I9 9900k

[Distracted]

Wondering if I set 1.340v in bios but highest I ever see is 1.312. If I could set it higher, say 1.4v it would only show about 1.350v in windows. Wonder if that would be the way to go to try and improve OC.

People have run older CPUs at higher vcores than that without issue 24/7. The voltage headroom is yet to be proven, but since it is supposed to have increased with coffee lake it should be safer than in the past.

 

[0121danwilliams84]

I followed der8auers overclocking guide. I increased all the power limits he showed. I am currently on LLC 5. 1.340v 4.9GHz. I don't think its 100% stable either

 

[Distracted]

Der8auer uses llc 6 or 7

I was using 7. Theres a lot of droop, so under load the cpu might be underpowered. Try raising it.

 

[moorhen2]

I followed der8auers overclocking guide. I increased all the power limits he showed. I am currently on LLC 5. 1.340v 4.9GHz. I don't think its 100% stable either

LLC5 is too low, more vdroop, up it to 6 and test for vdroop again, keep testing until you get the voltage under load that will run a given multiplier.

 

[Distracted]

Don't forget that a z370 is going to potentially have more power issues than z390. Like i said before, testing from SL showed that Z370 boards needed more voltage compared to z390 to hit the same clocks.

But whack that llc up a couple of notches.

 

[Sc00p007]

I think we all started our testing with x43 on the cache. I know people are beginning to increase that now, but if you want to compare your cpu to the rest you are seeing then leave the cache at 43 and see how you get on.

As you've already seen with the change from 45 to 43 it makes a difference to the voltage you need.

If you are worried about losing performance don't be. I tested with cache at 43 and then at 50 and there was no difference in cinebench score at all so i dont think you lose any noticable real world performance.

I use a Gigabyte board and cache was 47x as default in bios f4 and f5. Installed f7a and its been reduced to 43.

 

[moorhen2]

Core at 5.3ghz, 1.350v under load, take FPU/AVX out of the equation and it needs less voltage and temps are manageable.

 

[Distracted]

I use a Gigabyte board and cache was 47x as default in bios f4 and f5. Installed f7a and its been reduced to 43.

The higher cache on auto for gigabyte boards was limiting the oc people were getting by 100mhz or so compared to asus boards from what i have read.

By reducing the cache to 43 they match the asus settings and it should allow higher stable core clocks for the same voltage.

 

[0121danwilliams84]

LLC5 is too low, more vdroop, up it to 6 and test for vdroop again, keep testing until you get the voltage under load that will run a given multiplier.

Cheers guys. I am "almost" stable. I can game for hours and then I get a WHEA error in HW info. I put my SA IO ram voltages back to auto to rule that out. and Increased ram voltage to 1.360 as it was on 1.344v on XMP. I will up the LLC to 6 and tweak voltage more I guess and report on stability!

 

[Distracted]

Been doing more testing and it seems i have a pretty good chip...
so for 5.2 AVX i only need 1.28v to be Realbench stable

Spoiler

for 5.3 AVX @ 1.3v it passes cinebench and 3dmark but not realbench will probably need more volts or maybe reduce AVX by 1-2 but temps are already a problem

Spoiler

https://www.3dmark.com/spy/5098074
https://www.3dmark.com/fs/17110802

i tried 5.4 @ 1.32v and it passed 3d mark but not CB wont even attempt RB

https://www.3dmark.com/fs/17110927

i reckon i can get it CB stable at 5.4 i guess that kind of clock speed needs more tweaking than voltage and LLC only
maybe in afew days i will try it


now for my 24/7 setting things are pretty damn good
5Ghz prime and realbench stable (overnight runs)
when its not using AVX it needs 1.244v when i run AVX load volts jump up abit to 1.269 all rock stable so far
not tweaked any other settings just set adaptive voltage to 1.24v in the bios with LL6

Spoiler: RB

Spoiler: Prime

That does look like good silicon!

Did you say you are using Adaptive voltage instead of manual?

 

[moorhen2]

Looks like he is using adaptive voltage and speedstep enabled.

 

[Distracted]

Just finished 30 mins stress test with realbench.
No WHEA, all results match no bsod etc.



Now running 2 hrs.

Voltage in cpuz is reported at 1.008 at the moment while testing. Hwinfo showing 1.119.

5GHz no AVX offset cache 43, ram 3200 cl14.

Cpu hottest core 80 degrees during spike, averaging high 60s low 70s across all cores.
Vrms 42 degrees

1.17v or 1.18 v in Bios i think, llc7


Still on a single 40mm 360 rad with a 1080ti and vrms in the loop.

Probably my last test before i pull it all apart since I think I've got everything i need to finish the loop and add the mo-ra now.

Edit - 45 mins in and i've started getting cache errors so I'll need to up the voltage.
No errors on the shorter run. Possibly due to temps or just showing instability because of the low voltage and sustained load?

Edit - checked vcore

 

[fornowagain]

You might want to try adaptive. Once you're reasonably sure of the Vcore needed. Seems to hold nearer to the set turbo Vcore for the same LLC. Less droop. On mine I had to use a negative offset based on the full load VID.

 

[Distracted]

I'm going to stop and tear down the PC now.

