of your problems. Your high and uneven temps are most likely due to the original TIM drying up and no longer being effective. The thermal conductivity of liquid metal is far better than the thermal pad, so since there is very minimal chance of liquid metal having a chemical reaction between the CPU die and the inside of the IHS, the choice would nearly always be to use liquid metal for this interface.

If you wish, you can use the Carbonaut thermal pads or regular thermal paste between IHS and cooler. Either should be fine. I've not tried the Carbonaut thermal pads yet personally, but the reviews seem decent, so I'm actually very keen to try them out.

 

[ProfitMajin]

I don't think anyone meant for you to use the Carbonaut thermal pad between the CPU die and IHS. If there was uneven contact between the CPU die and IHS, then the the IHS may potentially require lapping (on inside and out, basically wherever it's not flat). However, this is unlikely the main cause of your problems. Your high and uneven temps are most likely due to the original TIM drying up and no longer being effective. The thermal conductivity of liquid metal is far better than the thermal pad, so since there is very minimal chance of liquid metal having a chemical reaction between the CPU die and the inside of the IHS, the choice would nearly always be to use liquid metal for this interface.

If you wish, you can use the Carbonaut thermal pads or regular thermal paste between IHS and cooler. Either should be fine. I've not tried the Carbonaut thermal pads yet personally, but the reviews seem decent, so I'm actually very keen to try them out.

Excellent thank you!

Yeah, I think I might well give it a go, and I suspect that you're right that the original TIM is likely the reason for the uneven core temps, but then saying that it seems like the pad performs well enough versus paste, and there is some QOL perks to be had with the pad.

Thank you once again!

M

 

[ProfitMajin]

Oh!

I did also forget to mention, I've been messing around with the various bells and whistles in BIOS, and I managed to encounter a very strange issue where my primary boot drive could no longer boot, either the PC would hang immediately after successfully passing POST, or it would simply go into the BIOS.

Without going into too much detail, I resolved the issue by disconnecting my other drives (so the PC only had 1 boot drive), and reconnecting after successfully reaching Windows. It seems like this issue is caused anytime I pushed the system agent voltage too high, but could also be caused by pushing the BCLK too high as well.

Currently, I've limited system agent voltage to adaptive + 0.199, which I've seen as a common upper limit, and my BCLK is 105 at present. I saw mention that increasing BCLK can cause instability on SATA drives, something to do with them not being designed to go faster than x1.00?

I'm keen to not encounter this issue again in the future as I lost a whole day to work this out! It is possible that there's something else going on, but I only found this issue when I was pushing for higher memory clocks, which is when I increased the BCLK and system voltage.

M

 

[an0nym0us]

Yes, increasing BCLK is going to overclock many parts of the system all at once and can cause instability very easily. You may be able to isolate the effects of the increase in BCLK to just the CPU and RAM if your motherboard allows it, but no idea if that's possible.

For your own sanity, it should be far easier to get a stable overclock by leaving BCLK at stock 100, then overclocking CPU and RAM individually.

 

[ProfitMajin]

Yes, increasing BCLK is going to overclock many parts of the system all at once and can cause instability very easily. You may be able to isolate the effects of the increase in BCLK to just the CPU and RAM if your motherboard allows it, but no idea if that's possible.

For your own sanity, it should be far easier to get a stable overclock by leaving BCLK at stock 100, then overclocking CPU and RAM individually.

I've managed to stabilise the system at 105BCLK, which has resulted in final clocks of:
VCCIN 1.98v
4.83Ghz Core with 1.28v
4.52Ghz Cache on auto voltage
1960Mhz Memory 9-11-11-28 T1 @ 1.64v
System agent adaptive +0.149
System Digital I/O Adaptive +0.149
System Analogue I/O Adaptive +0.199

I'm going to leave it at this stage for now as it seems to be a pretty good overclock for the 4790k, and will wait until I've delidded to try and push it further. It has passed Prime95 266 for around 70mins, can pass CineBench with a score around 965, HyperPI passes successfully, and I've run memtest86 overnight and that passes too. Temps during stress testing do not exceed 86C, normal usage is around 75C max load.

The thing that really confused me is that issue with the drives, it only seemed to happen when I pushed the system agent voltage higher, but I suppose instability is instability, and could have been caused by the voltage, or the BCLK, and hopefully I'll not encounter it again.

M

 

[an0nym0us]

Glad you managed to get it stable.

 

[ProfitMajin]

Glad you managed to get it stable.

Or so I thought!

I managed to improve the above by reducing my RAM and System agent voltages slightly, and had passed all stress tests, and played games for some time. I believed this to be stable until I downloaded the updated OCCT yesterday, 5.0.1, and long story short - my PC is unable to pass any of the OCCT tests at my current settings; PSU test, CPU 2009 test, and latest CPU 2019 tests all fail after around 10 seconds. I tried increasing my core voltage to 1.3 (from 1.28), and my VCCIN to 2.06 (from 1.98), and only just about reached a 'stable-ish' level, where the test could run for a short while before triggering thermal limit, which is to say it isn't stable.

It seems like the CPU can run at this speed, but I'm currently not able to keep it cool enough at the required voltage. My question is - why would all the other stress test tools pass? - I've used Prime95 266, CineBench, 3DMark, HyperPI, and have played games for several hours, all without any noticeable issues. I'm not saying the test is wrong ofc, I do now believe my system is unstable, but I'm confused how OCCT is the only tool that has caused the system to fail, and damn fast too, barely reaching the 10 second mark each time I tested.

