[SOLVED] Fine tunning OC with i9-10900k at 5mhz

[Crimson Morrow]

Hi there!
So I overclocked my 10900k to 5ghz, and when I play games or work on music in a DAW(processor demanding stuff) it nicely bounces around 4.8 and 5ghz (it runs at that frequency even when it's idle cause I set it to constantly be at that speed), with temps being under 70 degrees ( 50-60-65ish most of the time(fans are barely spinning too)). But when I run a stress test with aida64 and it goes to 100% load but the frequency drops down to 4.69ghz. Aida64 says its not throttling thought. Any idea what is happening here? Is it normal?
And yeah, the temps while stress-testing jump around 84-86, which from what I've seen on the internet is normal. But would also like to hear your thoughts on that as well.
Im using a Noctua NH-D15 with it.

One more thing about the temps. I used to have a problem where it would run in mid 70 to mid 80 in games before and would go above 100 under stress test, but then I reapplied the thermal paste and it become acceptable, 84-86 as I said earlier.
So yeah. Looking forward to hear what you guys think

SPECS:
MSI MAG Z490 TOMAHAWK
i9-10900k (with Noctua NH-D15)
3080ti
Kingston hyper x fury 32gb
Corsair 850w

 

[rgd1101]

power issue?
I think you mean 5Ghz.

 

[Crimson Morrow]

power issue?
I think you mean 5Ghz.

Yep Ghz, changed it. thnx.
I'm running on a 850watts corsair, should be enough though. Isn't it?

 

[rgd1101]

what the motherboard?

 

[Crimson Morrow]

power issue?
I think you mean 5Ghz.

what the motherboard?

MSI MAG Z490 TOMAHAWK

 

[Karadjgne]

You haven't OC'd the cpu, you've raised the Turbo limits. Meaning the cpu will be power limited to the Turbo long/short duration power/current limits. Once you hit that power limit, the cores get downclocked to maintain core use and power use.

In other words, if your limit is 250w, it's going to hit 250w with 6 cores at 5.0GHz or 10 cores at 4.5GHz. That's something different to being thermally throttled.

With your OC, it's more than just extending the multiplier and locking the cores, you must also change the internal settings in bios to be able to deliver the power requirements of that multiplier. Without both, you'll get one of 2 results, downclocks under a stress test or a bsod.

 

[Crimson Morrow]

You haven't OC'd the cpu, you've raised the Turbo limits. Meaning the cpu will be power limited to the Turbo long/short duration power/current limits. Once you hit that power limit, the cores get downclocked to maintain core use and power use.

In other words, if your limit is 250w, it's going to hit 250w with 6 cores at 5.0GHz or 10 cores at 4.5GHz. That's something different to being thermally throttled.

With your OC, it's more than just extending the multiplier and locking the cores, you must also change the internal settings in bios to be able to deliver the power requirements of that multiplier. Without both, you'll get one of 2 results, downclocks under a stress test or a bsod.

Thank you for your reply!

Yeah I know that, I've set the core voltage to 1.320 which should be enough for 5ghz. Isn't it?
The other two values are set to 1200 (don't remember what they are, but one is responsible for the memory exchange in the processor between the components, as far as I understand, and the other one, I frankly don't remember what it does lol)

 

[Karadjgne]

You'll need a lot more than that. LLC, turbo long/short values, current limits, disable phase controls, disable c-states, Ring voltages, VID, AVX offset, and more besides that.

Just bumping up VCore and VCCIO/VCCSA and the multiplier isn't an OC, nor is it stable, nor is it healthy for the cpu in general.

The fact that you have no clue what you did, why you did it, or honestly what you are doing, says volumes.

You can't apply half an OC and expect to get the benefits of a fully applied OC.

 

[Rogue Leader]

You'll need a lot more than that. LLC, turbo long/short values, current limits, disable phase controls, disable c-states, Ring voltages, VID, AVX offset, and more besides that.

Just bumping up VCore and VCCIO/VCCSA and the multiplier isn't an OC, nor is it stable, nor is it healthy for the cpu in general.

The fact that you have no clue what you did, why you did it, or honestly what you are doing, says volumes.

You can't apply half an OC and expect to get the benefits of a fully applied OC.

THIS.

The CPU is behaving as expected for the settings you have applied. You've jacked voltages for no reason, which are going to just wear down the CPU for no reason, and in fact over time make it less stable. If you're going to OC you need to read and understand what you're doing and why and apply settings accordingly, and properly test.

Reset your bios and run it stock, it will probably run better.

 

[Crimson Morrow]

Thanks guys!
I'll dfntly look deeper into what I'm doing and why

 

[Crimson Morrow]

Even thought this was the vid I watched the first time i put my hands on overclocking.

View: https://www.youtube.com/watch?v=X26WU900IAQ&t=2814s

And the other info I'm finding on the internet doesn't really say a lot more than I've already seen there.

 

[Karadjgne]

You really aren't going to learn anything from that video, other than how to get yourself into trouble, easily. The best place I know of in learning how to OC a 10900k is the Asus ROG forums. The actual bios or board or vendor is irrelevant, it's the theory and particulars you are after. Vcore, LLC, vccio, VCCSA, current high/low etc are all the same for Any motherboard or bios or vendor, even if the names are not the same.

