7820HK Overclocking + undervolting

There is not a ton of info out there on this. I find my self creating a lot of different profiles to use in different situations with a overclock and a heavy undervolt.

What have you guys achieved overclock with undervolt?

3.6Ghz will maintain stability with -0.155 undervolt. Does not seem very good. But that is with +100Mhz overclock. Any less and it is unstable.

4.2Ghz all cores, and 4.2Ghz cache can run -0.125v undervolt. 24/7 this seems kind of nuts. stays pegged at 88C under 100% load, with no liquid metal. only 50% fan speed.

What are you guys getting? There is such little info on this.

 

Undervolting completely depends on 2 things:

1) whether you are using static or adaptive voltages,
2) whether your Bios adjusts voltage by raising the CPU's "Default" VID at load with adaptive or static voltages.

Most AMI Bioses adjust the cpu voltage (VID, actually) up depending on current, beyond what is needed for baseline stability. This ratio peaks at the highest turbo multiplier (in your case, x39), because the CPU's default VID is programmed in steps up to the highest turbo multiplier, and is designed to work with adaptive voltage. The CPU's preprogrammed default VID (set at Intel from the factory, where it's tested table) is then boosted by a certain amount. When you undervolt, you 'remove' that boost. But you also remove the same amount at idle (Where the boost is virtually nonexistent) too, so care must be taken that you don't crash when loading windows or on the desketop..

Some bioses also do this "vid boost" thing with static voltages (they are not supposed to do it, it's supposed to be ignored when using manual voltages). MSI and apparently alienware laptops do this. eVGA has been tested not to do thiis. I don't know about yours.

I can do a -100mv undervolt (at all Bios stock settings) at 4.2 ghz, but the system will fail stress testing with FMA3 prime95, but nothing else.

 

Is that VID boost connected to LLC setting?

 

IA AC DC loadline.

Directly from Raja @ Asus:

(seems to be the same as Kaby Lake/Z270 from posts I found about Intel coreboot and Kaby lake Bioses, CM238 laptop chipsets may be identical or slightly less, but no less than 1.80 mOhms).

It's basically like having a 1.80 mOhm resistor soldered to the AC (and DC) circuit line.

The problem is, this setting (at least the AUTO setting) is supposed to be IGNORED when using manual voltages. @Vistar Shook 's eVGA Board correctly ignores this setting (no difference in power draw, watts, volts if he uses an Auto value or a value of 0.01 mOhms). But it seems that some ODM's are applying this with static voltages, which causes EXTREMELY high temps and excessive overvoltage. MSI definitely is using this improperly.

 

Everything you see regarding overclocking a 7700K applies since its the same CPU design. 7820HK is just slightly worse quality individual chips set with lower default clock multipliers, much lower power limits, and packaged for direct soldering.

Which is why you're 'overclocking' to stock 7700K x42 all core multi which is achievable with that kind of undervolt or more on a stock 7700K. It's 88C because the cooling is not good enough. It may be a combo of poor heatsink mounting, maybe non performant paste, maybe its designed around a 45W TDP cpu and lacks heatpipe and rad to cool anything more. 88C as a stress test maximum is fine, since it won't go that high in normal use or games (unless it's a shared heatpipe setup and added gpu heat cooks the cpu)

The big impediment to 7820HK above about 4.3ghz is not so much the CPU, but the motherboard its in and the cooling, which is largely dependent on the model of the laptop. Always hard to know if its a chip needing added voltage, or the vrm not being able to provide stable voltage at high draw and needing to 'overshoot' its VID to compensate for vdroop.

 

Yea every time I get the voltage to low the system will hang. Unfortunately it reloads those same settings on restart. So you end up in a boot hang restart loop every time. You gotta race before time and uninstall xtu before it applies the settings again.

4.3 seems ok at -0.085v

4.3Ghz requires much more voltage though, I’m still fine tunning my Aorus X7 V7

ea

 

If you are experienced enough and have an unlocked Bios (or can unlock your own bios menus in APTIO 5, with AMIBCP), then adjusting the IA AC DC loadline setting is far more effective than undervolting, as you will never have an idle BSOD this way randomly (unless you are just totally unstable period).

 

I tell ya what though. Huge difference in 7700HQ and a 7820HK. I’m fortunate I ended up with a unlocked laptop. Some games even at 4K are slightly cpu limited showing 90% gpu utilization when stock speeds of the 7820HK are used, then run 4.3Ghz or better and it’s 99% gpu utilization!

