Clevo Overclocker's Lounge

Mostly likey either the CPU is too hot or the voltage is too low... or both... a that point near the end of the benchmark. Try bumping the voltage up a little bit and there is a good chance that will take care of it unless it is already way too hot. If you're not sure use a monitoring utility to see what the CPU max temp reaches during the benchmark.

 

@Mr. Fox @Phoenix @Johnksss@iBUYPOWER @lctalley0109 & other having the RAM issues.... I didn't go "crazy technical", but finally got around to re-testing running TM5 on my 2x16GB G.SKILL 3000MHz on XMP1.

When I first tried to run it, halfway though I got an error... 42x4 CPU, 42x Cache, (-100mV adaptive). This happened both with SMBus enabled & disabled.

Spoiler

Accidentally hit the screen lock key trying to ctrl-PrintScreen for the snipping tool...

So I tried various combinations of running HWiNFO64 as admin/not admin, max fans, overclock fans, rebooting and couldn't get it to pass anymore. I realized that when I previously ran the TM5 successfully that I never ran HWiNFO64 simultaneously at all, so that could be some contributing factor. Also, I also have recently been trying out a lower overclock/undervolt @ 42x4 (-100mV adaptive) than I was previously doing when I tested TM5 at 45x4 (-50mV adaptive).

So I moved my clocks to 42x4 (0mV stock adaptive), and it finally passed with SMBus disabled.

Next I tried back to my 45x4 (-50mV adaptive), with SMBus disabled... Passed again!

Spoiler

However, at 45x4 (-50mV adaptive) and it would not pass with SMBus enabled...

So I went to 45x4 ( -0mV adaptive) with SMBus enabled.... PASS!




So based on these observations, I'm thinking that HWiNFO monitoring adds some necessity for extra voltage on the cores and/or cache. Because I have previously run TM5 standalone without any monitoring successfully at lower voltages, and upping the voltages resulted in successful tests. In addition, running HWiNFO with SMBus enabled requires even more voltage than with it disabled.

I'm not a "low level" electronics person (though I would like to be) but could it make sense that some of the hardware components that HWiNFO utilizes within the hardware requires more voltage to be "stable" (without errors). Because it seems without the overhead of the 1 second interval hardware monitoring less voltage is required to "successfully" run a TM5 test.

During all of this testing I didn't alter the RAM timings at all, and only rebooted once early on in the middle of many failed tests since I had been working all day on the laptop without a restart and thought that could have somehow been throwing off my ability to pass a test.

So maybe the issue is somewhat CPU/Cache stability related and not totally RAM related as far as TM5 passing testing??

 

Yea, it was right at 10 hours... 4 passes

Spoiler

 

make sure not to mix up different memtest tools, there are at least two that im aware of

 

Well, that was a waste of time. I used CLU on CPU and GPU and also applied CLU to the heatsink, however seems the heatsink still did not make contact and both the CPU and GPU were overheating. idle temp for GPU was 60-70C and CPU jumped to 100C as soon I started a stress test.
I changed CLU back to Coolermaster Mastergel Nano and temps are back to normal.
Funny thing is the CPU temp is 1C worse than it was when I just used a tons of Mastergel paste and did not even care (when I first tested the delid). Oh well, I will not take it apart again just to gain 1-2C.

 

Well at least you gave it a shot. I really think the only big difference you would see is in the delid anyways.

 

During all my testing I was able to get stable at different clocks by just tweaking the adaptive voltage. Really strange but for instance at 4.3 ghz I can run games, programs and Aida 64 at -140 but with testmem5 I had to go all the way to -80 adaptive before it would pass a few times in a row.

 

Different tests stress different parts of the core which will have different frequency vs voltage curves depending on loading.

 

Man, I can hardly wait for the @Prema vBIOS...

http://www.3dmark.com/3dm/16685920

 

http://www.3dmark.com/fs/11085948

 

http://www.3dmark.com/fs/11086024

 

i think its just simple to say that faster ram, in memory test will be harder on the cpu, hence need more voltage.

 

the first thing i said was "wow much better quality" and it was definitely not a "plastikklumpat" (austrian term for plastic ****) like the Delid DIE Mate

i thought about that too to apply Conductonaut everywhere but i thought about the contact issue. I can recommend kryonaut, Gelid GC Extreme and IC Diamond 24, last one should be the best for our notebooks cause the paste is the thickest and shouldn't pump out that fast, so far kryonaut didn't dissapointed me either but i heard some pumping issues from other users.

would be a nice christmas gift, right , i can smell the potential of the 1070/1080 in our books but ngreedia/clevo made sure they aren't gonna get faster than the desktop counterparts, how would that look

 

IC Diamond will be best if the cooler is not making proper contact, which seems to be true for a lot of models.
So the time has come to laser cut copper to fill the gaps, this is what i´m doing now, will report as soon as i get results.

