Clevo Overclocker's Lounge

Dont forget about easy 5.4Ghz+ validations if this worked as it should.

 

Because @Prema magic was still able to remove all other sources of power throttling. The problem is the 91W TDP limit can't be bypassed unless an actual power profile exists for the 7700k + 1080 system, due to how MSI coded their cancer firmware. Also Mr Fox's EC was using a value of "B3" in register E3, while the stock 16L1 uses "33" for GTX 1080. But both "33" and "B3" don't seem to be linked to a MSI power profile, so Bios Power Limit (PL1, PL2) overrides are ignored (rather, overridden by PECI power limits enforced by the EC). Maybe MSI left it blank on purpose or maybe they were incompetent. I don't know. Or maybe they felt "no user will ever exceed 91W TDP so we won't even bother..."

That's pathetic when a 7820k turdbook can draw 110W at 4.5 ghz...

 

There are still board level limits for the voltage.

 

If it were made correctly and a good chip, and a good motherboard with good firmware, it probably would not do that. I think the high wattage is more like a symptom of having to run it with all of the settings needing to be goofed up to trick it into overclocking sort of OK.

Hard to say. I'd lean toward deliberate, but there is so much incompetence and lack of regard for doing things right with notebooks it could be a little of both. The end result is the same whether ignorance/accident or intentional.

 

@Mr. Fox no it's not that.
The power draw is correct. 105W is 8 thread small FFT prime 95 FMA3 load.
Try that on yours and you'll pull just as much.

The problem is the "AC/DC loadline setting in the bios that's hidden from normal users.

However this is only on FIXED settings that normal users can't access (setting AC/DC loadline to 1 manually).
With the MSI automatic settings, something is completely broken.

if you use the MSI default settings, the system just shuts off with small FFT AVX load (>4.1 ghz).

Intel reference for "AC/DC loadline" is supposedly 2.10 mohms. But if you set it to 210 manually, or use the auto setting, the power draw is through the roof. It works fine IF you use adaptive vcore (the default VID is extremely low), and this setting seems to be tuned for "automatic stability" settings--adaptive vcore--stable automatically (e.g. 4.5 ghz overclock), system shuts off during AVX. However if you set AC/DC to 01 (the proper setting for desktops when using manual vcore), Adaptive vcore instantly BSOD's at 4.5 ghz (as it should), as the VID is actually reported correctly and 1.08v (because of vdroop which isn't shown as there is no vcore sensor). The problem is, the VID is not reported correctly at all when using Automatic AC/DC loadline.

Explain this screenshot (next post as it is on my laptop).

 

This is ac/dc loadline (in Intel i/a domain) at 1 (0.01 mohms), 1172 mv voltage in bios. power draw normal avx disabled in prime.
(AVX can be enabled in prime here. it will 'drop a thread' at 4.4 ghz unless I keep the ambient temps down. This is expected behavior). NORMAL proper laptop behavior. Temps settled at 76/74/77/75 after iteration 12k finished.


This is ac/dc loadline at auto (the default MSI setting that is hidden to normal users), 1135mv in bios. AVX is disabled still. see the huge heat and power draw? the vid does not correspond to what is shown. This is JOKEBOOK BEHAVIOR. and these temps were STILL RISING!! it would have reached 95C or more by the time the iteration finished and switched to 8k...

No wonder the system shuts off instantly if AVX is enabled under these conditions. It could be putting 1.3v or higher into the cpu, but its hard to say without a multimeter.

 

When we removed all limits on the LGA MSI, as it is SOP on all the PremaMod Clevo's, the system hardware couldn't handle it, power starved and shut down in the 5.1-5.2Ghz range.
We decided to give the extra power to the GPU rather than killing end-user systems out there for 200Mhz more on CPU...

 

And I thought becoming a chess player would give me a fun life
Modding systems, overclocking and building your own firmwares is where it's at !

Thank you Prema.

*Edit* going to edit my above post to show an avx 4.3 ghz load.