Yea I’m happy with 4.2Ghz. It runs around 78C with 100% fans under a stress test full tilt.

I’m going back to liquid metal paste in a few days. Then I’ll run a nice stable 4.6-4.7. I had to order supplies for long term liquid metal!

Pretty sure Falkytyne has the most BEAST 7820HK around! Lol

He’s gaming at 4.9 I think or 4.8 I can’t remember

 

I was mostly curious of what you guys are achieving.

I’m a desktop enthusiast my whole life. We always add voltage! Never used such a thing lol

What’s under-volting?? Lol

 

These days the chips come pre-overvolted. Probably to ensure stability in motherboards with crap vrms that have heaps of vdroop under load.

 

Easier to compare undervolts and temps between the same model, since like falkentyne pointed out, hard to say how much voltage is actually being applied. The VID reported is not the actual vcore. My EVGA needs +5mV offset @4.2GHz to be rock solid stable, but since the IA AC DC does not affect this system like the MSI, hard to compare. I assume this system has much higher vdroop and thus needs more voltage. The MSI although apparently requiring less voltage and thus a negative offset, Falkentyne has shown through extensive testing that it might be using up to 300mV more than the reported VID. For stock paste @4.2Ghz, your temps are good if those are when under full load.

Enviado de meu Pixel 2 usando Tapatalk

 

I Try to avoid gaming at 4.9 ghz. 1.380v vcore degrades CPU's, even if slowly. Just a few bench runs to make Papusan vomit less (you KNOW how much he hates BGA) and some PUBG to make myself happy. Plus there's the problem with the EC being locked preventing too much current from going into the CPU. It's read directly via the voltage line, so IMON SLOPE doesn't affect it (EC RAM registers D8 and D9). If you draw too much current, the laptop powers off and resets. That's why I can only do 1 or 2 cinebench runs before the EC shuts the laptop off.

@Mr. Fox had this problem as well on his 16L13. I'm not sure if @Prema was aware of it, but its' not a VRM limitation. it's the EC causing it. I wouldnt have known this it it wasn't for @BeastsForever.TheDragon speaking to a MSI engineer on the phone. While I have NOT seen a 16L13 EC, I suspect the exact same code is in it (since a stock 16L13 has battery boost). And even if the Prema God removes all throttling, the entire laptop is coded so strongly around NOS and the EC that this shutoff can't be avoided. VRM's don't shutoff like this. They just explode and melt.

P.S. Mr Fox, this is the same reason why when you set in VR Current Limit, PS Current Threshold 1=512, the laptop shuts off at load instantly.
This value tricks the EC into triggering shutdown (I believe the cutoff is "299").

I'm too demoralized to even contact Xotic PC and ask them to talk to a MSI engineer about allowing me permission to use the unlocked EC for overclocking purposes. I've had enough bad experiences to know that I am nothing but a "peon" to MSI. They don't care. I KNOW what will happen without even trying. They're too afraid I'm going to distribute it or blow up my apartment or something.

And I'll say it again.
I am 90% sure the EC protection is stored in these variables via "Conditions" relating to whether you are throttling or having overcurrent "Trip".

CPU_CrtT = 1.CPU_ThtlT = 1.CPU_ThtlT = 0.SYS_CrtT = 1.SYS_ThtlT = 1.SYS_ThtlT = 0

I honestly feel as if no one cares to help me no matter how much I find out, but if SOMEONE were able to CRACK the SHA-256 checksum at the end of the EC, disassmeble it in Linux, change the two "CURRENT TRIP" flags to 0, recompile it, verify the checkSUMS (plural, folks, PLURAL, there are MULTIPLE) validate to avoid a complete brick (reflashing a bricked EC isn't easy), then MSI cancer would be completely defeated. But who here has the skills to do that, besides Prema ?

