Those on Raptor Lake... What is your Vid cap? (Intel's default - or something else? - not updating at all??)

[RootinTootinPootin]

its a bit hot tonight here, but its doable on both my chips. (44k) what else could you suspect me? lol

 

[ShrimpBrime]

as I have said, I already tested the other way around, it just makes things worse for me, even auto sends it to 1.4v even my VID is still low. There's also no mention of such rule that the VID should be equal to vcore. never heard of that mention.


wdym?

The Ecores are running 4.7ghz with 5.8ghz/6.0ghz. The CB23 score should be closer to 50k.

I mean I don't have a 13900K/S (yet), but I'm close to 40K with similar clocks 14700K only 28 threads.

I'm saying your score seems low and possible cause could be effective clocks are drooping. Benchmate will show the effective clocks during the benchmark. I am only curious.

 

[chrcoluk]

Yeah, I can turn CEP off, but I still have undervolt protection on the new ucode or I could select the old ucode and it would be the exact opposite, undervolt protection gets turned off and CEP is on ( Well I assume - I don't know if it was cep I just know it was acting like CEP). Besides thats what my 90 ac loadline setting basically is, a small undervolt, which would be considered on the VID level wouldn't it? However like I said before I can't go much further without losing performance ( undervolt protection?).

I think to get it so they are both off I would have to at least go back a couple bioses. And I feel like its just not worth my time. The last time I had it working was the bios right before they first added the intel profiles, that I remember. But my undervolt got unstable and my machine was acting REALLY weird. Lots of things were happening, for example, at one point it looked like somebody was doodling all over my windows log in screen with paint. Thats when I said you know what maybe I should update my bios and ucode LOL.


I wasn't trying to say that what you were doing it wrong. I just meant that specific feature of the update wouldn't have much effect on you, thats all. I'm not telling you how to live your life. You do you.

There is a UV protection for security purposes, but that only blocks post boot undervolting, so it will prevent software like XTU from undervolting, but should still allow bios undervolting. But anyway given your past not great experience with UV, maybe its just the lowad line adjustment then combined with maybe a clock speed cap to get your VID down if that is bothering you.

 

[RootinTootinPootin]

The Ecores are running 4.7ghz with 5.8ghz/6.0ghz. The CB23 score should be closer to 50k.

I mean I don't have a 13900K/S (yet), but I'm close to 40K with similar clocks 14700K only 28 threads.

I'm saying your score seems low and possible cause could be effective clocks are drooping. Benchmate will show the effective clocks during the benchmark. I am only curious.

my 14700k does 38k, 50k seems too farfetched from what I see on OCN my scores are pretty normal.

 

[iameatingjam]

There is a UV protection for security purposes, but that only blocks post boot undervolting, so it will prevent software like XTU from undervolting, but should still allow bios undervolting. But anyway given your past not great experience with UV, maybe its just the lowad line adjustment then combined with maybe a clock speed cap to get your VID down if that is bothering you.

Really... cause gigabyte did a whole press release about how b boards and non-k cpus can undervolt now since you can turn CEP off (since ucode 123 I think) and linked to it in my very bios update. I got super excited and the first thing I tried was an offset... didn't work. Like it didn't even apply, didn't take. And in gigabyte's example they use AC loadline for the undervolting so I assumed thats how I was supposed to do it. But in gigabyte's example they did ac all the way down to 40. I get crippled even at 60. So yeah.... I dunno whats up with that. Maybe its just a weak cpu... but you'd think that would cause a crash not poor performance. ( I assumed that it was undervolt protection but gigabyte both claimed it worked and now you claiming UV protection doesn't apply to the bios, makes me unsure)

 

[ShrimpBrime]

my 14700k does 38k, 50k seems too farfetched from what I see on OCN my scores are pretty normal.View attachment 359798

It's not what I was asking for lol.

Here's an example. It's only 16 threads, but my point is observing the actual effective clocks which are recorded by hwinfo64 through benchmate.

I noticed when effective clocks droop, so does the actual performance. If I'm making sense....

So here's an example. Note max core and max effective frequencies.

 

[Zach_01]

I think that if Intel wanted for this issue to go away once and for all, had to do 3 things

1. Loose 100-200MHz on single boost permanently.
2. Limit default max power limit way under 253W like 200-220W
3. Make TVB actually work and allow max speeds only under specific thermal and current(A) conditions.

I guess it all depends if you believe that all this was some kind of misconfiguration on Intel part or just that they pushed the silicon over its safe operating limits for making the products competitive and hope for the best.

