The great big rambling investigation of intel load line settings.
HTML-код
- Опубликовано: 11 сен 2024
- My Patreon: / buildzoid
Teespring: teespring.com/...
Bandcamp: machineforscre...
Twitch: / buildzoid
Twitter: / buildzoid1
Facebook: / actuallyhardcoreovercl...
#overclocking #intel #Z790
First?
Going to watch in the morning
lol
If your here after the dinosaurs. You are not in fact first.
The mad lad pinned it. Nice
@@aliensounddigital8729 but dino aint here
1:01:12 "I think that covers everything I wanted to cover in this video". *checks timeline* 38 minutes remain. Whew, I was worried.
Framechasers gets butthurt everytime someone mentions your channel or overclock because u explain everything for free the things he sells in his discord 😂😂
It should be pretty obvious to anyone with half a brain that he's 200% confidence and 10% actually knowing anything.
Even for someone like me who used to subscribe to his discord. He was an a$* hole
Where? Never seen it.
framechasers disabled my comments so they don't appear on their channel what a guy
He likes to lock everything behind $$........
summary; buy cpu > set llc low/med > forget about it?
(with 0x129 & intels default settings, setting llc low/med will undervolt the cpu + have an okayish vdroop/not overshoot into voltages that aren't safe = 1 setting saves cpu from degrading by working at lower volts than default volts, and keeping the voltage at a decent 'balance')
i might be missing something tho?
Yes p much
I didn‘t see the whole video yet, but from former videos i know he‘s using fixed voltages? What if i use dynamic, standard one?
Thank you. Was chuffed to see you did upload the 1h explanation.
Its astonishing to they manage to get these things to run this hard, that some of them even can run this fast without damage.
Seems pretty clear at this point this wasnt just reaching for that last 1%, but for yields. Seems like as many as a third of these chips have VID charts that on a moderate LLC were gonna do fine, which is still wild to me considering how hard they were run to try to beat AMD on single core.
I dunno if ur like me but as an old school overclocker, i kinda admire the engineers that got them to run this hard at all, if not the validation team that decided to push the yields and VIDs.
Yeah this seems like the claim of "too many i7 chips are being binned into i9 chips" to meet demand and boost profit margins is 100% true.
That makes a lot more sense if the claim that these chips apparently cost $300-350 to make when they first launched and that number hasn't fallen by all that much since. Yikes that's expensive.
mashing my DEL key in solidarity brother zoid
I love these videos because they encapsulate the insanity I too have personally experienced dealing with Raptor Lake CPUs. Shout out to my 13900K and 14900KS currently being RMA'd. Microcode update stopped my BSODs, but I still get crashes with MSI AB and other background apps. DEGRADATION IS VERY REAL.
EDIT: My 14900KS came back a couple days ago via cross-shipping. My 13900K is still being processed since I did the regular RMA option. Fortunately no more app crashes, voltage is stable, and my temps are about 20 degrees cooler. I went ahead and set some conservative temp and power limits for the time being. Boosting to 6.2GHz on stock settings. Yay!
Your faith is impressive
If you are getting crashing its RMA time. Might as well clock & volt the F out of it before you do for lulz tho
Degradation always was real. But it wasn´t in the form like it seems to be currently for some people.
My 13900k had a bit different behavior. I think its because I have actually ran the machine idle most of the time, but it degraded into a situation that without having any load, it was crashing. While gaming, it was stable. The idle crashing (discord, chrome, DWM) I needed to handle with running some load to avoid before the 125 microcode. The only game crashes I got was Senua's Saga, it was not able to start first even (doodle compression errors), until the 125.
It is now in RMA, thankfully I have documented my struggles on a public forum, and when these news came to light just sent them the link. They did not even try to sweat me, accepted the RMA right away basically.
if your CPU is crashing with the intel default profile and the 0x129 microcode. RMA it.
Best techtuber, hands down. 💯 I'm here for the delete key smashing!
My core group for tech on RUclips is BZ, GamersNexus, level1techs, Der8aur, and Hardware unboxed.
Yes
Same
just hold shift on restart hhnnnnggg
@@The_Noticer. Just hold down the delete key while boooooting. Or finally create that "shutdown.exe /r /fw /f /t 0" shortcut. And maybe even assign a fancy keyboard combination to it, like CTRL+ALT+END
@@sp00nwhoa calm down Professor
So another funny shit to remember is that "VID" in HWInfo is not THE request but request corrected by DC LL parameter.
You must have had So much fun thinking this one up!
Thank you for taking the time to explain the questions that end users want to know the answers to, but rarely see anyone talking about them.
The "What If I do this?" sort of video.
nice pattern to the delete bashing this morning.
Thank you so much for making these videos. I've been watching this Intel series about 13th and 14th gen instability as I've been experiencing extreme 14900k crashes. I've just RMA'd my old cpu now that the beta bios is out. Hoping this CPU with the new bios won't degrade as it'll have the new microcode and will have never been touched by the old bios settings.
Literally my favorite kind of video just rambling for hours perfect.
1:36:38 same here, my 13700K on a cheap PRO B660M-A WIFI DDR4 mobo I can use mode 1 with additional -170mv undervolt in ThrottleStop without any stability issue, the systems (I have two of them exact specs) have been running 24/7 since June 2023. Now I'm confused if I really need the patch from Intel or should I just keep it this way?
Unless you're seeing spikes in voltage to high values it shouldn't *really* be needed, but it's probably better to do so anyway.