Attaching EC 17A1EMS1.109 and 17A1EMS1.108 to this post, even though I don't expect any results

screenshots:


CHECKSUM (SHA-256? OR SHA-1?) at the end of 17A1EMS1.108

E5 E4 B6 F8 59 6A 45 8C 37 C6 AB F9 65 6A D2 B5 92 2F C9 F2 6D A3 2F AB 91 06 9C 7E AC 46 BE F7 C5 2C F6 07 81 43 5A F1 22 BC 9A F1 EB 76 41 23 83 CD 78 CB F6 8A 1D F6 0D 29 49 23 38 6C C6 26 3E 07 2E E5 91 A6 73 BC FB 5E EB F6 0A F6 F6 EB 60 D2 7E 50 01 96 B6 D1 F0 F5 13 85 4C EF AF 7F DB 8D 40 05 FB 30 54 FE 76 0D 0C 6C 49 9F 32 0A 14 6D 4F 42 8C 28 CD F6 C7 38 F4 76 DF 82 FE 8F 43 77 A7 DC 6E B6 E4 F4 A8 20 6A 78 B9 DF B5 CA 87 58 0A 6D 18 75 AA BA BA 22 8A C8 2A B1 EB D6 96 43 3C 8A 43 B8 1A 85 77 8F 71 E1 5F 6D AD F9 AA 62 3B 5F E3 C9 34 27 D4 83 8D F2 B8 26 C0 ED 03 6E 90 18 83 C4 CE 18 F4 C9 E8 CD C9 4B A0 93 B4 0D 2A F3 52 E2 9D 50 82 28 7B 49 83 5B 86 D3 50 AA 52 8D C7 36 88 C8 5A C8 CF 23 85 91 65 CA 68 2C DE 3C E3 C0 CC 3F 29 2D 23 CC C2 8A 6A 36


Checksum at the end of 17A1EMS1.109
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


Comeon people. So many very intelligent and skilled people on this forum get crackin !

 

I'm currently running 4.2ghz with a 110mV undervolt. Temps in pubg are highs of 80c

Sent from my SM-G955F using Tapatalk

 

Wow a few people actually downloaded the EC. Do you mean there's actually light at the end of the tunnel? Hope?
I can't do a 5 ghz Cinebench run. The laptop just shuts off almost instantly. The EC won't allow that amount of current to be drawn at 1.40v+ on 8 threads...

 

Hey are you gonna run 8th gen BGA? Whenever it comes out?

That’s gonna be nice! Better overclocking on more threads.

Maybe a 1,500 in R15 would be average.

 

@tps3443 no I'm done with BGA. Buying an 8 core 16 thread Clevo when the Desktop LGA chips get released and Clevo releases a laptop LGA version and a Prema Bios becomes available. And the new Nvidia cards should be out by then too.

 

So now that I've been begging profusely:

Are there any firmware programmers or anyone skilled with Linux or disassembly or checksums reading this thread?

 

This means +115w Cpu package Power.

 

I thought the Asus Ryzen laptop with the 1700X in it and RX480 was pretty cool for $1,499.

If only it had a Vega 56, or 64 in there or even 8GB of VRAM and a full power rx480.

It scores like 1,400+ stock

 

That sounds good though I may follow suit.

I’m looking forward to next gen GPU, more than the cpu’s. I’m ready for gtx 2070 to come out.

I need more power for 4K.

 

Never been a 95w chips in Asus. It’s the cheaper power gimped 1700.

 

I wouldn’t hate on it. It’s really a test in the market and I think it’s great. It destroys our laptops for a lot less money. It’s a good value, I hope to see more in the future.

I wish I had the power that cpu had.

But yea that’s right, it does have a ryzen R7 1700. But , you know who really cares it’s still cool! It still gets like 1400 in R15 though and slams any bga or 7th gen i7 for that matter. It will even keep up with laptop 8th gens that are pushed.

Sure you can squeeze a 8700k in a laptop to keep up with it but they are expensive, and should be an 8 core it’s self for that much.

Plus, people have dropped a 1800X right in it with undervolting, It’s a socketed desktop chipset with a soldered Polaris 10 gpu

If they’d release another one with Zen 2 in it using less power, with a little better IPC, and a gtx2070 ampere, that would be very exciting, and tempting to me and many others.

 

For high refresh rate gaming and general system/app responsiveness and latency, where single-threaded performance is still king, your 4.3GHz BGA i7 destroys a stock clocked 3.2GHz R7 1700. Ryzen = Haswell level IPC + Nehalem level clocks and lots of cores.

 

AMD doesn’t have slow single thread like people make it. It’s not that bad. A threadripper 1900X at 4.3Ghz (1) core hits 181 in R15 matching a 7820HK @ 4.3Ghz


A 7820HK averages 155-162 in R15 single thread stock. The unlocked 1700 overclocked sightly to 3.7Ghz achieves 156 in R15 single thread.

