XPS 15 9550 temperature observations (undervolt + repaste)

Wait, I have get done some Photoshop... I didn´t take any pictures of the repaste cause I wanted to finish it as quickly as possible. Only pic I could provide is:

Spoiler: Pic of CLU

But if you look closely at the brush you could see some metal left (but I cleaned it )

So now I got my temps right but the machine is still getting into PL1 Limit in Realbench after about 12min. Does undervolting have an impact on this?

 

Thank you, looking forward to the tests results.

Undervolting helps quite a bit with temps, i used throttle stop but i heard XTU is fine too.

-140mv should be ok (my limit was -144mv but i run it at -140mv to have some margin).

edit: misread your comment, if the undervolting is too high you system will crash, it will not throttle down.
Could you please post the stress test temps graphs?

 

Never happened, you may as well be a smack head showing off his next hit

 

Excess undervolting typically shows via crashes. Doubt that is the issue but you could reduce the undervolt a bit just to see...

Are you still limiting CPU to 35 watts? That seems like a potential cause.

On my 6300HQ, I killed PL1 by raising Wattage limits/times in the XTU turbo settings.

 

Hm, I reduced my undervolt to -110mV and put back the TDP to 45W. Realbench stresstest still trigger PL1 Limit going to 15W. Temps were ok (CPU about 83°C, GPU 72°C ambient at 80°C).

HWinfo Reports Max Vr Voltage ICC max "yes" , Ring: Package Level Rapl/Pbm PL1 "yes" and IA: Package Level Rapl/Pbm PL1 "yes".

Sorry for missing pics but I won´t take the heatsink away for now

 

Those temps are basically the same as with high quality thermal paste.

If you used liquid metal i'd suggest you to remove it asap since it doesn't show (imho) worth benefits (maybe 2 or 3 degrees) and the longer you keep it the harder it will be to remove later.

 

No, I wont put it away by now. The temp are only this extreme in realbench in other situations the temps got down by 5°-6°C for me. Maybe there is some kind of burn-in of the paste as the pressure of the heatsink is quite low.

I read that this was a problem with CLP and should be gone with CLU (but didn´t see it myself) .

 

Liquid metal should be in-fact more than double the performance of a regular TIM based on the w/mK ratings... It's only marginally better because of a couple of factors...1. There's not a complete even layer between the die and heatsink..meaning some copper does come in direct contact with the die, thus reducing the need for more thermal paste. 2. The better conductivity lowers the overall resistance of the heat flowing from the CPU/GPU to the end of the heatsink fins, thus reducing the overall system temperature and it will even out because the BIOS select the fan RPM based on temperature... yea it may hit a similar temp but the performance is much better because I bet the fans spin way lower than they used to...

 

I forgot to mention, on my 6300HQ, when ambient (vrm temp sensor) hit to 78-79*C, PL1 triggered. That might indicate vrm temps are your bottleneck, and additional vrm cooling needs to be considered.

But first, check your other benchmarks as they should score at least as good as before and CPU/GPU several degrees cooler assuming the liquid metal properly mated.

 

I checked my other benchmarks and the results are better with (way) better temperatures. In fire strike i got 5580 with CPU 67°C, GPU 64°C and Ambient 69°C. Heaven Benchmark now is at 2883. Core temps are even. So it seems the paste is mating fine. But I think the VRM cooling still needs some attention... And Realbench seems to get my machine to ist knees. I´m really looking forward for Philaphlous values!


Edit:
Philaphlous, you mentioned the amount not being enough. I put a nice even layer on the die and an additional layer on the heatsink in the area the die has contact. How do you do it?

Edit: Forgot to mention that the fans are now kicking in quite a bit later. This could be a Problem as the CPU gets hot before they are working at full Speed. I will try to get them running and then start the test.

 

This may be true if the temps were lower but if it hits the same temperatures the fans run at the same speed. (considering the test lasts 15 mins)

I missed the connection between system performance and fan speed.

Also, the last 15 pages of the topic are about realbench stress test, this means that we gathered experience with that test and i can't see many gains there.