Awesome video and testing. A lot of the 13/14th gen issue(s) and factors that are important for just running the CPU within sensible limits (which is not the defaults currently still), let alone undervolt, come together in this and the oscilloscope is the proof. I think this shows why (oversimplified) "high AC LL causes degradation" is true, unfortunately. Some interesting choices by Intel... But the workaround is clear. That said though, much respect to the engineers and all the testing involved, it's awesome tech.
If Intel would put as much effort as you do in this problem it would´ve been fixed 2022 already. lol
They should literally be paying someone to do what BZ is doing. That’s how QA testing works.
*effort
@@samiraperi467 sorry that english isn´t my native language .... but i corrected it thx ..... hope you feel better now.
@@hammerth1421 guess the QA for 13 and 14th gen was just "Power on, POST works, fine" :D
the funniest thing is ignoring all the 'yadda yadda I'm sure Intel had their reasons to let their cpus get zapped'; 1.6v has been the rapid death zone for ambient cpus for decades now, at least since the early 2000s.
Perfect timing, as I am revising my load line settings!
Practical video next?: What to do for dummies?
For Asrock Owners: What I did to my 14700KF:
Bios with 0x129 microcode
Intel baseline loaded
CPU Vcore Compensation: Level 1
Base Frequency Boost: 125 W
CPU All performance Cores 5500 Mhz
-100mV offset
Loadline Calibration Level 5
PL1: 125W
PL2: 188W
Running Aircooling max 80C with Cinebench 2024 and Vcore is less than 1,4V now.
Good Luck!
Honestly I am using ASRock Z690 PG Riptide is a total mess, most settings aren't explained clearly; also most people buy MSI not ASRock; what is IA AC/DC LL no explanations, they're set to 1.3
My AC DC LL dynamic loadline is about 1.5 and my cpu vid peaks 1.6+ sometimes.
This is for an asus rog strix g16 laptop with an i9 14900hx. Mainly used silent profile for low turbo boost ratios cuz I had consistently high temps
Did my cpu degraded and what should I do.
Never undervolted untill now to -80mv but it wasn t very stable.
I do get occasional 0xc000005 crashes on plenty of apps randomly.
So why did intel say intel 14th on mobile aren t affected. And how should I proceed? :(
I need yr expert opinion. My unit costed me about 3300 usd in a developping country cuz high taxes 😢
PS : I only had it for 1 month so far, not even 2 months passed yet
email the laptop brand manufacturer if you bought directly from that brand or go meet the shop/seller.
One word, warranty
I'm working on a new website that should be up in a week. Going to be just a dump of random knowledge and stuff i do. But I wrote a post about Gigabyte Z790 Master X Bios Setting Definitions that will be on there. It is going to be more like a series of reference/ cheat sheets starting with DDR5 terms and definitions. Because the Bios descriptions that Gigabyte uses and terms for that matter are not helpful at all and I'm sure a lot of people would appreciate knowing what things do, might feel more tempted to play around. Your videos always help me a lot especially when you use this motherboard, since I'm running a Gigabyte Z790 and i feel they don't get the attention as much as Asus and MSI does when referencing the BIOS options. Thanks, This video helps me a lot.
I have an MSI board so cant set the max voltage limit. MSI has refused to implement it according to a forum post. MSI does have CPU lite load however, and that seems to help spikes stay lower if you are able to set it to a low mode. That of course depends on CPU quality, not everyone can.
set CPU Lite Load 1 and call it a day
The MSI Lite Mode just sets the AC and DC load lines. You can check this by setting a Lite Mode, rebooting, going back into the BIOS, setting the Lite Load Control to Advanced, and then you can see what values have been applied. For example for my MSI Z790 board Lite Mode 1 is AC/DC 1/1 (0.01 mOhm), Mode 2 is 5/110 (0.05 mOhm / 1.10 mOhm), Mode 3 10/110, going up in 5 point increments, so 0.05 mOhm, until Mode 9, which is 40/110, and then after that it goes up in 10 increments, up to Mode 16, which is 110/110 (the default 1.10 mOhm for both AC and DC). And after that, both AC and DC are incremented by 10 points, with the highest mode being Mode 23, which is 180/180 (1.8 mOhm).
@@sp00n yeah i noticed that today, too. On default it gives me mode 23 which made my cpu spike in an easy running game desperados 3, with 1,65v. I thought it will explode any moment. It's insane. Overclocking has become so complicated due changing voltage, multiplicator, turbo and different states. Back in the day I would just set a constant voltage and multiplicator and call it a day.
almost two hours of Actually Rambling Overclocking. I love this. *Sits more comfortably on train bench, while putting on headphones*
unrelated but when bz restarted the computer i flinched as my brain was like NOOOO DON'T RESTART MY PC WAAAAIT..
what a stupid brain i have.
Really glad to see this as after following the Raptor lake disaster, I've been curious as to how all this seemingly convoluted system works.
Can you please comment on my understanding of LLC and Vdroop?
1. CPU VRMs are not perfect and the voltage sags on transient loads. This is just physics, and some VRMs handle this worse than others (voltage sags lower).
2. To prevent voltage sagging below the "stability threshold" there are Vdroop and LLC.
3. The amount of Vdroop is set by configuring LLC.
4. The CPU is configured by setting the AC load line to match the LLC (amount of Vdroop) of the motherboard.
5. The Vdroop and LLC force CPU to *request higher voltage* on idle (base VID + "Vdroop calibrated offset"), so when a transient load hits it, the voltage sags right to the necessary "base VID" point. Basically we create a safety buffer for the voltage to sag, and the *lower the load, the bigger the buffer* .