An engine is an engine, doesn't matter if it's called a motor, thing under the hood, HP maker, go getter etc.

Learn what the bios settings do, what they don't do, relationships, affects, what helps, what hurts, limits, everything you can. And that applies to any 10th Gen cpu, OC on a msi board with a 10600k is the same as an Asus board with a 10900k, different actual numbers maybe, and different names maybe, but the practice of theory is identical.

 

[Crimson Morrow]

You really aren't going to learn anything from that video, other than how to get yourself into trouble, easily. The best place I know of in learning how to OC a 10900k is the Asus ROG forums. The actual bios or board or vendor is irrelevant, it's the theory and particulars you are after. Vcore, LLC, vccio, VCCSA, current high/low etc are all the same for Any motherboard or bios or vendor, even if the names are not the same.

An engine is an engine, doesn't matter if it's called a motor, thing under the hood, HP maker, go getter etc.

Learn what the bios settings do, what they don't do, relationships, affects, what helps, what hurts, limits, everything you can. And that applies to any 10th Gen cpu, OC on a msi board with a 10600k is the same as an Asus board with a 10900k, different actual numbers maybe, and different names maybe, but the practice of theory is identical.

Thanks!
I'll dfntly go check it out!

 

[Crimson Morrow]

Interestingly enough, when I stress test it with CPU-Z instead of aida64, it works at 5ghz at 100% AND under 80 degrees... hmmmm

 

[Crimson Morrow]

I also been reading this and that's pretty much the same way I got everything set up as well https://siliconlottery.com/products/10900k50g?variant=32927018582102
Everything is almost identical (the voltages that I have a lower) except for CPU Load-line Calibration, instead of level 3, I got level 4, which is a flat line, meaning that the system is going to run more stable in various cases.

 

[Crimson Morrow]

After reading more about all the values, I still have a question about IO and SA.
I'm only running it with 3600mhz RAM, and I've read that that frequency will only need 1.15v at max (I keep mine at 1.2, which as far as I've learnt recommended for 4000mhz). While there's other instances where people say that they keep their IO and SA at 1.2 with only 3200mhz RAM.
Can I please get a comment on that?

P.S.
Btw I tried lowering the vcore voltage down to 1.29v and it actually brought more stability to the system. The fluctuations are far less drastic than with 1.3v.

 

[Karadjgne]

Aida64 natively uses AVX instructions in its test, which can and do equate to @ 130% cpu, much higher power usage than would be considered normal by any means. Prime95 small fft is a good thermal indicator with option to disable AVX technologies and provides a constant, stable 100%, basically worst case scenario for a cpu under Normal circumstances.

There are games out there that do use a small amount of AVX, and a very, very few (there's one mod in Skyrim that I know of) that'll use AVX2, which is even more demanding than AVX. But thats considered part of normal usage because your cpu won't be at above 100% to start with.

So yes, Aida64 and some others will be brutal on a cpu, and the resultant temps, but in no way should be considered 'normal' for practical purposes.

You should define your tests. There's a big difference between a stress test and a temp test, even if those areas overlap. For temp testing, Prime95 small fft, AVX disabled. Some will also use it for testing stability, but on modern cpus, that's not exactly a reliable test. Aida64 or Asus RealBench, Cinebench loop, blender are much better suited to Stability testing as they inherently do not apply a constant use, in any one set of instructions, but rather bombard a cpu with differing levels of use, differing instruction sets, from multiple angles. Which only gives a one-time, spike use temp, so is unreliable for common purposes.

So temp test or stress test or both, but don't rely on a single test to give unequivocal results for both.

 

[Crimson Morrow]

Aida64 natively uses AVX instructions in its test, which can and do equate to @ 130% cpu, much higher power usage than would be considered normal by any means. Prime95 small fft is a good thermal indicator with option to disable AVX technologies and provides a constant, stable 100%, basically worst case scenario for a cpu under Normal circumstances.

There are games out there that do use a small amount of AVX, and a very, very few (there's one mod in Skyrim that I know of) that'll use AVX2, which is even more demanding than AVX. But thats considered part of normal usage because your cpu won't be at above 100% to start with.

So yes, Aida64 and some others will be brutal on a cpu, and the resultant temps, but in no way should be considered 'normal' for practical purposes.

You should define your tests. There's a big difference between a stress test and a temp test, even if those areas overlap. For temp testing, Prime95 small fft, AVX disabled. Some will also use it for testing stability, but on modern cpus, that's not exactly a reliable test. Aida64 or Asus RealBench, Cinebench loop, blender are much better suited to Stability testing as they inherently do not apply a constant use, in any one set of instructions, but rather bombard a cpu with differing levels of use, differing instruction sets, from multiple angles. Which only gives a one-time, spike use temp, so is unreliable for common purposes.

So temp test or stress test or both, but don't rely on a single test to give unequivocal results for both.

Thank you so much, that info actually helped a lot.
I also tried other stress and temp tests and the processor reaches the set value every single time. So yeah, thanks again, have an awesome day