When would you guys consider a system stable? I saw above that Sem said he was realbench stable after 15min testing. I know moorhen2 is using Aida.

Do you have a standard set of timed tests you go through before you stop and say you've done enough?

 

[Distracted]

You might want to try adaptive. Once you're reasonably sure of the Vcore needed. Seems to hold nearer to the set turbo Vcore for the same LLC. Less droop. On mine I had to use a negative offset based on the full load VID.

I'll definitely look into it, thank you.

 

[moorhen2]

I'm going to stop and tear down the PC now.

When would you guys consider a system stable? I saw above that Sem said he was realbench stable after 15min testing. I know moorhen2 is using Aida.

Do you have a standard set of timed tests you go through before you stop and say you've done enough?

If your pc does what ever you tend to do, either heavy duty gaming or whatever your day to day computing consist of, and it doesn't crash, its stable for that use I think. But stability is a personal thing, what one person believes is stable, another will say its not, so its subjective. JMO

 

[Sc00p007]

This avx thing has made all this harder to analyse.
Now there are multiple lev

I'm going to stop and tear down the PC now.

When would you guys consider a system stable? I saw above that Sem said he was realbench stable after 15min testing. I know moorhen2 is using Aida.

Do you have a standard set of timed tests you go through before you stop and say you've done enough?

This is why it's getting harder to analyse overclocks.
At the moment i feel i can't claim a stable oc at 5.2 because i know there is something out there that will crash my computer that wont at stock.
Could this cause my computer to crash if something i use in the future induces the similar stress.
Ive had aida64 24hr tests prove far from stable over time. Ive had realbench 15-30 min stables to prove problematic also.
Sometimes it can be a good few months of stable gaming, followed by scratching head when i crash a couple of times. Because i thought i was stable, it was lower on my troubleshooting list.
Now you have reports of 1.2v vs 1.35 for the same oc. But the reality is they are much closer, its just the stretching of the word stable.
This is why i tried prime, then reported my findings. To balance. Yes i can do 5.2 at good v and temps with aida and realbench. However instacrash at same settings for prime small. So i know i am not stable as prime doesnt crash my comp at stock. I should report i have a system that allows me to play games without crashing this week. But it is not prime stable. This means i would not be surprised if i crashed once or twice every couple of months.
I wonder how well it would go down, if intel sold stable chips that instacrashed on prime at stock...

 

[Distracted]

This avx thing has made all this harder to analyse.
Now there are multiple lev


This is why it's getting harder to analyse overclocks.
At the moment i feel i can't claim a stable oc at 5.2 because i know there is something out there that will crash my computer that wont at stock.
Could this cause my computer to crash if something i use in the future induces the similar stress.
Ive had aida64 24hr tests prove far from stable over time. Ive had realbench 15-30 min stables to prove problematic also.
Sometimes it can be a good few months of stable gaming, followed by scratching head when i crash a couple of times. Because i thought i was stable, it was lower on my troubleshooting list.
Now you have reports of 1.2v vs 1.35 for the same oc. But the reality is they are much closer, its just the stretching of the word stable.
This is why i tried prime, then reported my findings. To balance. Yes i can do 5.2 at good v and temps with aida and realbench. However instacrash at same settings for prime small. So i know i am not stable as prime doesnt crash my comp at stock. I should report i have a system that allows me to play games without crashing this week. But it is not prime stable. This means i would not be surprised if i crashed once or twice every couple of months.
I wonder how well it would go down, if intel sold stable chips that instacrashed on prime at stock...

There certainly is a lot of variation out there.

I'm not knockung anyone's methods. If someone says they are 15min realbench stable then theres no ambiguity there and you can do what you want with the info. Likewise if someone says they are 24hrs stable in the latest prime with avx. It is all useful so long as enough info is supplied.

The problem i have is knowing what benchmarks will give me the best indication that my system is 24/7 stable for my intended use. Hence me asking others at what point they stop testing and call an oc stable, and what testing they use to get there.

 

[Sc00p007]

There certainly is a lot of variation out there.

I'm not knockung anyone's methods. If someone says they are 15min realbench stable then theres no ambiguity there and you can do what you want with the info. Likewise if someone says they are 24hrs stable in the latest prime with avx. It is all useful so long as enough info is supplied.

The problem i have is knowing what benchmarks will give me the best indication that my system is 24/7 stable for my intended use. Hence me asking others at what point they stop testing and call an oc stable, and what testing they use to get there.

It is difficult to know.... I mainly game, have been realbench and aida64 stable, but definitely had crashes because of instability. 15 mins realbench stable to me is far from a confirmation of stability. But I know from experience. Many who are new to this may think it is a lot closer to stable "via the terming" than the reality.

edit* Maybe it would be useful to post 2 settings for a new chip. One for your aida's, and one for your primes. That way you give a better idea to far more people as to what a chip is capable of. The guy who is using 1.35 for their chip won't feel they lost the lottery when reading all kinds of voltages ranging from under 1.1

 

[moorhen2]

The thing is can we ever say our OC is fully stable, there are many things that can come into force that will cause your system to GSOD, its not just an unstable OC, drivers can be a major cause of crashes, W10 is quite a stable OS, but it only takes something to happen it doesn't like and it will crash, and its not always the hardware that causes it.