Obviously I’m on the latter.
I understand that Intel can’t go and just chop off speeds and power all over the place. It will cause huge reactions and even legal moves from customers and not talking only about the individual end users.

IMO they are trying now with semi-measures to extend maybe a little more the longevity of CPUs and try to keep boost and performance the same.

Have the cake and eat it too

I apologize for the intrusion but had to say it.

Here's an example. It's only 16 threads, but my point is observing the actual effective clocks which are recorded by hwinfo64 through benchmate.

I noticed when effective clocks droop, so does the actual performance. If I'm making sense....

Perfect sense to me.
The performance of these highly dynamic chips (both Intel/AMD) is directly related with effective clock that contain all CPU states and not just the active one.

 

[sn2x]

Never changed VID cap, I'm on a 13600K, under current BIOS the max voltage I ever saw was 1.349, which I'm pretty sure is safe.

Currently overclocked also.

 

[ShrimpBrime]

Never changed VID cap, I'm on a 13600K, under current BIOS the max voltage I ever saw was 1.349, which I'm pretty sure is safe.

Currently overclocked also.

The 13600KF I got about release time paired with an Asus B660-G on bios version 2012 was able to BCLK. I was squeezing 110mhz from that bad boy. I made the mistake of updating ME and Intel made sure no more BCLK. And they call these unlocked chips. Pfft.

 

[Arctucas]

I just set manual voltage in BIOS with negative offset.

5700 MHz, Vcore = VID = 1.265V +/- .007V loaded.

 

[chrcoluk]

Really... cause gigabyte did a whole press release about how b boards and non-k cpus can undervolt now since you can turn CEP off (since ucode 123 I think) and linked to it in my very bios update. I got super excited and the first thing I tried was an offset... didn't work. Like it didn't even apply, didn't take. And in gigabyte's example they use AC loadline for the undervolting so I assumed thats how I was supposed to do it. But in gigabyte's example they did ac all the way down to 40. I get crippled even at 60. So yeah.... I dunno whats up with that. Maybe its just a weak cpu... but you'd think that would cause a crash not poor performance.

Well you mentioned an additional UV blocker on top of CEP, thats why I mentioned that.

I think CEP and the UV security feature are two separate things.

As I mentioned before there is two offset's, one for the VRM and one for the CPU VID request. Buildzoid actually did a very recent video on this, and since he has a gigabyte motherboard, it might be an idea to check as he specifically shows the setting to use on gigabyte board's to alter the VID offset. You need to alter the VID offset to undervolt with CEP enabled. Undervolting with AC/DC is limited in its capability and seems to mostly impact multi core load droop levels.

If CEP is disabled, then it shouldnt matter which offset you change, however BZ in a video I also recall mentioning he thought disabled setting wasnt working for the CEP on his gigabyte board. (meaning was on all the time).

 

[Zach_01]

The performance of these highly dynamic chips (both Intel/AMD) is directly related with effective clock that contain all CPU states and not just the active one.

BTW

FEATURE - Effective clock vs instant (discrete) clock

It has become a common practice for several years to report instant (discrete) clock values for CPUs. This method is based on knowledge of the actual bus clock (BCLK) and sampling of core ratios at specific time points. The resulting clock is then a simple result of ratio * BCLK. Such approach...

www.hwinfo.com

 

[ShrimpBrime]

BTW

FEATURE - Effective clock vs instant (discrete) clock

It has become a common practice for several years to report instant (discrete) clock values for CPUs. This method is based on knowledge of the actual bus clock (BCLK) and sampling of core ratios at specific time points. The resulting clock is then a simple result of ratio * BCLK. Such approach...

www.hwinfo.com

I love when I mess around and don't set up my daily OC.

Power up to the number 36 everywhere. lol. I think I was aiming for 125w. And that's a HUGE difference from 253w haha.

 

[RootinTootinPootin]

I noticed when effective clocks droop, so does the actual performance. If I'm making sense....

you are correct on this observation, that is pretty normal natural (happens on AMD as well, clock stretching), that's why you disable C-States when benching, IDRGAF nowadays much on E-peen points, but I still post a few from time to time. I just try to enjoy these beasts as that's what I bought them for.

Hope nobody gets offended.

 

[iameatingjam]

Well you mentioned an additional UV blocker on top of CEP, thats why I mentioned that.

I think CEP and the UV security feature are two separate things.