Here´s what i think might be my end solution:

Delidded CPU with Liquid Metal - Conductonaut, and re-sealed with high temperature silicon.
Correcting cooler gap with laser cut copper to fill in the space.
Lapping the cooler.

Will soon be able to show results

 

looking forward to those results

Sent from my Huawei Mate 8 NXT-AL10

 

I am at a loss here...I have changed the thermal pads on my VGA and now it is overheating.
I have tried less and more thermal paste too. Less paste made it warm up faster, more made it warm up slower.
My GPU temp is about 12C worse than with the stock pads despite using better W/mk pads and better sizing them so they would completely cover the part.
My GPU temp now tops at 92C after 1 Fire Strike Ultra run...

When just stressing the CPU is is 15C better than it was before the delid.
Just stressing the GPU with GPU-Z render test is tops at about 72C..
However in Fire Strike Ultra teh GPU reaches 92C within minutes.

I am out of ideas, I just don't know what is wrong.

 

are your paste jobs on the gpu side satisfactory? when u take off the heatsink, does the gpu show a thin, even coverage with paste? do you see indentations on the thermal pads, being a sign of proper contact with their respective parts? did you screw down the heatsink properly? carefully measured the thermal pad thicknesses needed? has anything changed in your ambient settings? temperature, laptop location, etc? did you check gpu clocks with stock pads vs. new pads? maybe its boosting far higher now, which would also reflect in benchmark scores.

Sent from my Huawei Mate 8 NXT-AL10

 

Doublecheck if you are using the right pads in terms of their thickness, not that you are using e.g. a 0.5mm Pad where 1mm Pads should be

Gesendet von meinem Pixel XL mit Tapatalk

 

I checked the thickness of the pads and used the pad with the correct thickness on every part (1-1.5-2mm) so contact should be fine.
What I have tried to far is to remove the heatsink and wipe off the paste from the heatsink (making it less to contact) -> 90C but much faster than before
Then I remove again and without wiping it ott I applied some more on the GPU -> 90C but slower than with less paste

I tightened the screws as hard as I can and more would damage the screws.

Pads before applying the heatsink for the first time (still had CLU on there when I made the picture):



Temps:

 

I am also curious about the results.

Especially to see if you can fill the gaps at all?

 

Yeap thats the thing, i need to check it for each model if there´s a gap size that will work for all. Becase if there is then anyone can get pre-cut copper shimms, else you will probably need to cut get them at different sizes, whatever it will be.
This is going to be a lot of work, and Christmas time is probably not the best time to go this deep.

 

well THERES your problem: dont just reuse old paste, especially when wiping it "half off" and mixing it with fresh paste!!! use isopropanol to properly wipe off the old paste, only THEN apply new paste and let the heatsink spread it properly... clean BOTH the heatsink AND gpu/cpu surfaces!

 

I´ve seen at least three broken units from clients that did exactly this. Well i saw all three because they where sent back for RMA, no BIOS post!

On those three units the Liquid Metal actually dripped to the motherboard shorting it.

I kid you not! Probably under a long period of heavy load the Liquid Metal gets more liquid, then probably after some hours of work you if you pick up your laptop and go somewhere, and the position of the laptop changes to vertical (a backpack or bag) the liquid metal will start sliding and dripping to the motherboard.

The good news was that after cleaning the mbo it came back to life, but i think it was just luck, it could have easy killed the mbos.

I remember this was all in P7 units, and it probably happens because there is still some gaps in the cooler.

We never sold units with liquid metal because of stuff like that, specially after seeing it with my own eyes.

 

Some more pictures of the pads, the contact seems fine to me:



@John@OBSIDIAN-PC I have checked for CLU drips and did not see any and I already removed the CLU from the CPU & GPU and did not have any problem with not posting.

@jaybee83 Will give it 1 more try ...

ps: Is there a chance that the CLU damaged the heatsink ?

 

I also tested the Gelid and the Fujifilm 14 w/m K on the GPU.
For me, the same effects have occurred.

The original pads should protect the RAM and other electrical components from heat.
This is at least my experience at the time!
You have to really get an understanding of the functionality with this beast.

I think that the original thermal pad not only protects against heat, but is also electrically non-conductive.
Above all, one does not even get the original pads even in the normal trade.
At Eurocom you can at least buy it.
Thanks to Eurocom!

I only waited so long with my publication because I did not rule out a mistake.