Having lower temperatures in other benchmarks doesn't mean much unless those with thermal paste run the same benchmarks and post the results or rockstar post his results with realbench's tool.

 

You're right on the process for applying the paste. You paint it on both sides then mate them together... You definitely don't want to put too much. I believe when you apply too much it might even create a barrier and as it heats up..some does evaporate and dry out...this can cause worse temps over time...So keeping it still as thin as possible is the best method...

I'm hoping to get the liquid metal on this weekend. I have to outline the cooler and die contact so I apply it properly...

back story... I recently found out I might possibly lose my job depending on some future circumstances...so yea. Contingencies are taking up my time right now...so not great. But now I can use my new laptop for a good purpose!

 

I think it´s not quite easy to compare as all of us have different machines, different undervolts etc. So comparing own Benchmarks does make sense as somebody could compare his own values before and after. Before I had Kryonaut and CLU seems to get my temps down about 5°C on load.

 

Thanks for your reply. Then I got it right, puh! Sorry to hear about yout job though.

 

As you know, the benchmarks are just proxies for real-world scenarios. Improving benchmark scores and/or temps could open up bottlenecks one faces playing games or running other intensive software.

The ROG StressTest is a good all-terrain benchmark the 9560 just seems to struggle with. It gets close but not quite. Getting that sorted could open up more bottlenecks, unleashing more performance.

PRELIMINARY CONCLUSION - vrm is area still is ripe for additional tweaks

 

Well if you tested it on the same settings and temps were lower by 5° it is an improvement over thermal paste. If it still throttles down and if it's not vrms' fault than there's nothing else to do.

Did you try setting iGpu to maximum battery life in the power plan settings?

 

This is what I thought too. I still have to look at some games etc. the next days. For me the VRMs seems still to be the Problem...So there is something left to be done

 

This is a problem as there are few (no) before & after deltas posted. But raw scores are still useful:

- Provides some baseline measures for 9560 troubleshooting

- Clearly shows vrm bottleneck still needs to be addressed

- 9550 google spreadsheet in this thread provides indicative deltas for different pastes/undervolting in this chassis

- Deltas may not be so important in the 9560. Undervolting potential & variance in different 9560 machines are both (likely) lower as the Kaby Lake processors are essentially similar to overclocked SkyLake processors

The raw scores indicate the XPS' chassis thermals are being pushed much harder in the 9560...

 

This came to my mind too. And then I thought about people yelling for the 1050Ti or 1060...
How does the razer perform in realbench? Does somebody read something?

 

Yeah i'd be interested in that too, there's no way a 1050ti or 1060 (i imagine at least 20w tdp more than a 1050) work for more than a few minutes without throttling down unless razer's cooling system is much better than dell's.

How did you outline the cpu/gpu on the heatsink to spread the liquid metal there too?

 

I had my old paste on the heatsink and saw the "imprint" of the dies. It´s about 2mm from the black cover. So I got the form of the CLU layer. I read before of somebody drawing it with a needle on the heatsink but I didn´t want to do that

 

I'm convinced after just having it for less than a few hours to play with that if we properly pad the VRM and do the paste right we shouldn't have an issue with throttling... Just have to find the right combination... It seems like the heatsinks are well designed and the heat is really drawn away quickly from the chips...

 

Inside, the Razer Blade looks remarkably similar at a glance. Unfortunately it is also known to throttle under load It probably has a somewhat better cooling capacity, however that gets compensated by the higher GPU TDP.
It uses direct heatpipe cooling with no copper shims welded to the pipe (so no Conductonaut with it). Supposedly so that thicker heatpipes can be used.
If I see right, its VRMs are under a heatspreader mated to the heatpipe.
http://www.notebookcheck.net/Razer-Blade-2016-Notebook-Review.186751.0.html
http://forum.notebookreview.com/threads/repaste-those-razer-blade-2015s-at-your-own-risk.771707/
https://insider.razerzone.com/index.php?threads/razer-blade-engineering-design-and-thermals.4515/

 

So fixed my downclocked CPU issue... Got that taken care of. Also checked out a little of what my GPU can do. Looks like boosts to 1730MHz...gets really hot and then I stop the benchmarks... Memory overclocks to 1850MHz so far no problem! Definitely going to need to check the heatsink seating on it...