6. This is configurable because some VRMs have terrible transient response (voltage sags a lot), and this requires higher average voltage for the CPU (so in the worst high-load scenario the CPU remains stable). And vice versa: for amazing VRMs these values can be set lower since the voltage sag is not as terrible.
7. None of this actually protects you from voltage upward spikes on load release, but more steep LLC may result in less pronounced spikes compared to the "average" low-load voltage (which is actually higher than average voltage with a shallow LLC, so the peaks are actually the same).
Thanks for your amazing work out there! I really love such passionate people as you.
So with all of that said, would you recommend using voltage offsets to undervolt vs AC Loadline? Default AC on my ASUS board is 0.4mv, if I set SVID behavior to trained it goes down to 0.26mv, both are stable on my 13700K. With that said, I can achieve the same load voltages simply by using a higher adaptive voltage offset with AC LL and the CPU's LL set to the same.
adaptive mode voltage offsets are compatible with CEP. You can even set different offsets for each core clock ratio if you want to really fine tune it.
@@ActuallyHardcoreOverclocking, and if I only have a "static" offset option (MSI B760M Mortar II) I guess the only difference is that with AC loadline tuning I'll get higher low-load voltages and lower high-load voltages than with static offset? While with static offset the whole curve goes down.
Or am I missing something and all offset types are actually CEP-compatible? I'm not sure if it's different on the B-chipset.
I also have the option to set the offset "by CPU" or "by PWM" whatever that means. Man I love that every BIOS has its own names for the same things.
Great video. In summary, 2 issues. One is the motherboard manufacturers are undervolting the CPUs. For me, undervolting is got consent from Intel at least in an implicit way, because it makes benchmark better. The 2nd issue is that the CPUs request too much VID and get degraded. Thus, those are all intel’s fault.
thanks buildzoid i watched the whole video, the windows calculator is a mighty tool!
Extremely usefully: now I understand the connection between the upper-undervolt regulator and the actual physical V-limit and how they interact.
As always, marketeers are cancer for any quality control system: Intel was not half this bad when the 13th gen launced
I think intel actually started implemnting this VR control scheme with 10th or 11th gen.
1:35:00 yeah makes a lot of sense, that is why my 13700KF gets crazy with the new firmware/microcode on Asus and drops the clocks to the stone age with my lower voltage config and request a lot of voltage on auto
"If you're gonna exceed something, exceed it a lot"
- Buildzoid
Finally a normal lenghty video again. Was worried for a second that the Zoid went for sub 30min RUclips Shorts.
Is going above 5Ghz really worth all this trouble.... I'm very tempted to just down clock to 5Ghz on P cores and 4Ghz on E cores
B660 locks a 13600k to 5.1. Basically set it and forget it.
So... If i understood correctly. And TL:DW, i guess...
1) Both AC and DC loadlines are EXTREMELY questionable decision. That must be set equal to value that VRM loadline is set to (or balanced around it).
2) For some reason AC loadline doesn't work as expected, even if set to value equal to VRM loadline motherboard as it corrects voltage based not on actual current measurement, but on some theoretical current value. Which is dumb. Meaning that it either overvolts or undervolts CPU's under heavy load. And of you were to balance it, resulting value won't be actually equal to VRM loadline.
3) DC loadline is shitshow. Don't understand reason for it's existence. So this one actually works based on proper current draw (or at least it seems so), but it only affects VID request reporting (and power draw). WHY?! What reason does it have to exist?! CPU makes VID requests without it, why not just report it? (But oh well, you sure can make some unrealistically low voltages by setting AC loadline too low and DC loadline too high).
4) VR voltage limit tells CPU that it cannot request voltage higher than X. But it doesn't prevent VRM to output voltage higher than said X due to voltage overshoots on transients.
5) CEP (aka Clock Stretching) really won't like if you use Dynamic VID offset to undervolt CPU, as CPU won't know about this offset (because it is done directly at VRM). Instead you must undervolt through Adaptive Mode.
6) Intel uses high loadline reference, which, when overlapping AC loadline shennanigans and VDroop prediction VID spikes doubles down on voltage spikes.
I hope i hadn't missed anything?
Intel Meeting: "Why don't we hire Buildzoid?" "No, because he knows stuff"
People at intel knows there stuff very good. I am allmost sure people who created 13-14th said to their leaders that 1.6V and all these 6.0Ghz is not good idea, but than marketing part came and said we have to beat 7950X in cinebench or something like that, so 1.6V and 6Ghz with 400W TDP was only way to do it and engineers had no decision about that.
They probably dei hired someone.
The core engineers do not even design the products. They design the architecture and features, and then a wholly different team decides how to bin them, what features go in what products, and how aggressive to be on the frequency / voltage curve. Those teams get leaned on massively to keep up with AMD / outperform previous generations. As said, I suspect this architecture was never meant to be pushed this hard. Also, Intel's QA probably has internal motherboards that stick closely to the Intel spec, and what the board partners do comes as a surprise to them
@@peterwstacey which shouldn't be the case because Intel should have been validating the default settings that these motherboards come with.
@@peterwstacey Intel does not mandate, but only recommends to their board partner, because they want them to push the CPUs to the limit to win them some benchmarks against AMD. Intel is really afraid to set any reasonable limits fearing it might hurt their benchmark numbers.
and does it work with z790 aorus elite ax ?? i have same as u i9 14900k
Thanks for this and especially for doing it on a gigabyte mobo. This cleared a few things up for me, still got a lot to learn on modern cpu regulators.