 

Still waiting on @Falkentyne to run something other than Prime95 with AVX off.

 

I dont like running prime with AVX because i degraded two desktop 2600k's doing that. Granted one started at 1.38v and slowly degraded cuz i wanted to be in that "cool OCN 5 ghz stable club", the 2nd i was running 1.52v through it. Please understand. I'm on SSI, i dont have a lot of money to waste on hardware like some of you rich people do. I'm not well off. not trying to turn this into a crybaby thread but PLEASE understand my situation.

Ok heres the 4.3 ghz AVX screenie (4.4 ghz is much harder to pass, need very cold ambients).
Notice i'm pulling 102+ watts from a turdbook? This is with MANUAL vcore 1172mv and ac/dc loadline at 1.

IF I HAD AC/DC LOADLINE AT AUTO with AVX/FMA3, the system would shut off in less than 5 seconds!.

Now I need to explain something.

the VID shown seems be accurate when AC/DC loadline is set to 0.01 mohms. And the system works as you would expect, besides the power limit and battery hybrid shenanigans from EC RAM register E3 and possibly other places only @Prema the mighty would know about from the 16L1. System does not shut off.

But heres the problem.
If I use Adaptive (Default), 4.3 ghz with default AC/DC loadline of auto (All msi defaults), small FFT prime will shut off the VRMs at 4.3 ghz in about 30 seconds to 2 minutes. The VID is reported as 1.14v and the CPU wattage is reported as 91-95W.

Yet the VRM's shut off. And temps are reaching 90C also. Higher temps but lower wattage reported and VID reported lower?

Yet at higher reported watts, temps are lower, VID shows higher, but VRMs are happy.

Somethings not right.

@sirgeorge said that he suspected the GT73 is misreporting the VID. It might also be misreporting the power draw too since everyone knows here that the CPU has no direct way of knowing how much power its using. It gets this from the VID (or vcore) and the "Imon" signal from the VRM's. You can set a negative imon offset to make the CPU report its using less power than it is (this is useful for if ur being limited by TDP only rather than by EC enforced total system power +TDP).

 

Ummmm, almost all the cpu benching software is pretty much free. And the rest being your free time.

I'm also on OCN and I do remember all the 5ghz this and that from then till now And on Xtremesystems it was using linX instead of prime95/Blend.

So you running a few benchmarks which last 11 hours and 50 minutes less than you running prime95/small FFT for 12 hours should pretty much be a walk in the park my friend.

Side note:
Ever since you brought up this ac/dc stuff I have been waiting to see it in action. To see if it will vary from my own findings, but since I do not have your setup. I have to wait for someone else to do it. If I had your setup, I would have already done it by now.

 

This is an expensive turdbook, that I need to game on. I don't want to mess it up running stress tests. Making the VRM's shut off is not healthy.
That being said what I'm trying to get at is this.

Intel says the 'reference' for AC/DC loadline (meaning, the Auto setting) for the Z370/8700K CPU is 2.10 mohms. If the 8700k is just a kaby lake with 2 extra cores, it's possible the 7700k and 7820HK is also 2.10 mohms.

Now, all the kaby lake guides say that you should set AC/DC loadline to 0.01 when setting manual vcore. That way, the CPU VID is the same as the target vcore.

Raja @ Asus said that

Remember this does not account for vdroop. You need "loadline calibration" for that, which the GT73 does not have. Instead you have "VID boost".

BATTERY BOOST!! @Papusan

Now with the default MSI settings the average user doesn't have access to, this means, trying to run OCCT (small sample size), or Prime95 (small FFT+ FMA3+AVX) with an overclock (without a downclock/negative offset) is going to shut off the VRM's, or give absurd high power draw, and the VID (or even the wattage) may be reported wrong. I'm sure that if any GT73 user sets their CPU to 4.5 ghz, sets ICCMax to 200 amps, and tries prime95 small FFT, without access to the IA AC/DC loadline setting, the system will shut off in seconds. OCCT small sample size will fail instantly or shut off instantly.