As I mentioned before there is two offset's, one for the VRM and one for the CPU VID request. Buildzoid actually did a very recent video on this, and since he has a gigabyte motherboard, it might be an idea to check as he specifically shows the setting to use on gigabyte board's to alter the VID offset. You need to alter the VID offset to undervolt with CEP enabled. Undervolting with AC/DC is limited in its capability and seems to mostly impact multi core load droop levels.

If CEP is disabled, then it shouldnt matter which offset you change, however BZ in a video I also recall mentioning he thought disabled setting wasnt working for the CEP on his gigabyte board. (meaning was on all the time).

Yeah they definitely are different things. Intel sure put a lot of work into making sure people don't have a way to fix their mistakes. Anyway I'll check out the video for sure just maybe later I'm kinda sick in bed right now. In my experience when it comes to offsets, UV protection doesn't allow them, CEP allows them, it just cripples them. Thats how I thought anyway, I was able to tell the difference. Anyway have a good one, man do I got such bad headache right now.

Well you mentioned an additional UV blocker on top of CEP, thats why I mentioned that.

I think CEP and the UV security feature are two separate things.

As I mentioned before there is two offset's, one for the VRM and one for the CPU VID request. Buildzoid actually did a very recent video on this, and since he has a gigabyte motherboard, it might be an idea to check as he specifically shows the setting to use on gigabyte board's to alter the VID offset. You need to alter the VID offset to undervolt with CEP enabled. Undervolting with AC/DC is limited in its capability and seems to mostly impact multi core load droop levels.

If CEP is disabled, then it shouldnt matter which offset you change, however BZ in a video I also recall mentioning he thought disabled setting wasnt working for the CEP on his gigabyte board. (meaning was on all the time).

What's the title? You don't mean the 'making 14900k fast even with 1.4 vid limit one'? I did already watch that one but I didn't bother trying because it's includes offsets. I saw another one with cep in the title but it was an msi board?

Nevermind I think I found it starting to watch now...

 

[chrcoluk]

Yeah they definitely are different things. Intel sure put a lot of work into making sure people don't have a way to fix their mistakes. Anyway I'll check out the video for sure just maybe later I'm kinda sick in bed right now. In my experience when it comes to offsets, UV protection doesn't allow them, CEP allows them, it just cripples them. Thats how I thought anyway, I was able to tell the difference. Anyway have a good one, man do I got such bad headache right now.


What's the title? You don't mean the 'making 14900k fast even with 1.4 vid limit one'? I did already watch that one but I didn't bother trying because it's includes offsets. I saw another one with cep in the title but it was an msi board?

Nevermind I think I found it starting to watch now...

Posted anyway for the benefit of others.

 

[ShrimpBrime]

you are correct on this observation, that is pretty normal natural (happens on AMD as well, clock stretching), that's why you disable C-States when benching, IDRGAF nowadays much on E-peen points, but I still post a few from time to time. I just try to enjoy these beasts as that's what I bought them for.

Hope nobody gets offended.

Disable C-states does not eliminate the droop.
But if you don't care, you don't care. I get iit. Neither do I. Enjoy the beast!!

 

[iameatingjam]

Well you mentioned an additional UV blocker on top of CEP, thats why I mentioned that.

I think CEP and the UV security feature are two separate things.

As I mentioned before there is two offset's, one for the VRM and one for the CPU VID request. Buildzoid actually did a very recent video on this, and since he has a gigabyte motherboard, it might be an idea to check as he specifically shows the setting to use on gigabyte board's to alter the VID offset. You need to alter the VID offset to undervolt with CEP enabled. Undervolting with AC/DC is limited in its capability and seems to mostly impact multi core load droop levels.

If CEP is disabled, then it shouldnt matter which offset you change, however BZ in a video I also recall mentioning he thought disabled setting wasnt working for the CEP on his gigabyte board. (meaning was on all the time).

Okay thats funny. I sat through all of BZs rambling only to find out this vid way of undervolting, is the way I've always done it. Even before learning about the ac/dc loadline. LOL. I guess its the b board that makes the difference.

However, I learned something useful... its the intel profile that stopped 104 from allowing the offsets. So all I have to do to make undervolting work and not f up performance is deselect the intel profile and select 104. I just tried it, and it worked, it did not reduce performance but actually increased it. and vcore is indeed lower (though weird its about 50mv lower when I set 100mv offset... must have been some other setting altered by using the gigabyte profile.)