I was playing around with my LLC and AC/DC load lines on my z790 master last night and I found the gigabyte default load lines pretty much perfectly matched the VID requests to the die sense for my board which was a surprise.
As an Intel 13980hx laptop user I fully support you getting a laptop.
Try an MSI laptop like mine. The unlocked BIOS had a ton of overclocking options, but half of them are janky or don't do anything. You'd have fun!
I'm trying to find optimal AC loadline and negative cpu core voltage offset settings - without the benefit of knowing or controlling LLC and VCORE.
It's like hard mode!
Im running LLC @ medium and AC 16 - DC 50 on gigabyte Z790 gaming x ax , why my vid and v core from gigabyte sensors match , but my VR OUT v core is much higher from ONSEMi NCP81530 then VID under load ? Its strange! Under load ViD 1.250 , V core 1.255 but VR OuT reading 1.390 ! i don’t get it
Sadly, while I understand some basics, I don't have a background in electronics and have a question. At 56:23 I don't understand why the DC loadline offset isn't used to change the actual VID requests. What do you mean by "used to calibrate the power measurements"? If not to calibrate the requests then what is it calibrating? Just lying in the hardware info reports? Is it just a design mistake that the readouts don't match the actual requested voltage?
OC3D TV says in "Intel CPU Undervolting Guide" that a high LLC value of 7 on ASUS MB does not overshoot AT ALL.
How is that statement compatible with this video, showing over and undershoot on high LLC value ? Its very confusing.
Can we have both ?
I actually asked the guy this question:
"WHy use LLC 7 on ASUS with zero explanation? This will give your voltage more overshoot than a lower LLC value."
and he answered:
"Doesnt overshoot at all. There is still a small amount of droop but its as close as you can get without over volting it (because of too much LLC)"
Can there be any conclusions drawn if a 14900KF can't run Cinebench r23 multicore if everything in the BIOS is set to Intel Defaults / Auto, except for the AC and DC load line, which is being set to 1 / 0.01 mOhm (to try to measure the default LLC mOhm value, as shown in this video)?
interesting stuff!
Hey BZ, can you check any MSI board with 129 uCode update? Supposedly MSI did not implement IA VR Max Voltage at all and there's confusion whether or not the uC actually do anything.
Thanks BZ. Ive got an adaptive offset of -0.090V on my i5 (with a significant frequency OC thats been stable for a year) hopefully thats enough to stop it cooking itself. Ive just left the Loadline settings on auto for my MSI board hopefully that's fine.
At around 58:34 you discuss how the dc load line is used for readings, not for the “actual VIDs”- do you have a sense for if the IA VR voltage limit is applied before or after the DC load line adjustment is applied? Based on everything else I’m assuming it is not taking dc load line into account since it is a voltage regulator limit.
Edit: I think actually the later portion of the video where you show the voltage limit testing has helped answer some of my questions about how it works. Thank you for what you do!
Intel says DC load line must match AC load line. I think DC higher is minus readings and lower is plus readings.
correct, higher DC LL reduces lower DC LL increases, for what i understand what you want is to set your VID averages as close to the Vcore averages voltage at full load
Z790 APEX:
I applied IA Safety features on: TVB enabled, ICCMAX 389A. I tried with these enabled & also added in a .9/.9 iA/DC Load Line. This resulted in almost no difference to a Manual VRM Vcore setting of 1.375. During testing with R23 runs after adjusting VRM Vcore to automatic, the voltage readout in HWiNFO didn't change much. The temps did! 5400MHz on CPU with boost to 5600MHz, the temps went from 94 on hottest core @ 5400 (during R23 and manual vcore 1.375) to 100c on 4 cores. According to ASUS EC: Max Amperage during test 196A during Auto voltage. Also VOUT up to 1.39. Why would the automatic VRM voltage affect the temps so much?
Thank you Man for your great explanations and your work.
"why didn't they mandate a loadline spec when there is so much headroom for undershoot". What if they expected faster degradation than normal and wanted to bake that ~100mv headroom in so that as a processor degrades, it takes a lot longer before performance/stability degrades to the point of anyone noticing?
Thanks for the explaination on this. I have a i7-14700k that I haven't used much and thankfully doesn't have any issues currently.
I was previously using a higher vdroop loadline and using the AC/LL for undervolting with IA CEP disabled.
I found out what the value of each LLC setting for MSI Z790 boards is and used your explanation of modifiers to calculate what to set my AC/LL and DC/LL to, in order to match a shallower LLC. I then enabled IA CEP and slowly increased my negative voltage offset until my Cinebench cores started regressing.
Now my CPU scores a few percentage points higher in benchmarks while staying a little cooler and drawing less power. Amazing.
LGA1700 is a mess, and I'll be glad when it's dead.
@nighthound2714 what values do you have set for your cpu?
@@raiku9161
I found that the multipliers he listed for the Gigabyte LLC are the same for MSI. LLC 8 is 1.1 mOhms and LLC 1 is 0.01 mOhms.
I have boost behavior and multipliers set to Intel defaults (No multicore enhancement, PL1=PL2=253 W, and IccMax=307 A), along with TVB, TVB voltage optimization, and IA CEP enabled.