Using a manual vcore and auto AC/DC (default) is even worse. Instant shutdown.

Setting IA AC/DC to 0.01 (the basic setting that SHOULD be used) with stock adaptive voltage will cause an unstable CPU (BSOD) with a high overclock. To compensate for that you will need an offset or manual vcore.

But the VRMs will not shut down. VID and power draw will be reported accurately.

Same for manual vcore. Only thing not reported is the vdroop from vcore monitoring sensors as it doesn't have one.

This chart by @sirgeorge shows you something is messed up with the power reporting.

quote by sirgeorge:

 

Nevermind.
Not sure why you posted all that stuff again, but it's cool. I'll figure it out later when I get time.

You ran 12 hours of prime which could do far more damage to you system than running something simple like cinebench that last all of 1 minute

Thanks, but don't worry about it. It's all good.

 

That was years ago, like 6 years ago.
I don't run 12 hours of prime on my taptop.

 

So, let me get this straight. You are fine with asking people to attempt to fry their systems while you are not?

 

Where did i ask for anyone to fry their system??????????????????
Do you know what I hate more than anything else?

Someone who falsely accuses me of something I didnt do.
My abusive stepfather did that to me. People have done that crap to me my entire life. People do that to me in my videogames. I dont need it from YOU.

 

Every time you ask someone to run a Prime95.

All i'm doing is trying to understand.

 

You kept bugging me to run prime with AVX instead of non AVX so i did and now you're flaming me.

Are you seriously trolling? I'm 2 seconds from putting you on my permanent ignore list.

 

P95 ain't a good software on laptops. Bro Johnksss work with them every day. I'm sure it was intended more as a joke. He of all, don't and never will advice people to run P95.

 

Have at it.

I never once asked you to run prime95 anything. You need to pay attention. I said Wprime155 which has nothing what so ever to do with prime95.

So ignore on! I'm done.

 

Yeah, i'm done Bro. He just doesn't understand and I can't seem to explain this simple thing to him.

 

And I'm not talking about simple things. I'm talking about "shenanigans" in the MSI firmware which @sirgeorge found out, that shouldn't be happening. Like "VID boosting" and reporting wrong VID and showing the results of that (VRM shutoff). that's all.
I'm not talking about stress testing.

If i want to stress test ill run prime with AVX disabled 8 thread, because that's' something games can actually put similar load on.
That's all.

 

@Falkentyne - run wPrime 32m and 1024m. Set the thread count to match your CPU.

Also, run Cinebench R11.5 and R15.

All three of these benchmarks will be excellent stability tests and actually give you results to measure with.

All Prime95 does is makes the CPU stress and get hot. You don't benefit in any way by running Prime95, or AIDA64 or OCCT stress tests, especially if it lasts longer than Cinebench or wPrime 1024m.

 

Like IBT/Linx (another storically avx based software) P95 is useful just to test out max icc (wattage) draw by extreme test vrm or bios settings (to check of limits are removed also).

There is no other cpu intensive software that push so hard.

Also tested while running a 3Dmark is really stressful and check system stability in 15 seconds ^_^

 

You know. All I wanted was to see how AC/DC effected a 7820HK. Don't even know where all this extra stress testing stuff came in at. You and I both know what AC/DC does and how it affects the system, but at the end of the day the max voltage still seems to be 1.375V for a 7700K in a P870DM3. We can make it look like it's 1.175 or 1.51 but either way you look at it you can not get any more overclock using it. You in the end are just manipulating the numbers to show up different in cpu or any other monitoring program, but not manipulating the performance of the cpu. It still comes out the same in the end. Whereas on a DT board, you see the difference.

That's what I was after....but then you already knew this.

 

Sorry but I get that AC/DC works like a LLC, limiting vdrop and similar. Or, on my Tornado F5 acting like this.