 

um....all thermal pads are electrically non-conductive....

 

If you really want to exclude the possibility of dripping, you need to do the following:

Do a 30m run of AIDA Extreme stability test, pooling both CPU and GPU to 100%.
After that turn off the unit and put it in a vertical position (you probably should do this with the battery out).
Wait for 10m, check for dripping

Do the same test, sit the unit vertical in other position, check for dripping.

Do the same test set the unit vertically and shake to simulate a car drive / walk in a backpack.

Because i dont really know how it happen with the clients, but it did!

 

Will do that, but as for now I am focusing on fixing the overheating

 

Hopefully you do not have to spend to much more time to get this under control. If you have the correct size for the pads what thermal pastes have you tried and what was on it before you put new pads? You may end up being stuck with only using IC Diamond but I think you said in a earlier post you used tape and it appeared to have a good fit so I would think using a different paste such as GC Extreme would be fine.

 

I used Coolermaster Mastergel Maker Nano before and using it now too. It also has nano diamond particles like IC Diamond.

 

I won't use Phobya Liquid Metal or Conductonaut any more for this reason. Too watery even during application. I have seen both of those drip on my own systems and I won't touch either one any more.

I never had a problem with CLU myself and this is a truly amazing product. What you said in the previous post about fit is relevant. P7 are known for poor heat sink fit. As wonderful as CLU is (nothing better) it cannot work unless heat sink fit is excellent.

Using anything except ICD unless you have a solid contact is not going to be good. ICD is definitely best for sloppy fitting heat sinks. Some can perform as well, but they are not as durable.

 

Master Gel Maker Nano is quite good.
I have already used.
But Conductonaut is the best.
Just like Liquid Ultra!
Since you only have to protect the GPU - so that the thermal paste does not leak.
Very important!

 

Contact of pads and paste:

 

I'm sure you're probably grounded off camera. But be careful with the exposed electrical components near that blanket, could cause some static electricity discharge from you to them, and you'll have more than just temp issues on your hands...

 

thats quite a lot of paste, but contact seems to be fine. also, vRAM indentations in the thermal pads are a good sign, same goes for mosfets on the top pad. i might be imagingin it, but the gpu core seems a tad exposed on the very edges of the corners. also, as mentioned before, after each heatsink removal, properly clean off every last bit of thermal paste residue (both heatsink surface and gpu/cpu surfaces), then apply fresh paste on cpu and gpu (cross method recommended).

 

That is a big amount of paste!
You should use a lot less. And i can assure you that your system is not making proper contact, that amount of past would spill the hell out if it was making proper contact.


Now if you really have problems with the temperatures here´s what my experience as showed.

1st is there proper contact?
The poor guy way of checking for bad contact is to apply a VERY small amount of paste in the center, bolt it, tight, then remove it and look at the cooler.
Did it spread well? If it did then you probably have no contact problem. If it did not then you have a contact problem.

If you have poor contact, you can get away with adding a thin sheet of copper, 0.3 to 0.8, need to test them all. Try to lap the cooler before applying the shims.

2nd
Delid the cpu might help too, there i have no problem going crazy and using liquid metal, why? because you can use silicon to close it and it won´t leak.

3rd
Now about the cooling system design, there is a fundamental problem for me, which is, you can´t isolate the CPU from the GPU, there are two heat pipes connecting both. My problem with the current design is those crappy console port games which will pull hard on the CPU, eventually the CPU heat will go to the Graphic card, making it hard to keep under control.

As for the CPU there´s no way it won´t throttle without a small voltage offset, my test system with -120mv offset and laser cut copper shims + delided CPU with Kryonaut (not liquid metal), pooling 100% CPU and 100% GPU with AIDA Extreme:

Automatic Fan Profile- CPU- stays at max 89ºC
Automatic Fan Profile- CPU- stays at max 86ºC

Now there´s also something else that keeps busting my head, the automatic fan profile is slow reacting to the temperature and never goes to max speed.
The Overclock fan profile will get you way better temperatures but it´s just too sensible for me.
So the EC could be better tuned.

Also the same CPU can work very well at -150mv which give you way less heat. And remember the lower the temperatures you get from the CPU the lower it will be for the GPU also.

If you still get throttle in the GPU you can also try to create a profile for gaming in throttlestop, one without turboboost for example, without turboboost some CPUs seem to work fine even at -220mv and less.

As soon as i finish all testing i will post it.

 

So this is part of what I've been doing. I shaved a little of the bottom of the lid too so that it can get in a touch closer to the die to compensate once metal goes on. You'd be surprised how concave the IHS is out of the box too.
Just waiting on some more CLU to drop and I'll try for a full metal job.