I'm using LLC 5 with AC/DC LL set to 22. This still gives me a bit of vdroop to aid in the undervolt without triggering CEP. I can't say for sure about low load voltage spikes, but I get a max of under 1.35 V according to the VR Vout sensor while Cinebench opens. My actual undervolting is done with offsets on adaptive voltage. The offset I use doesn't matter, since it's going to vary depending on silicon quality.
I'm being conservative with my offset. My goal is maintain "stock" performance, while avoiding thermal throttling during realistic loads like cinebench multicore benchmark and OCCT stability test.
@@nighthound2714 is iccmax just the current limit? Is 366A not the default? Why is yours so low?
Do you mind still telling your current offset? I know it depends on the cpu but just to have a rough idea.
I have a b660 f asus motherboard. And I currently have an offset set of -0.11V. (I only have crashes veerry rarely).
In cinebench my vcore is around 1.28V I think. But my system throttles in cinebench 23. I get around 31-34k points. Which is fine for me.
In games with 40-50% cpu load. I have a vcore of around 1.42-1.46V. Is that normal? Sounds high. No throttling, just high temps.
I'm new to llc acdc. I want to have lower temps and not such high vcore numbers. I will try some settings out.
Have the same cpu
@@raiku9161
307 A is the value for the Intel performance power profile. Lowering your iccmax value will reduce both power consumption and boost frequencies. The latter will help your stability issue.
I'm still testing different configurations, but I have an offset of -0.020 V for the settings I mentioned before.
Intel made things more complicated than it needed to be, but you'll learn by experimenting. Don't worry, not enough voltage won't cause any hardware damage. I suggest disconnecting from the internet so Windows doesn't try updating with an unstable undervolt.
@@nighthound2714 okay. I will try 307 A out. I will set a vcore limit to 1.5V just to be safe while testing the settings out.
What is your vcore in games? Are my numbers mentioned above normal?
Thanks a lot
Is the bios you're currently on a beta bios or a full release? Still waiting for a full release on my board :\
I've been having a blast with AC DC on my 13600k on an ASrock z790 pro RS D5 ( this is so confusing ) !!!
I'm so happy you uploaded this when you did. So if I'm at medium-low LLC which is the auto setting on my board (level 4 of 1-5, 1 being least vdroop or extreme equivalent), what should my AC DC really be set to exactly? I have the exactly .3 and 1.1 that you mentioned for AC and DC respectively, that is their auto value as well.
I believe Intel is adding this voltage because of the extreme frequencies they’ve been running and the exponential voltage required per bin. Because they know the higher number sells CPUs. And they’re reluctant to use clock stretching because it affects benchmark scores.
BZ talks, we listen! Thanks for the uploads.
Aren’t you supposed to match your DC LL to your LLC for correct readings? Your AC LL should dictate what voltage is requested and LLC how much droop
Basically, match DC LL and LLC, and set AC LL as low as you can while remaining stable?
Isn't there a performance penalty to clock stretching? So I can see why Intel is trying to use a method that avoids that.... too bad they can't actually get any board maker to actually set the LLC correctly.
Thanks 👍👌 I think my 12400f is using a lot of Clock Stretching when trying to undervolt via Offset on B660 Strix 😅.
Power and CB Points go drastically down, Clock stays same xD
Have to read in this AC DC Loadline thing more on my B660 Strix 😅🐢
How are you undervolting on a B660 Asus board? I'm trying to do the same but it seems that mine doesn't have the option for offset, it can only set a static manual voltage.
Turn CEP off and validate your UV without CEP. If it crashes the voltage just wasn't high enough and CEP clock stretches to safe it but doesn't actually perform any better.
So imo check effective clocks in HWInfo while running Cinebench, check your scores and if scores and effective clocks go down your undervolt ist too aggressive
Use HWiNFO64 to identify clock stretching. You will see core clocks and core ratios, and core effective clocks. Compare those datapoints under all-core workload like R23. If Ratio is commanded high at what youd expect but effective clocks are down, you are clock stretching.
Buildzoid, perhaps you have the background knowledge required to explain why the prior trend up to about 90nm was "smaller node =lower voltage" has been broken?
The 486 started at 5v, Pentium was 3.3,then 2.8v. P2/P3 gradually dropped under 1.8.the p4 dropped to 1.4.
Since then, it's fluctuated, and it feels completely crazy to run the same voltage as a 90nm Prescott on a "7nm" chip today.
Why idea why this trend broke?
Different transistor types or fundamental physics limitations would be my guesses.
Your hard limit is likely the 0.7v of the silicon FET body diode iirc
From simple transistors to better mosfets - and now we are scratching against the electrical limits of silicon-based diodes: You can get them as low as 0.5V but then they have rather poor characteristics otherwise. For FETs in CPUs we are looking at ~0.5-0.7V like mycosys said. At low frequencies the CPU can go to 0.8V or there about. But for higher frequencies you need higher voltages cause the FETs basically act like tiny capacitors that need to charge up before they can conduct.
madeup example:
At 0.8V it might take 500 ps till the gate-capacitance is charged enough to conduct. with 4 gates in a row that would be just about fast enough so that at 400MHz clockrate all transistors have finished switching.
If you want to reach 1GHz you need those gates to charge up to the same 0.8V in less than 250ps instead of 500ps. The gate-capacitance, trace inductance and resistance are pretty much fixed for a given manufacturing node-size, so the only way to charge the gate faster is to apply a higher voltage.
So in the near future you will keep on seeing ~0.6V as minimum voltage when idling, and mostly less than 1.7V at peaks.
