Offset won't go under 1.296-1.315V AsRock Z370 Taichi

[Shou Miko]

I was wondering at what voltage my i7-8086K would do 5GHz on all cores bit with a stable offset of -50 I still see CPU-Z reading os 1.296-1.315V and a little drop to 1.264V.

I was wondering do I really need a fixed voltage to try to go down to 1.225-1.250V to test if my i7-8086K can do 5GHz at that voltage?

I am not complaning about my Cinebench R20 core is around 3794pts but I want the voltage down to reduce heat on my 240mm AIO Fractal Design Celeius S24.

I am running bios P3.20 modded by @R-T-B waiting on a newer release by him since I do not use Intel MEI.

 

[Vario]

I have the same board as you and I think there are a good bit of bugs with it that do not effect the stability of the board but do impair some of the overclock fine tuning. My Z370 Taichi acts strange with offset voltage too. I don't remember the specifics but I seem to remember it not moving voltage in a granular way with offset, it was harder to dial in as a result because it wouldn't change even with +/- 100 according to my sensor software. I also can't overclock my uncore beyond 4300, it simply doesn't change past that point. I can change my uncore all I want in Windows using XTU and go quite high with it, like 48X no problems, but not in bios, not past 4300, any additional uncore multiplier does nothing even if my CPU multiplier is above that amount.

Overall I can't complain, its a stable system that has never blue screened on me and my 8600K sits at 4.6 all cores no avx offset, 46x, and voltage sits at 1.15 in most games, 1.18V peak AVX. It overclocks ram really well too. Everything runs really fast, responsive, stable, and cool.

Also my onboard audio has a ground loop that picks up the mouse movement, but I use a USB audio card instead.

Edit: Also my vdroop could be better, even if i run a high LLC setting it droops a good bit, almost as though the LLC doesn't exist.

These issues don't really matter to me, I like having a system that never crashes ever. If I was a hardcore overclocker, however, I'd be frustrated.

 

[Shou Miko]

I use a Creative Sound Blaster Z with my Audio-Technica A550Z if I use onboard all audio even with mods sounds horrible because the ALC1220 cannot drive these headphones.

Hmm I wanna play around with this more because I do now want to run like 1.225-1.250V on my CPU 24/7 I don't think that's good for it.


Looks like -80 gives me around 1.264V so maybe going to -100 will make it around my target of max 1.250V without getting unstable.

But I am glad to hear @Vario I am not the only one thinking this is weird.

Just made a cinebench r20 run with a offset of -110 that gets CPU-Z to 1.248V under load with no crash at 5GHz on all cores if I do -120 in offset the CPU-Z just freezed out during Windows 10 so I guess I hit the end of the line unless I wanna go under 5GHz.

 

[Vario]

I use a Creative Sound Blaster Z with my Audio-Technica A550Z if I use onboard all audio even with mods sounds horrible because the ALC1220 cannot drive these headphones.

Hmm I wanna play around with this more because I do now want to run like 1.225-1.250V on my CPU 24/7 I don't think that's good for it.


Looks like -80 gives me around 1.264V so maybe going to -100 will make it around my target of max 1.250V without getting unstable.

But I am glad to hear @Vario I am not the only one thinking this is weird.

Just made a cinebench r20 run with a offset of -110 that gets CPU-Z to 1.248V under load with no crash at 5GHz on all cores if I do -120 in offset the CPU-Z just freezed out during Windows 10 so I guess I hit the end of the line unless I wanna go under 5GHz.

Guess so! I think 1.25V is perfectly safe, if its anything like past generations, you need to worry more about temperatures. Also check your VCCIO and VCCSA to make sure they are not overvolted. 1.05 V should be adequate for these, the specification is around 1.0 V for them. Sometimes these things get overvolted to 1.1 V and beyond.

 

[Shou Miko]

VCCIO is set up auto and shows a reading of 0.950V and VCCSA is also set to auto and shows a reading of 1.05V.

I guess I am going here and no need to set a value.

 

[Pantidria]

Have a look in Intel's datasheet.
VCCIO: 1.2V +/- 5% (1.14V-1.26V).
VCCSA: 1.05V.
On "Auto" some motherboards will pump too much voltage into the CPU. To protect your CPU from your motherboard set the Long Duration Power Limit and the Short Duration Power Limit to 95 or less.