The trick you're refferring appear to be when use IA VR domain "+" or "-" and the wattage result lower(or higher) only on load (on monitoring software but not in a 20$ kill-a-watt meter )

 

Yeah, I don't know where that came from either. Seems to be popular now, especially with people that have BGA turdbooks, (which is totally counter-intuitive since the CPU can't be replaced if you ruin it,) to run Prime95, OCCT and AIDA64 stress tests. I don't get it. It they did this to animals they would be arrested on cruelty charges, LOL. Conceptually, it's similar to throwing a cinder block on the accelerator pedal of your parked car to test how long it is going to take for the engine to blow.

Yup, on the voltage limit. I took the 7700K in the DM3 as far as it can go, just as you have. End of the line for number-chasing. Impressive compared to other laptops, but not quite what 7700K is capable of in the hands of a talented overclocker installed in an enthusiast desktop board with full voltage and better cooling.

 

It's all this Tripod mess last year who is one the culprit. The trend started to increas when Dellienware went further with their sailing down the drain.



Yeah, very popular by Turdbook owners. They should rather stop buying Trash!!

 

Let me put it to you like this.
1: Vdroop does not drop if set to 1
2: Vdroop drops if it's set to 0. Drop on my machine is just under .020Mv.
3: Watts drop if you leave the voltage alone.
4: If you lower the voltage with AC/DC set to 0 you get the same effect, but slightly higher Temp.
5: VID is locked at what ever voltage is set. That goes for idle if you allow it to idle down.
6: Stock voltage is 1.2346V under load & 0.8048 at idle. Locked
7: Stock voltage moves about .001Mv on load and idle. Locked
The part in question.

1: Now that LLC/AC/DC is locked or so it would seem. Now the Vdroop jumps from locked voltage of (Difference of 0.34Mv)
1.315 to 1.349 @ 5.0 Ghz.
1.380 to 1.412 @ 5.1 Ghz
1.380 to 1.412 @ 5.2 Ghz
Then it gets really unstable. The exact opposite of what it's suppose to do.

This is why I say it does not work.

 

Thanks man!

So I've to Go to do some maths and trial, see you tomorrow and thx again

 

No Problem.

And I just did all that quick testing about 30 minutes ago.

Side note: I had the voltage locked at 1145 in bios and TS. And this voltage can range from 400Mv difference.

 

Good LLC settings would be the the thing that would most improve the results of the desktop notebook series IMO.

 

I don't get why CPU LLC is so bad. Nvidia GPUs since maxwell have had near flawless hardware level LLC. For some reason CPU vrms only check the vdroop across the buck converter inductors instead of the full vdroop all the way to the CPU core. This leaves it up to the BIOS to add a software level correction, which is the LLC settings. GPU VRMs simply read the GND and VCC vdroop from the backside of the core and avoid all this crap.

I don't know if it is an Intel restriction, or mobo manufacturers don't want to use a modern VRM for other reasons.

 

Futuremark Celebrates Newegg Partnership with Huge Discounts - $5 for 3DMark
PRESS RELEASE by Chino Wednesday, November 8th 2017 19:07
"Futuremark, the developer of the world's most widely used benchmarking software, today announced a new partnership with Newegg, the leading tech-focused e-retailer in North America. The partnership sees Newegg complement its comprehensive selection of PC components and complete systems with Futuremark's popular 3DMark, VRMark and PCMark 10 benchmarks. It's a winning combination: everything you need to build and benchmark a new PC in one place." RUN AND BUY!!

Sorry the of topic below... I was out for a walk with my dog IVAN as usual in the forest here home. I walk the same places every day several times a day. Yesterday I saw a single moose calf come close to me and I could almost touch him. Wasn't afraid at all. Today night I walked the same trip and sat down on a couple of timber logs as I do. I tourned my head arund and saw 3 moose calves together with the mother walking against me. Dang!! They Stopped 10 meters from where I usually sit. My dog was sceptical but he went to them and sniffed around. The big mama watched everything closely. This is very rare to see even here home... Almost never happen before that Moose can get 3 moose calves. Over 25 years ago last time this have happened here where I live.