@@ABaumstumpf so, are you saying there's an absolute floor on the voltage required to switch, but at higher speeds the voltage simply can't ramp up fast enough to achieve switching? So in essence, like led displays, they're being "over driven" to achieve faster ramp ups?
@@ABaumstumpf so, are you saying there's an absolute floor on the voltage required to switch, but at higher speeds the voltage simply can't ramp up fast enough to achieve switching? So in essence, like led displays, they're being "over driven" to achieve faster ramp ups?
BTW, my only conclusion so far regarding 13/14th for best performance (assuming you have average/shitty silicon) is (i use ASUS MB):
1. anything less then 1.0 AC LLC (or 100 for gigabyte) will be unstable under extreme heavy load (mine is rock solid stable only at 1.1), and DC LLC should be equal (or any in range or 1 to 1.1 as long as it still stable, this only affect the VID requests by the CPU) + disable CEP
2. ideal CPU load line is 3, if unstable then 4 but no more then that, the heat impact the performance a lot! (load line 3 is both stable and not over-heat) and this in turn translate to more performance
3. never exceed 400A ICC
4. limit your Watt usage based on your cooling solution, i can easily set it to 280 - 300 Watt and never reach 100c for all cores
5. use eTVB
6. from this point you can try lower the actual vcore voltage (using adaptive), but mine is not stable at 6Ghz when doing this
also, from that i conduct that less heat gives more performance then adding more Ghz by OC, you should almost always try reducing the heat with CPU LLC instead of other settings
I’d love to see what you think of MSIs CPU Lite Load modes that seem to reduce AC values. Most of them below default screw with CEP and heavily reduce perf.
What is a good AC / DC ratio to start with?
14700kf
they seriously couldnt design a more overcomplicated system if they tried, could they?
I'm not buying it that any of this nonsense actually results in better performance nor efficiency or is necessary for stability.
On my ryzen 3900x, by far the best setup has been to just set a fixed voltage and clock, the most aggressive loadline, which STILL allows the voltage inside the actual cpu to drop significantly, but to a manageable amount no matter how extreme the load is, without getting excessively hot. Which, funnily enough, is exactly how cpus used to work by default over a decade ago.
As for boost clocks: I dont have a single application that actually needs JUST one core but at insane frequency, because it turns out all modern games I care about actually use enough cores to negate ALL boosting behavior these cpus offer, meaning my fixed clock is higher than that.
And of course with how stupid boosting on cpus works, I cant actually adjust the curves to still be able to have single core boost at the same time, making the entire system utterly useless.
Can you please share how exactly to check the VRM load line calculation you did, from the HWmonitor reading? you made it look to easy, and math is my Krypton :)
so.... why don't we slap extream lodline calibration and minimum ac load line? 1:05:00, because we are scared of over/under shoot. there is going to be voltage spikes on transitions.
from what i can see : 1:27:39. voltage spikes due to transient doesn't look to be huge (~~50 mV)
I'm wondering this as well. If the undershoots aren't a problem anymore (as they apparently were in 10th gen), wouldn't be a LLC with (almost) no Vdroop + a voltage that's close to the operating voltage under full load prevent some of the crazy high VID requests?
How dead do you think I would be If I drank every time you say "voltage"?
I tried to identify the mOhm value for my MSI Z790 and 14900KF, but I'm severely limited by temperature. Don't know why, I have a 360 AiO and re-seated/pasted it multiple times, but I still hit the temperature limit almost immediately when running Cinebench r23.
Then I thought maybe if I disable the E-cores and only run with P-cores for more thermal headroom. But then Cinebench just crashes if I set both AC and DC LL to 1 (0.01 mOhm), which is really weird when everything else is still on Auto / Intel Defaults.
Can you also do this measurements with a negative Vcore offset, or will this interfere with the calculation?
What are your settings? LLC mode, lite load mode, power limits, core voltage? If you have no power limit and high LLC it's quite possible to hit 100C with this chip even if you have 360 AIO.
I would start with lowering power limits. start with 150W and go up with increments of around 20W to see when you hit thermal throttling. Assuming you have correct power readings and for that you need correct DC LL (for me it is 1,1 mOHM on MSI Z790 Tomahawk Max). Basically - too low DC LL = overreported power consumption, too high DC LL = underreported power.
@@Savigo. I'm basically limited to around 220-230 Watt Package Power, I've never seen it hit the 253.
But in most videos they easily exceed that value with a 360 AIO, so that made me really wonder what's going on. The Cinebench score is sort of fine though, not reaching 40k ofc, but around 36k.
I was able to find a stable undervolt of -0.130v, but to determine the actual LLC mOhm value, I set everything back to default and set AC and DC to 1, as in this video.
If you keep having to raise the voltage in the design then at some point 100C is gonna be a feature.
@ActuallyHardcoreOverclocking Are mobile or HX CPUs really not affected?
I think I'm going to fall asleep while listening to this
I'm not following even 10% of what you are saying neither I have an intel cpu but I'm curious about this stuff and I always learn something (I think/hope 😂)
MSI PRO GAMING Z790 motherboard with 13600K with 0x125. I have voltage offset -0.125 and it never goes over 1.13V and 60º, with 5.2/4.2Ghz OC, 24100 CineBench 23. Noctua D12. When running CB the most it goes is 73º
My goals is to the understand what your taking about by the the time I hit next semester in my IT Degree
So the fix for Raptor Lake degredation is literally to turn on CEP and max out the load line setting in BIOS? That's pretty wild! LOL.