I'm reverting back to default BCLK, incase that's the cause for instability, and will see how that goes. I cannot currently keep the CPU cool enough when supplying 1.3v to the core, and so I'll have to back off of the overclock until I can de-lid. I don't mind that at all, as I'd rather the system be completely stable, I'm just stumped as to whether or not I've missed something here.

M

 

[an0nym0us]

I can't advise on whether your specific settings are reasonable, but please double check the safe voltage limits as these will vary from one CPU architecture to another. The VCCIN sounds high, but I don't know, so best to double check all the voltages.

Since you were originally getting large variations in temperature between cores, I definitely would recommend being more conservative on the voltages and overclocks you apply, then settle the temperature issue as a priority.

 

[pp111]

Agree with the delid recommendation. The 4790K is a very nice CPU but the thermal interface that intel fitted is just poop with the end result that the CPU runs too hot and there is not a lot you can do about that other than delidding it.

 

[ProfitMajin]

Hi Guys,

After further investigation I've been able to stabilise the system, it seems that the BCLK at 105, even with all other clocks being conservative, was causing the OCCT test to fail. When reverting to BCLK of 100, with voltages and clocks being mostly the same, the test is now able to pass, and does not overheat.

I'll stick to these settings until I can de-lid, then I'll revisit. I'm not sure why the BCLK is causing instability in the OCCT test, yet is able to pass all other stress tests, but it's absolutely not worth the hassle for a 30Mhz increase on core. So my settings as of now:

BCLK 100
Core 4.8Ghz @ 1.285v
Cache 4.5Ghz @ 1.28v
RAM 2,000Mhz @ 1.57v
VCCIN 1.9v
System agent offset +0.1
Max Stress Test Temp 83C

So all my voltages are now within safer limits, in that they're white rather than yellow/purple as they were before. Oddly enough though, I've found that if I set BCLK to 100 in BIOS, in Windows it shows as 99.7. Even setting the BCLK to 100.2 makes it 99.9 in Windows, yet when I set 100.3 in BIOS it'll be 100.3 in Windows... I'm not too concerned about this, but I do find it odd!

Thanks again for the replies, I'm looking to delid this weekend or the next. I'll be using Kryonaut instead of the Carbonaut pads as I'll practice on my old 3770K first, and I'll use the Conductonaut and Kryonaut on both CPUs. I'm thinking the 3770K will benefit most as it's very old by now, the TIM is probably dust in there... I'll do some temp testing to gauge the improvement for both CPUs and report back once done!

M

 

[pastymuncher]

You are best off leaving bclk alone on Haswell because it overclocks everything including the pci-e bus and that's most likely why it's unstable. Your overclock of 4.8Ghz is excellent for a 4790k and would put it in the upper range for that cpu. There is the possibility of 5Ghz after delidding if you give it more vcore, ideally no more than 1.35v although you should be ok up to 1.4v vcore if you can keep the temps under control. Sadly no matter what I tried my old 4790k wouldn't go any higher than 4.6Ghz and 4.4Ghz cache.

 

[ProfitMajin]

You are best off leaving bclk alone on Haswell because it overclocks everything including the pci-e bus and that's most likely why it's unstable. Your overclock of 4.8Ghz is excellent for a 4790k and would put it in the upper range for that cpu. There is the possibility of 5Ghz after delidding if you give it more vcore, ideally no more than 1.35v although you should be ok up to 1.4v vcore if you can keep the temps under control. Sadly no matter what I tried my old 4790k wouldn't go any higher than 4.6Ghz and 4.4Ghz cache.

I think you're absolutely right here, all the other smaller elements being overclocked seems to have affected something that tripped a failure in OCCT. Overvolting by a significant margin seemed to begin to stabilise this, until thermal limits ofc, but even then I was likely losing performance by trying to hold the BCLK overclock, and reverting to BCLK 100 has made it far more stable, and voltages are much more reasonable now.

With regards to your 4790k, I don't know if you're encountering the same issue I had, but my Cache clock will absolutely not go above 4.6Ghz unless it has a very high voltage, whereas keeping it to anything under 4.6Ghz it's much less troublesome. It may be that your cache is hard limiting you to 4.4Ghz, but could your core potentially climb any higher?

I know all CPUs are different, for mine I was able to boot into Windows with my CPU core running at 4.9Ghz as long as I kept my cache under 4.6 - but it wasn't stable, it'd hit 94C in tests, although the delid would help with this.

After the delid I'm hoping I'll be able to supply 1.35v to the core and see what rate it can reach, I'll probably need to up my VCCIN though, what's a safe upper limit for VCCIN for 24/7 usage? I was thinking of topping out at around 2.05, which is just a little under the last purple setting in my BIOS, after which it reads in red which is just seems like a bad idea

M

 

[Danny75]

Around 2.0v sounds okay for VCCIN. You may end up not needing that much, play with bringing it down after you finish messing with the overclock. I can leave VCCIN at 1.8v and it makes no difference. Boards and CPUs differ of course.

When it comes to Cache, some mobos set 1.2v as the "red". I'd go up to 1.3v on that as I know plenty have, and I've been running 1.24v for about 6 years on a 4770K Cache. But you may not want to go as high on that as with the Vcore.

Whereas 1.4v for 24/7 sounds like a good limit for CPU Vcore, temp-dependent, as Pastymuncher already said.