 

Something I tried on my 9550 with success was putting thermal pads on the ends of the pipes (by the fans/vents) so they touched the case. It's far away from the VRM area and might increase the effectiveness of the fans. You might want to try that and see if it helps or hurts.

 

That area looks cool in the thermal image, in fact cooler than it need be and surely unused cooling potential, because there is the fan exhaust before it goes through the fins.
I think it might be a great place to lead the VRM heat to with a copper band, as thick as can be squeezed (maybe removing the black thin one stuck to the case bottom, and maybe even by raising/bulging the bottom cover by a half mm with some spacers and washers).
I'm not sure if this band could be simply padded to the heatpipes, or it would be better isolated with thin spacers and an air gap for cooling by the vents. If the VRM heat is the bigger issue, I guess isolated. With a pad you likely draw a bit more heat from the HP.

 

Here's my Extreme HD +200memory overclock Valley benchmark score and temps...


Fans do ramp upto 4900RPM which is pretty high... I'm hoping to check out the paste tomorrow morning and hopefully be able to fix some issues....


So that's a +200 on the memory


It should technically be "free" performance gain since memory overclocking has little to do with the requirement for additional power so there shouldn't be much heat generated from the overclock... Haven't gone more yet.

On stock clocks for the GPU core I'm bouncing around 1770MHz down to around 1720MHz by the end of the test. GPU never goes above 70C so that's promising thus far! CPU is undervolted to -0.100 and iGPU is -0.050

Pretty promising so far!! Really loving the new laptop! Bottom of the frame around the VRM isn't really getting warm so my guess is the thermal pads aren't quite thick enough.... I believe I have 2 pads stacked which are 2mm each, so 4mm total. 5 or even 6mm might be your best bet...

 

Thanks for testing, could you do a basic preset testing on unigine heaven (not valley) to compare it to ours?

I have 2 stacked 1.5mm pads on vrms + another 1.5mm pad that covers both the vrms pads and the chokes for a total of 4.5mm and it touches the laptop cover for sure since it leaves some outline.

 

Sure can this weekend. Stock Valley was 30.6fps and memory oc +200mhz got me to that 32.2 or whatever the number is. I had some issues at +300mhz for the memory but I don't think it was related to the overclock...I think I was throttling since the cpu wasn't undervolted.

I'm pretty sure I don't need to do anything with drivers anymore...cpu is running great and according to benchmarks my performance is right there. Did a quick test of bf3 for like 10 minutes this morning and I was a sold 60fps with no drops at ultra spec.

One thing I am bummed with is how you can't oc the monitor... I was hoping I could get more hz out of it but Intel won't let me...I kept my desktop monitor so I know that'll get me 85hz.

Gonna check the seat of the cooler tomorrow morning and figure out the pads for the vram hopefully... If I can figure out the pad situation I can do my clp paste so I'm planning on that

Few things I will say after reviewing my temps...
1. Not sure where all these "ambient" sensors are from? I've read the DIMM is the ambient sensor near the VRM... It's just a guess...but there are 6 "ambient temp" and Temp3 sensors... They might individually be each VRM on the board...or maybe not since there's such a wide variation in temperatures.. anywho. Looks like the PCH (Platform Controller Hub) definitely needs a thermal pad to the aluminum frame...In all pictures the chipset is bare with no heatsink or thermal pads... Since the TDP of the PCH is so low... I think its like 4w, this shouldn't cause there to be much of any additional heat on the frame...

Since the thermal pads I have are so squishy...(silicon).... I'm going to trippple stack them...just hope this doesn't increase my cholesterol...