It would mean that degradation doesn't occur on non-Z boards since you cannot disable CEP on them. Unless you run some old microcode that still allows it.
@@Savigo. Wendel from Level1Tech was informed of degraded CPUs that were run on server boards (W series).
I love he sounds fed up with intel lol
I spoke with ASUS Support in France, and they confirmed that only desktop CPUs are eligible for the extended warranty support. I'm beyond frustrated, especially since I have a mobile CPU! We need to hold Intel accountable-they're fully aware that their 13th and 14th gen CPUs are failing, yet they refuse to acknowledge it. There should be a lawsuit against them because their actions are unethical and possibly illegal. This is the last time I’ll ever purchase an Intel CPU or any system equipped with Intel hardware!
Okay so maybe I am understanding this wrong but on my ASUS board I have LLC5 set which is auto for the 0x125 and 0x129 microcode update, then I have AC/DC LL at 0.73, the DCLL and LLC are synced in the ASUS boards, I have Intel Extreme Profile set, CEP on for IA and SA, IA VR 1400 limit as well, the issue I have is when I try to set a global core offset voltage set to adaptive -0.110 I will get BSOD, when I turn CEP off I still get that issue so I lowered my undervoltl to -0.05 which isn't much but just to test. The BSOD gets triggered by one thing each time which is the vgc.exe client, riots anti cheat..... but it only crashes when the game is NOT running, cinebench23 will pass and I got y-cruncher to pass as well, so is it due to the -0.110 undervolt or the vgc.exe? 96sp is pretty potato so it would make sense for that undervolt to be bad but even BZ sp on his Asus board was worse and he set a more aggressive undervolt, still trying to understand all of this but I just want this to work lol.
The double boot might be to do with your ram, i have a kit that forces my pc to double boot as well, might be worth a test while your at it
nope it's an OS thing This happens with this OS on every single gigabyte motherboard I have DDR4 or 5(haven't tried with a DDR3 board).
Excellent video! I have a 13700k CPU and an Aorus Elite AX motherboard. After a 0x129 flash, during Cinebench, my CPU reaches 97°C, but before 0x129, it was 85°C. I have an H170i 420mm cooler. Could you help me lower this temperature without performance loss (aside from the 'official' performance loss)?
Without setting some manual stuff i realized that as well with 125 already. (im on an aorus z690 elite d4, there is no 129 yet) but after setting the -0.100 offset it feels like back where it was :) Using an arctic lf 2 360. As long as we not thermal throttle it´s all fine for me tbh ^^
@@TigerNationDE I see, thanks for the info!
I have two 13700K systems but haven't got the 129 patch yet, but I did undervolted them since day one using ThrottleStop, all I did was just offset the CPU Core and CPU Cache to -170mv (may vary for your system)
Also, in the MSI bios, I put the lowest mode for the LLC, which was mode 1. This is important because the higher the mode, the higher the voltage and temps.
Now I'm not sure if I need to patch them because using ThrottleStop I can switch their speed to any GHz I want anytime with just a preset change in ThrottleStop, which further reduces voltage.
For example, when I'm just doing nothing (browsing, watching, etc) I just switch to the disable turbo preset which will lower their clock to 3.4GHz and disable turbo which reduces the wattage to around 30ish.
When I'm editing, converting, gaming, I just switch to 4.7GHz or 5.2GHz whichever needed.
just follow any good undervolt guide - nothing has really changed since 12thG (apart frm it getting even more useful)
Try to set the pl1 pl2 limit.
Great video lots of info 👍 still a bit confusing what to actually set to get the most efficient running cpu at stock boost levels . Would love to see a vid with an undervolt the coorct way mixing offset voltage and LL settings . Stable and efficient with best scores .. 👍 I7 13700K ⚡
I think I may have switched to AMD at the right moment.
30+ years of Intel, with maybe a Cyrix or AMD one now and then. :D
Intel has been good to me over the decades. I'd hardly jump ship after them putting out a bad model. That'd be like me not liking BMWs anymore because they made a bad model one year.
@@DingleBerryschnapps I hated Nvidia for a while now, I've had their stuff since they started. So I wanted an AMD GPU and since their CPUs are pretty okay right now an all AMD system made sense. =)
there're several point in time where we should've abandoned intel, first is when we are told that 10nm+++++ was lagging behind competitor, then people calling for boycott on it due to geopolitics, then 13th gen debacle. The last people to leave win 1million$
I’m not updating to the new 0129. I’ll manually configure my bios. This whole fiasco is just frustrating. Intel needs to refund every one of these pieces of silicon garbage instead of beating around the bush with USELESS “placebo” fixes in the hopes everyone just forgets about it. No more Intel builds for me. I just cannot risk CPU failures in my builds.
Clock stretching probably causes some nasty stutters in game is my best guess
when what you've implemented requires an oscilloscope to understand I think it's fair to say you've failed. How on earth they expect your average user to first of all understand what to configure and then to make correct settings is beyond me. Great video though, at least I now understand what the DC load line does.
in my case, the IA VR voltage limits dont work after 0x129 code... i set it to 1350 mV (or 1.35V) and CPU still overshoots this setting and consumes over 1.4 Volts in normal computing tasks, and higher in games/heavy computation. what am i missing here?
Maybe the minimum base performance the cpu needed was more than that? What cpu did you use?
