Question OC Strategy for 13th Gen?

[Grimnir]

I've always done manual OC with fixed voltages before. Now I'm thinking it would be nice with a more energy efficient OC, so I'm trying to figure out how adaptive voltages work.

Are there any good guides on this? And, what's your preferred strategy?

One problem I'm having is that the voltage range is huge. At the moment, all core load is around 1.275V, whereas it peaks at 1.465V. I would like to bump the all core voltage, and actually lower the peak, but don't know how to go about it.

(Of course, this is because all core is at 55x multiplier, and I've set a 58x multiplier for fewer cores)

Also, are there any important supporting voltages I should play with, to get a stable OC?

 

[Kocicak]

The easiest way to get rid of the voltage peaks is to prevent the cores to go to 5.8 GHz. The easiest way how to improve efficiency in a heavily multithreaded load is to limit power draw.

 

[Grimnir]

Sure, but I'm aiming for the highest frequency at the lowest possible voltages.

 

[Kocicak]

There is some extra voltage built in, so you can go a little bit higher frequency without changing anything else. You can change maximal frequency of P cores to 5.6 or 5.7 GHz, while limiting power draw to prevent overheating with your cooling solution.

 

[Carfax83]

Well if you're talking about the highest frequencies with the lowest voltages, you will want to do a negative offset voltage. From my experience so far with Raptor Lake, the CPUs respond very favorably to under volting. I currently have mine at -150mv and it's rock solid stable. I tried -165mv but it crashed during idling or low CPU usage scenarios.

If you want to hit high frequencies though, power draw is the most important factor. I have mine limited to 215w and it's capped at 5.2ghz on the P cores and 4.3ghz on the E cores. I game at 4K though so the additional headroom isn't very useful to me and 5.2ghz is already blisteringly fast.

But for you, you're going to want to have the highest power draw that your cooling can comfortably handle, and then tune down the voltages with a negative offset until you're stable. You can also increase the LLC level to limit the voltage droop.

 

[Storm-Chaser]

The easiest way to get rid of the voltage peaks is to prevent the cores to go to 5.8 GHz. The easiest way how to improve efficiency in a heavily multithreaded load is to limit power draw.

Of course, on the flip side of this, your system is going to take a rather substantial performance hit (depends on how much you limit power draw, of course).

Why are your recommending against 5.8 on the best two P core chips? They are rated for that clock speed and he can likely use them at the same speed with less than factory supplied voltage.

 

[Kocicak]

Why are your recommending against 5.8 on the best two P core chips?

Because OP wanted to avoid voltage peaks.

 

[Grimnir]

There is some extra voltage built in, so you can go a little bit higher frequency without changing anything else. You can change maximal frequency of P cores to 5.6 or 5.7 GHz, while limiting power draw to prevent overheating with your cooling solution.

I understand that lowering the frequencies will result in lower voltages. But again, I'm after maximum performance. Controlling voltages by lowering frequencies is clearly not the way.

Well if you're talking about the highest frequencies with the lowest voltages, you will want to do a negative offset voltage. From my experience so far with Raptor Lake, the CPUs respond very favorably to under volting. I currently have mine at -150mv and it's rock solid stable. I tried -165mv but it crashed during idling or low CPU usage scenarios.

If you want to hit high frequencies though, power draw is the most important factor. I have mine limited to 215w and it's capped at 5.2ghz on the P cores and 4.3ghz on the E cores. I game at 4K though so the additional headroom isn't very useful to me and 5.2ghz is already blisteringly fast.

But for you, you're going to want to have the highest power draw that your cooling can comfortably handle, and then tune down the voltages with a negative offset until you're stable. You can also increase the LLC level to limit the voltage droop.

Would such an offset be added to the V/F curve?

 

[Kocicak]

... lowering frequencies ...

P cores are limited by 5.5 GHz, I wrote about increasing that limit. I believe that overclocking those two selected cores does not make much sense, because most of the time when you do something practical with a little higher load, all the P cores fall back to 5.5 GHz.

 

[Grimnir]

P cores are limited by 5.5 GHz, I wrote about increasing that limit. I believe that overclocking those two selected cores does not make much sense, because most of the time when you do something practical with a little higher load, all the P cores fall back to 5.5 GHz.

No, it does make a difference. Many loads are lightly threaded. Here is just one example.