Going to whip up some dough...no joke... as a measuring thickness putty for the VRAM thermal pads... according to this pic the VRAM toward the fan need a lesser thermal pad because it lifts the cooler slightly off the GPU... I do have pretty high GPU temps so I'm wondering if this is causing an issue... we'll find out tomorrow

Spoiler

Sent from my iPhone using Tapatalk

 

You guys can follow my photostream on flickr. I'm likely not going to post every photo I take but at least this will give you guys an indepth idea of all the pictures you'd like and then some...
https://www.flickr.com/photos/30149337@N04/albums/72157680615207736

 

Thanks for all the details and good luck with modding. Please keep posting results...

I also would guess your vrm thermal pads are too thin. That is difficult to measure as the mosfets are tiny and very far from the case bottom. You will see earlier posts where several of us tried the vrm padding (choke only) and it made both the case bottom & keyboard very hot. With 17 W/mk pads on 6 chokes (only), the case bottom got too hot to touch after just 4-5 minutes and the keyboard was uncomfortably hot. Very low rated pads seemed to work for two guys.

As you know, on p. 80 we have baseline scores with Heaven 4.0 basic. And I don't think anyone has gotten a 9560 through 15 minutes of RealBench StressTest yet without throttling. RealBench is a laptop favorite as it hits the both the CPU and GPU so highlights thermal issues.

As you mentioned, some "alternative" vrm cooling like the copper to fan base might help with the RealBench issues people are seeing...

The hottest ambient sensor is likely the vrm as you know. GoNz0 took some very cold air and identified only one ambient temp sensor in the vrm area. He took a picture of it and posted it here so you know what it looks like (H56 on the 9550). He said the newest version of Aida64 properly identifies the 9560 vrm sensor.

 

Yea, I'm planning on using my can of compressed air to check temps around the motherboard tomorrow morning. Along with using my infrared thermal gun to double check temps when its benchmarking...

 

Check out my Flickr album. I updated it this morning with a ton of photos. Looks good so far! Temps way down! GPU still needs CLP or some type of liquid paste but the CPU is definitely fine. Haven't hit 70C yet on the CPU...fans still ramp up but I think I fixed any type of throttling from VRM overheating...

https://www.flickr.com/photos/30149337@N04/albums/72157680615207736

 

Thanks for your pics! Did you put paste and pads on the Vram? On your last picture it looks like there are pads under the heatsink.

 

That's with +260 on the memory.

So something is up with Battlefield 1.... At first I thought it was my overclock but now I'm pretty sure its not... I see my system memory go all the way upto 99.7% used then it crashes... lol wut Good job dice.

Yes, I put paste on the VRAM chips... if you check out my album I checked the tolerance with some dough..no joke. Made a french baguette...lol jk. But it looked like at least 2 of the chips were so tight, even a 0.5mm pad might not do it...so I just used paste... 1 was really thick so I used the pad and paste on each side..the one over the heatpipes was uneven so I had to cut the pad and use paste on the rest.

 

That looks like a lot of paste on the CPU/GPU, did you take the heatsink back off to see how the paste was distributed?

 

With Gelid Extreme, I found best temps with very little thermal paste, about 1/2 grain of rice per chip.

 

Still could be undervolt-clocking issue . . . Try removing completely just to eliminate those factors

 

Sorry bud but your guide needs a "this is how not to do it"
Ridiculous amount of paste used, no idea what those white things are?

You are going to destroy your laptop at this rate, I can see the heatsink mounts being pulled out due to overpacking

Now I have seen a lot of messes in my time and I was quite generous in sorting said messes out when I could but if I ever came accross a laptop like that I would take a pic and invalidate the warranty.

 

GoNz0's way of gently suggesing you try less thermal paste. Don't lose hope - just give his advice a try...

Food dough to measure thickness of gap for padding

 

My jaw dropped when I saw the pics.
Also unsure food dough is wise due to the risk pulling the mounts out.

 

I just used the indentations showing on the pads to see how fitment was via trial and error. Light indentation in foam from all 4 sides of chip seemed to indicate adequate contact. Thermals worked fine with that method.