This is the place (same as in all pictures) I usually goes. This pict show almost daylight (around MIDNIGHT 00.30) in the summer...


Now its soon winter here ... Same place but winter time means fully darkness... I use my phone, so the pict is what it is.

Spoiler: Moose Mama and the 3 calves

 

Nice pictures. Thanks for sharing them.

I like that about Alaska as well. Long days (never dark) during summer and long nights (never bright sun) in winter. Also good that it is nice and cold for benching a laptop on the patio. I miss that about living in the Pacific Northwest as well. Even the coolest days of winter in Arizona are not actually cold (unless you ask the people that like hot weather, and they think my air conditioned house is cold during summer,) so never a very good place for benching outside. Laptop stayed colder on the AC unit in the house, LOL.

Are you allowed to hunt moose? Moose and elk are my favorite meats. Better than deer meat, and a lot more of it. Moose and elk can be very dangerous animals, especially a cow protecting her calves.

 

Moose/elk hunt is allowed. From Sept-Dec. But i'ts damn costly, + you need proper licence and a team. Not for everybody all places in Norway. Moose/Elk hunt is almost a National sport in Norway Thousands of hunters when the season start. And yeah, the meats from those animals is one of my favorites. All should try it.

The mama is a lot more dangeress in the spring when the moose calves is smaller. They are bigger now and Moose mama isn't equal angry if you go between them. But you never know. It's animals. And they are BIG. And they are fast. I have been in between the Cow and calves before. Not nice.

 

Quick question, I'm about to re-paste my CPU and GPU as I have too much on it (based on images from the last time I shared). I'm going to use my CPU shim as well. Do you recommend a line down the center, using the "X" pattern for the cpu? For the GPU what is recommended there?

 

It does not matter as long as coverage is complete. A little too much is far better than not enough. It wastes a little bit of thermal paste, but otherwise causes no harm.

I used to use an "X" pattern, but now I use a spatula and spread a thin layer across the CPU or GPU die. I like that method best, but this is almost all entirely personal preference.

 

http://hwbot.org/submission/3708193_papusan_wprime___32m_core_i7_7700k_4sec_12ms?recalculate=true


http://hwbot.org/submission/3708195_papusan_cinebench___r15_core_i7_7700k_1120_cb?recalculate=true

 

X with pushing down on the heatsink during install should get a very nice coverage

 

Very true, nice coverage and there is less air pockets which creates a strong vacuum. I have noticed this when removing the heat sink too.

 

It usually ensures and even contact for the thermal pads too.

 

So would you recommend this for cpu and GPU?


Sent from my iPhone using Tapatalk

 

The X method with CPU and GPU when using a desktop I7, it's different for the CPU direct die, that's just a thin line.

 

Just got a prema vbios that unlocked voltage and core clock on my 970m. Some weird things happened. First of all, my score in heaven is higher with a +135 core +400 vram +25mv then +380 core +0 vram +200mv. Why is this? Does heaven place more emphasis on vram then games and stuff? Secondly, it seems like I can't push the vram up to +400 anymore, even at +200mv. Does the overvolt apply to vram as well as the core? Is this caused by modded vbios or something else? I'm using nvidia inspector.

 

Check the actual clock speeds rather than relying on offsets. Offsets increase what the clocks are without an overclock. @Prema vBIOS is totally different and you may be running much high clocks with a smaller offset. Probably running higher that original "stock" clocks.

 

8700K 5.3GHz Validation: https://valid.x86.fr/fq4umn | http://hwbot.org/user/mr._fox/

 

Ok I'll check that. Any reason why heaven seems to really favor a higher vram OC then a higher core OC?

Sent from my Moto G (5) Plus using Tapatalk

 

Synthetic benchmarks can have odd loading patterns that when the hardware outstrips the original intention of the benchmark you start getting odd bottlenecks.