I havent update 0x129 , so my IA voltage limit to 1400 still work for my 14700k
@@user-tz7bl8vw6b i figured it out, after writing this comment... i inserted wrong values... but now everything is managed perfectly. i also use 14700k, power limited to 240watts, and now voltage limited to 1.4volts... temps remain below 85 degrees at max loads
@@realracing3specter295 same here. On game and working load onky 55 to 60
And idle at 30.
I think for now it safe to put the limit at 1.400.
I hope it stops degradation and any other stupid action
subbed. Thanks for the video
How does AMD CPU account for LL trim? It's not like AMD has any LL parameters.
I also why won't you start saying "AVP" instead of LLC since you've already done a video about it.
48:24 you mentioned ringing. There are sound mitigation options in the bios, are these related?
No ringing in electronics refers to an electrical effect not sound.
@@ActuallyHardcoreOverclocking thanks for clarifying
overcomplicated design is so "intel" 😂
as far i can tell clock stretching was originally designed for io clocks to prevent buffers from overflowing
and it was only later adapted to power management after dynamic boost for cpu's was introduced
Im getting 1.4mid high 1.4 on My 14900KS while Im watching this vid. is that normal for this CPU?
Okay, I have a possibly stupid question.
If 13600KF is just a cut down 13900KF...
Isn't it safe to set the IR Voltage Limit to 1400, if it's stable at those voltages since the substrate is the same?
To prevent degradation, I mean. And assuming I don't touch the ring bus setting at all.
(I get really good OC results under 1400 - 2 cores @ 6 Ghz, 5.6 multicore on mine, must've been lucky).
To be totally clear, I usually run at stock with an undervolt, but if the settings seems fine under 1.4...
It'd be nice to squeeze that extra juice out of the CPU
@@thelxr 1.4V limit should be fine.
@@ActuallyHardcoreOverclocking Thank you good sir. I'm a bit too cautious, so I just had to ask.
Am a big fan of your research videos!
Thanks!
Praise Buildzoid
1:24:26 so... if I use intel XTU to use adaptive voltage and test my undervolt. Am I doing it right when CEP is on? Or is CEP clock stretching and giving me more stability than it should?
Should I test my undervolt with CEP off? (I have the undervolt protection off in the asus bios).
Right now, I can am setting the max frequency and testing the CPU stability with XTU, and then lowering the voltage with adaptive as it goes. I am testing each of the VF curve points for how much undervolting it takes.
59x takes -0.08V
56x takes -0.20V (I was surprised it was this much...)
and so on down the VF curve list.
If I turn CEP off, should I get less undervolt results because it doesnt clock stretch?
CEP causes clock stretching but gives a stability at settings that wouldn't be stable otherwise. Best practice imo is to turn CEP off, check effective clocks in HWInfo and Cinebench scores to validate clocks and performance. Set your OC / UV accordingly and turn CEP back on after that.
If it crashes with CEP off the voltage wasn't high enough to reach the clock you put in
@@whatdafuq9 Fair enough... Will retest the undervolts. Thanks!
@@whatdafuq9 Clock stretching / lower scores in Cinebench with CEP enabled doesn't _necessarily_ mean that your setting is unstable, at least from my limited testing I did when I accidentally had CEP enabled.
I was doing some tests with a fixed Vcore of 1.3v, and with CEP enabled Cinebench r23 would give me horrendous results with LLCs with higher Vdroop, compared to when CEP was turned off. It was around 6000 points difference, and it at least completed Cinebench several times with CEP both enabled and disabled. (I didn't check any further for stability, went to check other settings with CEP off)
@@whatdafuq9 I came back to report that this is not how CEP works. The 'current excursion protection' seems to limit spikes and peaks, since you can undervolt further down with CEP off, meaning the chip is more stable with CEP off.
With CEP on, the chip doesnt go as low in the undervolts.
In Intel XTU I saw higher voltages with CEP off.
So if you plan on keeping CEP on after tuning, then you might as well tune with CEP on.
When i run Cinebench R23 with intel default settings hwinfo64 tells me that the cpu are power limited.
Using 14700K with 1503 bios that uses 129 microcode
Core Power Limit Exceeded, Package/Ring Power Limit Exceeded, IA: Electrical Design Point/Other (ICCmax,PL4,SVID,DDR RAPL), IA: Max Turbo Limit, RING: Max VR Voltage, ICCmax, PL4
CPU Package Power 260W, IA Cores Power 250W, PL1 Power Limit (Static) 253W, PL2 Power Limit (Static) 253W
You say "VRM is set to 1 mOhm" but obviously you mean that it has that physical property apart from LL trim.
Its not one game per Server bro, use of affinity etc.. run one game instance per perf core
Just curious why you don’t like LGA1700? Personally I would prefer am5 x670e/x870e for more pcie lanes.
because it's power hungry and the memory controller is a slot machine.
And you need to buy a contact frame for it
Im having trouble with My i9 14900KS I wish you can Make a video on an MSI Board
Can someone explain to me in a few words what I should do, I have a new i7 13700k that has been used for a month, I have not updated the BIOS on my motherboard, it has a 2023 version, it is a Gigabyte Aorus Elite AX DDR5, what steps should I take now with all the information available about the problem?
trying to mitigate the degradation what I did at the beginning was set the consumption limit from the bios to 200w maximum there the cpu does not exceed 80 degrees and a maximum frequency of 5.2 to 1.285v at rest it reaches 5.4ghz at 1.370v maximum I don't know if that's a lot of v for that i7 13700k cpu gygabyte aorus elite ax ddr5 motherboard memories at 5200 I have not updated the bios with the micro code.