Someone here recently said engineers used some special claylike substance for this type of depth measurement. I thought that might be tough to remove 100%, which might concern me for potential electrical shorts or thermal insulation. The food is biodegradable. Anyways, that is well outside my skillset.

 

it's called Plastigauge... but for getting a rough idea of the thickness for ram thermal pads even play-dough would have worked fine...

 

The pads thickness on the vram modules is an issue for me too; i'm not sure since it's been some days from last opening (and things may have settled better) but even with 0.5mm pads the heatsink might have contact issues on the gpu die (and i don't think my heatsink is bent since it's new, i managed to get a replacement one).

Since i don't have the factory pads i ordered some squishy silicone 1mm ones and also some k-5 pro paste that is a paste used to replace thermal pads (very thick and gummy), when these and the grizzly conductonaut arrive (i had some liquid ultra but i'd rather repaste with the best one once) i'll see what i can do.

As for the pads on other components everyone has their idea (i had a good experience with a couple of 3mm thick small slices on the gpu side of the heatsink) but i don't think they make much of a difference (other than some on vrms and the ssd).

 

lol. Took the heatsink off this AM. the CPU was perfect but the GPU was a little much like you said...

I said **** it and decided to use CLP this morning. Didn't plan on it but I hit 80C and throttled in realbench so I figured it was time...

No throttling in real bench at all with CLP. I think I might have hit 75C but fans were lower RPM.

Will post pics later! Got tons!

Pictures are up on the flickr site.

I had the most difficult time getting the paste to spread on the copper. It was probably 2x as difficult as the die. What's even worse is I threw out the paint brushes that came with it...fail. Either way I got it taken care of. I put a tad more on the GPU just knowing the contact. Dropped my temps 5C from the previous paste in realbench... I've been on battery power for almost an hour...fan has yet to come on...doing quite well

 

Have you aplied this K-5 paste before? It sounds interesting to replace the Vram pads with. Wonder the amount of paste being neccessary on the chips without making a mess with squished paste in that area.
Be careful with Conductonaut. I just can repeat what I read, but it has a reputation of being a little too fluid and will likely run of the heatsink in areas it should´t be...I for myself bought the CLU because of this reason as I was inexperienced with liqid metal. FYI I took my heatsink of today to have a look and it could be removed easily. A little gray shadow was left on the copper but everything else got clean as if it was never there

 

I never used the k-5 pro but from what i see it's just like a very dense thermal paste that somewhat hardens after a bit, to avoid spills (even tho it's not conductive so it's fine even if it spills) i'll apply it in the center of the vram chips and let the heatsink pressure spread it.

As for the conductonaut i'm going to test it on some old stuff first and then use high quality electrical tape ( this one) to cover all the possible electrical contacts on the cpu/gpu (like this and this).

Good news on the removal, it means we can fix it if it's needed; how was the contact patch when you removed it? Did you use a paper towel or some alcohol aswell?

 

I´m looking forward for your results. I used the same tape but cover the whole area round the dies (not just the little spots). The paste really sounds interesting and maybe just right for our purpose on the Vrams. Please report back how it works!

 

You might consider broader coverage of electrical components to help mitigate the risk of conductive and corrosive liquid migrating towards components.

These are instructive primers on liquid metal and about taping to protect components

http://forum.notebookreview.com/thr...-compatibility-with-copper-heat-sinks.800890/

http://forum.notebookreview.com/thr...t-vs-cool-laboratory-liquid-ultra-pro.791489/

 

Thanks Pressing. I think for now it's pretty well protected. I made sure to really press down the electrical tape so nothing would come up. My original idea of having a square thermal pad protecting the die and heatsink didn't work since the gap between the substrate and heatsink were too tight. Electrical tape was the next best logical answer. Fortunately nothing is fried yet so I think I'm good... I have CLU on an old dell for the past 8 months and that's good to go so I'm pretty sure this won't be any different...

I can loop valley benchmark without any throttling stable at 1730MHz on the GPU and I think +100 on the memory...I'll have to go back and check. But I think this is going to be my final expedition of thermal compounds and thermal pads... no throttling in realbench either now.