What’s the difference?

I know intels ipc has always been superior. And I’m not saying it’s not. But AMD is right there on them. And you don’t lose much of anything to say well, I really need 10%faster ipc so I don’t want 50% more multi core performance.


Your talking a 5% difference maybe 15% with good overclocking.

 

Still no one with EC reverse engineering experience? anyone?

 

I’m still in school for graphics programming. I’ll be there soon enough lol.

 

And it works for all? http://forum.notebookreview.com/threads/asus-rog-gl702zc

For the records... I like that the Asus come with socketed desktop chipset. A good and only viable choice. But the laptop is still fully crippled by the weak BGA Gpu package. + the 65w Ryzen chips aka not the X version make it as well as a terrible overclocker. It's a reason Asus went for this choice... The laptop can't handle high overclock with with subsequent higher Cpu Package Power. As well can't handle higher powered graphics.

I won't expect AMD paired with a Nvidia in Asus Ryzenbook. And the cooling in mentioned Asus notebook struggle already with a down crippled (see... 68w) Amd graphics.

Why not compare with a fully working i7 running 4.3GHz ?

Why not compare apples v.s apples ? You talk about a stock clocked BGA vs. oc'd Ryzen

And the 2nd gen Ryzen X (see also 105w TDP due increased Turbo boost) is still behind in gaming.

 

I only referred to overclocking because you did. You said a 4.3Ghz 7820HK vs that stock R7 1700.

I know it’s RX480 is gimped in it. No doubt, what a waste of a good GPU lol!

So what your saying is amd IPC at 4.3Ghz vs intel IPC at 4.2Ghz is within 5%? 181 vs 189.

It’s close is all I’m saying. I’m breaking 195 with my 7820HK. No AMD can do that. But, it’s only because there limited to how high they can overclock to.

Fortunately we have competition again. And I think it’s great! If it wasn’t for AMD we would have (6) core intel 8th gen in laptops. Or a (16) core desktop processor for $799 at microcenter that competes with a stock clocked $1,800 intel cpu

It’s awesome. And it’s not the same old boring tick tick rinse and repeat 3% IPC improvement every year, after year after year.

 

Games show a bigger IPC difference because they are more affected by CCX latency.

 

Update.
That isn't stable in pubg. Kept crashing every half and hoir or so.
Gone back to my safe 4.0 @-160mV

Sent from my SM-G955F using Tapatalk

 

@benson881 Do you have the unlocked/SVET (if msi laptop) Bios?
Do you know your CPU's default VID at all?
The default VID of the CPU can determine how much it can 'undervolt'.
But that requires unlocking bios menus like in this guide:

http://forum.notebookreview.com/thr...-clear-cmos-and-prevent-common-issues.812372/

(can be used on MOST APTIO 5 bioses ( @Vistar Shook did it on his), provided AMIBCP doesn't crash on opening the Bios capsule).

 

@Falkentyne I've got an aorus x7 v7 mate so no unlocked bios

Sent from my SM-G955F using Tapatalk

 

Can you dump your bios ?
with this command (run with command prompt in administrator mode):

FPTW64 -D Biosbackup.bin -BIOS

 

Well.. I really can't find any of IA AC DC on my BIOS that you have mentioned.. I have checked every single menu I have got

well my BIOS is APTIO 5 using AMIBCP

Check it by your self
https://drive.google.com/file/d/1Ji_n0NqiOKUsUtm2cQt_GmLqmUsyx4TH/view?usp=sharing

 

I'll check in a bit.

 

@THEBOSS619 I found it in the bios strings but I don't know what MENU it's in.

its under 0x210 VR Config Enable
0x212 AC Loadline
AC Loadline defined in 1/100 mOhms. A value of 100 = 1.00 mOhm, and 1255 = 12.55 mOhm. Range is 0-6249 (0-62.49 mOhms). 0 = AUTO/HW default. Uses BIOS mailbox command 0x2.

Ok it's not even in the Bios. It's in the STRINGS but they left the options out of the compile!
And i don't know if such options can be accessed via "memory" registers like in RU (you would need to know the exact address it's stored at). Boy talk about "we don't want you to mess with our property" locked down cancer....