 

[Kocicak]

I am not confident recommending running these CPUs at high voltage to reach high frequency anymore, I will let somebody else take this over.

High frequency needs high voltage, you need to make up your mind, if you want to go the risky route to ultimate performance, or a more reasonable way, remembering that these CPUs offer almost all the performance they can achieve out of the box already, they need to be even tamed to some reasonable power draw. Once they are, these CPUs are actually quite easy to cool.

 

[coercitiv]

One problem I'm having is that the voltage range is huge. At the moment, all core load is around 1.275V, whereas it peaks at 1.465V. I would like to bump the all core voltage, and actually lower the peak, but don't know how to go about it.

(Of course, this is because all core is at 55x multiplier, and I've set a 58x multiplier for fewer cores)

SkatterBencher has done the for us, extracting an entire manual on Alder Lake overclocking, which obviously still applies to Raptor. The great thing about adaptive voltage ever since Comet Lake is the Advanced Voltage Offset. This allows you to set a custom offset for the most important area of the V/f curve, including having stock voltage for low frequencies and having a custom voltage for the OC ratio.

In practice you would want to test for the lowest stable voltage for your all-core multiplier (e.g. 55x) and the lowest stable voltage for light threaded multi (e.g. 58x). You can also test for some values bellow 55x, but at some point around 4Ghz you'll just want to leave the values on Auto since the gains will be minimal (relative to your cooling) and you want stock voltage as you go down in clocks anyway. This way you solve the biggest problem with the static voltage offset, which is instability at low frequencies where undervolting threshold is limited. Keep in mind the number of V/f points is limited, but the algorithm uses interpolation to compensate.

Once you have an optimized and stable V/f curve for stock clocks, it's time to reap the benefits by either using less power or increasing clocks with a power/temperature limit you're comfortable with.

 

[Carfax83]

Would such an offset be added to the V/F curve?

Yes, using an offset, whether negative or positive applies to the entire V/F curve. This can make using a negative offset problematic because your rig can become unstable when idling as the voltage drops too low. I encountered this problem myself when I set the negative offset to -165mv, and I crashed a few times when doing basic low power stuff. At -150mv it's been rock solid stable though.

You can get around the instability at lower voltages by using higher LLC settings, or by just increasing the voltage slightly more like what I did.

 

[Grimnir]

I am not confident recommending running these CPUs at high voltage to reach high frequency anymore, I will let somebody else take this over.

High frequency needs high voltage, you need to make up your mind, if you want to go the risky route to ultimate performance, or a more reasonable way, remembering that these CPUs offer almost all the performance they can achieve out of the box already, they need to be even tamed to some reasonable power draw. Once they are, these CPUs are actually quite easy to cool.

You're completely missing the point here. I want to limit the voltages while pushing for the highest possible frequencies. Limiting the frequencies to drive down voltages is not the solution. Controlling the voltages is (and last time I looked, the frequency multiplier is not a voltage control).

 

[Kocicak]

... I encountered this problem myself when I set the negative offset to -165mv, and I crashed a few times when doing basic low power stuff. At -150mv it's been rock solid stable though.
....

You are JUST 15mV away from instability?! Wow, that seems careless. I would keep at least 50 mV buffer.

I hope you meant mV (millivolts), and not mv (what is it?). Distinguishing lower and upper case letters is important. For example, 1MV = 1000000000 mV.

Volts are V. ALWAYS. NEVER v !!!

 

[Storm-Chaser]

You are JUST 15mV away from instability?! Wow, that seems careless. I would keep at least 50 mV buffer.

I hope you meant mV (millivolts), and not mv (what is it?). Distinguishing lower and upper case letters is important. For example, 1MV = 1000000000 mV.

Volts are V. ALWAYS. NEVER v !!!

This guy keeps recommending lower voltages because thinks the CPU is going to degrade, and he has his mind made up, without actually doing the legwork to prove it. He ran three cores @ 1.45v @ like 6.0GHz and now claims his CPU is degraded. So he is pushing it on other people now. The claims are baseless.

Bit ironic that you are posting on an overclocking forum and this guy is telling you to do the opposite.

 

[Kocicak]

This guy keeps recommending lower voltages because thinks the CPU is going to degrade, and he has his mind made up.

Oh yeah. I am not putting more than 1.35V in one of these. I also decided not to run them hotter than 80°C.