BUT. I don't think I should keep it. Am I hurting or helping at this point? I am 1) using an old ucode which I'm pretty sure has bugs and was the ucode my other cpu died on 2. requires turning cep off. even though I thought it was off before. BZ kind of convinced me its a good thing to keep on if I don't want my cpu to die. Before I thought it was more of just a response to plundervolt. (Is it?) 3) no more intel profile. I can do my best to emulate manually but who knows what else its doing behind the scenes.

The difference is 1.35v vs 1.4v and at 1.35v passmark cpu test gave me 56,000 which is more than I've got in a while(otherwise I'm at ~54500). I mean intel says keep CEP on... they can't be saying that for performance reasons (since its bad for performance) they must be saying it for safety reasons, or security reasons I guess...? idk. Plus I haven't even done proper stability checks yet.

Yeah I think I better be a good boy and listen to intel's instructions.

 

[chrcoluk]

CEP will prevent the CPU from being fed voltages that are are outside of a reasonable range, so it requests lets say 1.30v and the board tries to feed it 1.15v, then you might get clock stretching or something. So probably for stability I think.

I cant tell you what to do, but your main concern seems to be stability, an undervolt the chip will be cooler and use less power, but its like an overclock in the sense you away from the default v/f curve so it "can" be unstable if the chip cant handle it. Using the latest bios and default settings, should keep you within Intel's guidelines, so for safety and stability that is probably the most prudent. Gigabyte's defaults on BZ's board with latest bios was inside intel performance spec.

 

[RootinTootinPootin]

Disable C-states does not eliminate the droop.
But if you don't care, you don't care. I get iit. Neither do I. Enjoy the beast!!

I kinda meant the consistency on the clockspeed..droop will always, infinitely be there.

 

[ShrimpBrime]

I kinda meant the consistency on the clockspeed..droop will always, infinitely be there.

I notice mostly when power is left to auto. Screen shots above are manual tuning. No droop on the effective clocks.

So played with the idea to leave v-core auto. Try a benchmark. At stock without increasing multiplier (frequency in general) the effective clock is as low as 5190mhz.

This is a little bit OC to try and bring the effective clocks up. Here's that result. This started under 2000pts stock clocks.

 

[iameatingjam]

I noticed that if I do the whole ac/dc 55 then llc high thing that I can actually drop my vid limit to 1450 without any performance penalty from 1550, when before even 1500 had a small penalty it seemed. I don't think this really does anything in reality, since vcore is roughly the same (at least its peaks anyway), its just because vid and vcore are closer together I think? I remember I tried this setup before and decided not to use it but I can't remember why. Well anyway it seems to be working well enough now, vcore doesn't pass 1.4 ( as reported anyway) which is I guess the best I can hope for without offsets or downclocking.

I notice mostly when power is left to auto. Screen shots above are manual tuning. No droop on the effective clocks.

So played with the idea to leave v-core auto. Try a benchmark. At stock without increasing multiplier (frequency in general) the effective clock is as low as 5190mhz.

This is a little bit OC to try and bring the effective clocks up. Here's that result. This started under 2000pts stock clocks.

View attachment 360102

Nice. Thats 50 points higher than the highest score I ever got with a 14700k.

 

[ShrimpBrime]

I noticed that if I do the whole ac/dc 55 then llc high thing that I can actually drop my vid limit to 1450 without any performance penalty from 1550, when before even 1500 had a small penalty it seemed. I don't think this really does anything in reality, since vcore is roughly the same (at least its peaks anyway), its just because vid and vcore are closer together I think? I remember I tried this setup before and decided not to use it but I can't remember why. Well anyway it seems to be working well enough now, vcore doesn't pass 1.4 ( as reported anyway) which is I guess the best I can hope for without offsets or downclocking.


Nice. Thats 50 points higher than the highest score I ever got with a 14700k.

And the VID looks pretty good too. I concentrate in the working v-core of all core loads rather than worrying about the idle VID for single and dual core loads/Boost clocks.

As you can see, the VID is close to 1.3VID during the benchmark. This is about the most that could be handled under ambient cooling from my experience.

I have 4 or 5 screen shots more of the clocks and scores before this one, or leading up to this one.

Did 1 more run with LLC adjusted to Lvl 5 and was hitting 344 watts with no increase in score and a top temp of 97c and a VID of around 1.35v and I did not take a screen shot of that.

So it looks like Intel has set an AVX max frequency of 5.2ghz. I enabled AVX2 for these runs and left the multiplier at auto to see how it reacts to small tweaks. So multi to 56 and a little bclk to go with it. All power settings auto but released to 4 thousand watts. Turned out to be 5.350ghz effective clocks.

309 watts there.