 

Thank you at least for your time.. I guess I better forget it for good then until I learn it so well... WOW another mission is to learn how to change strings.. here we go again... Google here I come again

Thank you again

 

@Falkentyne I tried that but it refuses to work buddy. I take it you did mean in cmd prompt and not a special program?

Sent from my SM-G955F using Tapatalk

 

You ever undervolt the cpu and then test for stability relentless like. And you swear it’s good. And it’s like the next day note pad or something dumb proves it’s not stable at all? Lol

 

Can you share your XTU profile?

 

I wont let me upload the xtu profile but heres a screenshot

 

Someone unlocked the bios before on a X7 V6.

He had xmp, voltage, etc. etc. and he was able to run DDR4 2666 xmp via bios. That was Skylake so,

I’m hoping I can unlock mine and run xmp ddr4 2800+ without having to bclk overclock.

 

Thanks I’m gonna try it out.

After running KDC it really puts your overclocks in check.

I need to go back liquid metal. I was moving my laptop around a lot, so I got worried. Now I’m back on standard white paste and it’s just day and night difference much hotter!

On liquid metal 4.5Ghz was like snapping your fingers max temps at 75C in games. Stress test would never touch 90C in OCCT at 4.5Ghz.

And I’d run a single core at 4.6Ghz.

 

@tps3443

Remember what I told you about LM?
If you want complete safety with LM, first, make sure when you apply NORMAL paste and you do a PRIME 95 (this is important, you want BALANCED load) with AVX disabled (disable AVX and FMA3, edit the file local.txt, , add CPUSupportsAVX=0 and CPUSupportsFMA3=0), then do a easy 3.9 or 4 ghz run with small FFT and max fans.

Then check for core temp differences.
if core temp differences is LESS than 5C, you are ok for LM. if it is HIGHER than 5C, you have pressure problem, either bad HS pressure, warped heatsink (this can be sanded, but very hard work and requires OCD attempts) or voodoo magic curse.

Then for LM, for CPU:
2 coats of transparent nail polish over exposed smd resistors.
(Cellulose based nail polish, its important!).

For GPU, 3 coats of cellulose based nail polish. This requires as LOT more careful work because the GPU has a lot more exposed stuff than the CPU. HDET CPU's also have a lot of exposed raised stuff too. If LM gets on these parts, even if a short does not happen, it will slowly eat the solder off and the resistor will just fall off .

If you don't want to use nail polish, then you can use Super 33+ tape.
I prefer nail polish because this gives lower thermals. @judal57 tested this himself. That's why I like nail polish. And looks cleaner!
But Super 33+ tape is the original way and works good.

The last part is the foam dams.
This is what gives some people trouble.
The foam MUST be a VERY VERY light foam and it must be THIN foam and must COMPRESS to the size of a human hair, with light finger pressure!! That's important. This is the foam you need.

Then you cut out the foam in the shape of the CPU or GPU.
like this:



or this:



Then the final part which gives people some problems:
use the same nail polish to 'attach' the foam dam so it doesn't move. Just put some on the corners and sides (6 spots) and then attach it. Then secure the heatsink

That LM won't be getting ANYWHERE after you do that. Worst that will happen is if a large bump happens, you may have to repaste if it gets disturbed on the CPU.

FOAM DAMS WILL NOT WORK UNLESS THEY ARE VERY VERY LIGHT, NOT DENSE AT ALL, VERY FLUFFY AND COMPRESSIBLE!

 

Yes I do remember thanks.

I’m gonna re do it probably tomorrow.

Ive got everything I need to do it again, I can barely hold 4Ghz on all cores in KDC it bounces around 82-91C with a stable -0.140 undervolt. My GPU sits at like 62C-66C lol with +100/+775

It’s bothersome because if it touches 93C which is rare, but it does happen the cpu will throttle and cause the tinniest micro studder so it’s very annoying because you feel it when it happens

I can run it at 4.3 stable in anything else. But kdc just really pushes a cpu.

 

Excuse my ignorance but what is kdc?

Sent from my SM-G955F using Tapatalk

 

Kingdom Come: Deliverance

 

Ahhh.... KCD

Sent from my SM-G955F using Tapatalk

 

Taking your estimates as gospel; A 5% difference is at least a year old platform - 15% is ancient...

That is not 'right there on them' in any meaningful way. Ignoring the real world results that say otherwise too (in certain workloads/workflows).