 

[Gregow]

This guy keeps recommending lower voltages because thinks the CPU is going to degrade, and he has his mind made up, without actually doing the legwork to prove it. He ran three cores @ 1.45v @ like 6.0GHz and now claims his CPU is degraded. So he is pushing it on other people now. The claims are baseless.

Bit ironic that you are posting on an overclocking forum and this guy is telling you to do the opposite.

Heh, on my previous CPU (8086K) I accidentally pushed over 1.9V.
It might very well have degraded because of this, but it’s been rock solid at 5Ghz for several years. It’s now going to serve as my server CPU.

As far as I know, it’s safe to push over 1.5V on these newer CPUs. According to Intel’s spec. Now, I don’t want to push that high because of temperatures and possible issues over time, but… well, we’re talking years - not short durations.

 

[Kocicak]

As far as I know, it’s safe to push over 1.5V on these newer CPUs. According to Intel’s spec. Now, I don’t want to push that high because of temperatures and possible issues over time, but… well, we’re talking years - not short durations.

I am sorry not to have some throw-away money to be able to really test this, because I believe I could KILL a few CPUs this way in minutes, not only slightly degrade, as I already managed with one previous CPU at lower voltage.

 

[Grimnir]

I am sorry not to have some throw-away money to be able to really test this, because I believe I could KILL a few CPUs this way in minutes, not only slightly degrade, as I already managed with one previous CPU at lower voltage.

Dude, please. You clearly have no idea what you're talking about. Others, meanwhile, have provided some very solid answers (a bit heavy on the reading though, but still).

I've been overclocking for well over 20 years. While I am unfamiliar with some of the latest strategies, don't take me for a -redacted- fool.

 

[Kocicak]

Dude, please. You clearly have no idea what you're talking about. Others, meanwhile, have provided some very solid answers (a bit heavy on the reading though, but still).

I've been overclocking for well over 20 years. While I am unfamiliar with some of the latest strategies, don't take me for a fucking fool.

Dude, the silicone from today and from 20 years ago is a completely different thing. Moreover, Intel 10 nm process may be less than ideal. Test it yourself.

(backing into bushes)

 

[Grimnir]

Dude, the silicone from today and from 20 years ago is a completely different thing. Moreover, Intel 10 nm process may be less than ideal. Test it yourself.

(backing into bushes)

Please stay in the bushes. I've already got some solid responses here (thank you guys!). What I don't care for are "advice" trying to bring me down to an idiot level. Honestly, that's -redacted- insulting.

Profanity is not allowed in the tech forums.

Daveybrat
AT Moderator

 

[Storm-Chaser]

Dude, the silicone from today and from 20 years ago is a completely different thing. Moreover, Intel 10 nm process may be less than ideal. Test it yourself.

(backing into bushes)

Oh yeah. I am not putting more than 1.35V in one of these. I also decided not to run them hotter than 80°C.

More power to you. Just don't go around telling everyone else to kill their processors performance in the interim. The OP is here to overclock not undervolt and underclock. You really need to back off and go back to the basics of overclocking, because I think you are having an existential crisis right now. If you can't figure out what you are doing wrong, I recommend you start your own underclock/undervolting forum so you can shout your misleading data from the mountain tops.

Please stay in the bushes. I've already got some solid responses here (thank you guys!). What I don't care for are "advice" trying to bring me down to an idiot level. Honestly, that's -redacted- insulting.

He is like a yoyo. Keeps coming back for more. I would actually give him some credit if he did the leg work, but he only has a sample size of ONE. Thats right, ONE chip. Not even worth it.

Don't worry about it. So long as cooling is good you should be able to get 5.6GHz all core. And I don't own a 13th gen, but another way to go about it is PER core (if this feature is available). Set your best two cores to 6.0GHz at 1.430 volts and set the remaining cores at whatever speed you discover that's reliable. Probably 5.5 to 5.6GHz. Keep your LLC up a couple notches to prevent voltage dips to the processor. Can you set VID for every core?

Another option would be turbo core. Gigabyte has just recently released a BIOS that includes an instant 6.0GHz overclock for 13th gen (just fyi). So you could on your own for a turbo ratio of 2x 6.0Ghz, 2x 5.7GHz and 4x at 5.6GHz or some combination thereof. I actually prefer turbo core overclocking in some instances. You can also experiment with voltages on your turbo mode and see what's optimal.