 

[Shou Miko]

@Pantidria when readings show what they show at auto why change them¿

 

[Pantidria]

With HWMonitor you can read out the System Agent Voltage and the IO Voltage of your system. On "Auto settings" they go up to over 1.3V with my motherboard.

 

[newtekie1]

Guess so! I think 1.25V is perfectly safe, if its anything like past generations, you need to worry more about temperatures.

Yeah, I've always been under the impression that anything under 1.3v is safe, as long as you've got the cooling to handle it.

 

[Vario]

With HWMonitor you can read out the System Agent Voltage and the IO Voltage of your system. On "Auto settings" they go up to over 1.3V with my motherboard.

1.3 for those two is too high. I assume you are not running that Auto setting.

You should be able to run at the Intel specifications or slightly above. I run mine at VCCST 1.05, VCCIO 1.0. I might also be able to lower my IO to the Intel .95. That is with 3200C14 B-Die on 1.35VDRAM.

 

[Vayra86]

I have the same board as you and I think there are a good bit of bugs with it that do not effect the stability of the board but do impair some of the overclock fine tuning. My Z370 Taichi acts strange with offset voltage too. I don't remember the specifics but I seem to remember it not moving voltage in a granular way with offset, it was harder to dial in as a result because it wouldn't change even with +/- 100 according to my sensor software. I also can't overclock my uncore beyond 4300, it simply doesn't change past that point. I can change my uncore all I want in Windows using XTU and go quite high with it, like 48X no problems, but not in bios, not past 4300, any additional uncore multiplier does nothing even if my CPU multiplier is above that amount.

Overall I can't complain, its a stable system that has never blue screened on me and my 8600K sits at 4.6 all cores no avx offset, 46x, and voltage sits at 1.15 in most games, 1.18V peak AVX. It overclocks ram really well too. Everything runs really fast, responsive, stable, and cool.

Also my onboard audio has a ground loop that picks up the mouse movement, but I use a USB audio card instead.

Edit: Also my vdroop could be better, even if i run a high LLC setting it droops a good bit, almost as though the LLC doesn't exist.

These issues don't really matter to me, I like having a system that never crashes ever. If I was a hardcore overclocker, however, I'd be frustrated.

I recognize this behaviour using the vcore Offset as well on my AsRock Fatality Z370. Which is practically the same board.

What you will find, probably, is that the offset vcore is governed by a third parameter: the actual frequency the CPU runs at. An offset still follows the voltage curve per multiplier as set by Intel. If you knock 200 mhz off and keep everything else the same, you will likely monitor a lower vcore as well.

At least, that is my take on this behaviour and I can reproduce it like that. Another big factor is of course the LLC setting wrt your load voltage. I am getting best results with looser setting (level 4/5), and it also keeps me away from (exceeding) the -100 offset, because high LLC will force you to lower the actual vcore/offset value.

Still is preferable to the fixed voltage though. There is quite a voltage reduction to find between tweaking the LLC level, offset value and the turbo clocks itself. I ended up dropping 100mhz just so I could end up a good bit lower on vcore and hit my sweet spot in heat/perf. If I wanted 4.8 Ghz all core, I'd be pushing 1.28-1.3v instead of 1.26v now (with level 5 LLC!), which makes my OC 'summerproof' as well My peak AVX load temp equilibrium is 83-85C, in a very warm room.

VCCIO/SA, yes they should be FAR lower than 1.3V. 1.2V and downwards, with 1.2V-1.25V being for serious memory OC's. You can usually drop SA a tad lower than IO. I run XMP 3200C16 2x8GB on 1.1V. Any lower and the system would run 'stable' (no BSODs or visible errors) but then you look at event viewer and notice that something is constantly throwing errors at you. So do check that too if you go low on VCCIO and SA. A CPU generating errors (OCCT is also very good at showing that in testing) will hurt performance in the end.

 

[R-T-B]

I am running bios P3.20 modded by @R-T-B waiting on a newer release by him since I do not use Intel MEI.

Thanks for the reminder, btw. Will get working on a beta build for z370/z390 taichis today...

 

[Shou Miko]

1.3 for those two is too high. I assume you are not running that Auto setting.

You should be able to run at the Intel specifications or slightly above. I run mine at VCCST 1.05, VCCIO 1.0. I might also be able to lower my IO to the Intel .95. That is with 3200C14 B-Die on 1.35VDRAM.

As I said AUTO is fine HWMonitor reports VCCST 1.032-1.040V and VCCIO 0.960-0.968V.

I don't expect to change anything this is during Cinebench R20.