In other news - For anyone wondering don't bother with the Liquid Metal pads. I'm using an old one I had one lying around as a placeholder under the Ultra arrives. They're barely useful even directly on the die to the IHS.

 

Is that the other IHS?

 

From what i´m testing on my side you do not need something as thick as that. But yeah that seems to be the way to go.
I´m just hopping to get a one size to fit them all so i can offer this free to our clients and keep the production time reasonable.

 

Haha yeah it certainly is. I have a new heatsink and they are fitting real flush. Fingers crossed they tighten up the same.

I'm very interested to see your solution

 

California mandates energy-efficiency standards for computers. Come into force in January 2018

"The standards, which come into force starting from from Jan. 1, 2018, focus on the performance of computers in idle, sleep and off modes rather than putting limits on when they are in active operation, said the California Energy Commission, which on Wednesday adopted the new standards."

http://www.pcworld.com/article/3150...nergy-efficiency-standards-for-computers.html

What next? Putting limits on when they are in active operation? Maybe from January 2022?

 

thats fine though, as long as a device aint really doing anything, no harm in restricting its power consumption to the minimum. no need to restrict performance during active times though

 

Intel and Nvidia fixes this alone. + OEM's Just add a new Turbo Boost 4.0 or 5.0 for future gen Intel i7 processors. And Nvidia submit new algorithm where the maximum boost reach max around 36C on the graphics card. Higher temp will decrease the performance as now, only much worse
Edit. BTW. Passed 8000 messages. Hope all posts have been read by the majority in our community

 

If you do this, remember that with the P7 series to check the ram area for shorting. When it dripped out, it went onto the memory lines connecting the slot to the motherboard.

Mine was my own negligence. I lapped to great contact, then benched in sub-zero ambient. When I later went to re-paste, some CLU hardened. I cleaned it up, but did not relap the surfaces. Some of the hardened CLU made a gap for my later application which allowed for it to slip out (hence my own negligence)! After cleaning the CLU off the lines for the ram, booted fine. That just as easily could have fried my ram or the board!

This is why I agree with @Papusan, use pressure film to check contact and do not ignore if contact is not perfect on the edges. Because I didn't want to relap to a perfect plane, I used indigo extreme and am the only person I know that got it to work. Didn't work as well as CLU with perfect contact, but it worked better or equal to gelid extreme. Note that IX biggest issue is that people cool down the heat too fast for the proper alloy. After the dip in heat, it needs run for a couple minutes after that reflow to fully melt the metals. Then, you need to shut the computer off without allowing the fans to fast cool the chip (so disable C-states before starting and don't allow the chip to go under your manual multiplier settings). I can't remember if I just held down power to shut it off with the stress test running or not, but I did use auto fans so they would not cool too fast with max fans. I then immediately wrapped in a towel and let it set to the side for 30 minutes (I think I started a show and waited 45 minutes to an hour). It is a safer alternative if you are new to liquid metal. If the gap is too great and pressure too low, this won't help you and you may only get partial coverage, which means a shim is your solution unless able to vary the height of the mounts lower or willing to do other mods.

Either way, be careful as it can be easy to apply too much downward force and damage the die (if delidded) or crush the socket or damage it.

Sent from my SM-G900P using Tapatalk

 

Did you read about Ryzen yet and its features (and head to head handbrake session against a 6900K)? I want to know more, including the cache prefetch, more benches, their 25mhz increment boost feature etc. Thought I would mention the last point since you mentioned boost...

Sent from my SM-G900P using Tapatalk

 

I have read only a little. It's time AMD "MAYBE" can achieve something. But we have to wait and see. Not all that glitters is gold

 

Another theory I have is that these new pads are not as soft as the factory pads and because of this despite having the correct sized pads, there might be a small gap when I put everything back together because at least one of them. I might even have a guess that it is the 2 pads at the "9" screw.

 

Of course right now they are doing a press push because they need a win. Handbrake can be slightly better than the blender test, but we still don't know enough. I found their auto overclock interesting as it is supposed to scale to your cooling solution (including ln2, although the articles and I suspect custom tuning to give the best results). But as you said, there is a lot more to it than their press materials and time will tell...

Sent from my SM-G900P using Tapatalk

 

This is also what I would guess based on your paste spread.

 

One of the reasons that the softer pads is preferable. See results from this guide from @jaybee83 14 W / mK Fujipoly / Alphacool pads can beat 17 W / mK Fujipoly / Alphacool pads http://forum.notebookreview.com/thr...e-recommendations-update-thermal-pads.796820/