XPS 15 9550 temperature observations (undervolt + repaste)

You are showing signs of what is technically called a piss poor heatsink mating
Pressing said a 10 degree difference, that is a sure sign it isn't mated. I do 2-3 test mates to see the paste is being spread so thin you can see the chip underneath when the sinks removed.

If you messed about with the VRAM pads and changed them that is probably the cause.

 

I have already undervolted and made a decent improvement on fans (I'm having some issues with graphical glitches with anything more than -105/-40 for CPU/iGPU but I have more pressing matters to deal with), I repasted only because the fans were occasionally on when during web browsing and light loads. I've done this plenty of times for desktops and done plenty of messing around inside laptops so I thought it would be a piece of cake!

Given how many times I've reapplied with the same results, would I be right in saying this is looking like a bent heatsink situation?

I have messed with the VRAMs, I edited my prior post with some details but the GPU isn't half as bad thermally as the CPU just now, it tends to level out around where it should throttle at 77C but stays at or ever so slightly under full boost.

 

Yeah buddy, my guess is the heatsink is bent. I've only encountered it with about a dozen XPS laptops, lol. Although they haven't been that bad, my guess is it got a little worse when taking the heatsink off (those darn VRAM pads can be sticky). Trust me when say that it's super easy to bend the heatsink, those damn heatpipes are fragile.

 

Hey guys! I've both 9550 and 9560 ( my 9550 was refurbished and after 11 months and 3 times dell reemplace the motherboard because of the coil whine and 3 times the screen, they offered me to send me a brand new 9560) i decided to compare temps!

Both are i5, 8Gb ram, repaste both and only on the 9560 i've put pad on the VRM. I've tested with prime 95 with stock voltages and undervolt to see the differences.

With stock voltages the 9550 power package is consuming 11w less and with undervolting same on both -140 core&cache and gpu -100 there are 7w less consumption on the 9550. Isn't supposed that Kay Lake consume less?

http://imgur.com/a/MbqyD 9550 Stock
http://imgur.com/a/CjdHa 9550 Undervolt
http://imgur.com/a/uWuVJ 9560 Stock
http://imgur.com/a/Qckj7 9560 Undervolt

 

The good thing is that you should be able to get one from dell pretty easily.

 

The wattage measurements are simple estimates that sometimes are very inaccurate.

You could measure power at the wall but with different GPUs and some other changes etc. that does not help much.

Also each chip's performance is a lottery.

You can consider the Kaby Lake chip a slight evolution to the Skylake chip; it runs at higher clocks and on average should be a bit more efficient (particularly with some 4k tasks for different reasons).

 

So I've made progress; heatsink was a bit bent and I think I've corrected that as the paste distribution is a lot more even.

That wasn't even my main problem though, turns out the 95% surgical spirit I was using to clean has canola oil in it! I've ordered some 99.9% isopropanol from Amazon and we'll see how that goes when it arrives.

I've gathered they're very good at this sort of thing, especially given the button in Command Power Manager to order another battery. Any idea what the prices are like? I'm in the UK but I imagine they'd be similar?

 

Temps are fine but both systems are being throttled by TDP after repaste so those temps are not a true reflection over prolonged load (because they're being throttled down to 800 MHz)

Sent from my SM-G935F using Tapatalk

 

There's no throttle, i just took the screenshots after i stoped prime95 and the 800Mhz is on idle.

 

Just thought I'd make a post here since I've experienced the VRM/power throttling in windows and recently got nvidia drivers working in linux.

Besides some nice dynamic thermal throttling that occurs when things overheat the annoying throttling happens when the CPU and GPU are fully loaded at the same time (furmark + prime95). A few seconds later I can see in HWInfo64 that the "Ring PL4/IccMax/VR Limit" warning turns on and the PL1 power limit drops quickly, usually to 15W, and then very slowly (over about an hour) recovers to theusual 45W. For the first few minutes the CPU clock is stuck at 800Mhz and everything is slow.

So the point of the story: it doesn't happen at all on linux. The usual thermal throttling takes place after things heat up but the CPU manages to stay at about 2.5Ghz with prime95 and furmark running and jumps straight back up to 3.4Ghz as soon as I stop furmark and things cool down a bit. Interestingly, the furmark framerate is higher on linux (67fps on 1280/720 vs 52fps max on windows).

So I guess for those trying to game on windows it might be worth investigating more on the software side - I would imagine its a driver rather than the firmware that is responsible for this throttling.

Also a side note - if I run furmark and prime95 on linux when on battery power, the power consumption gets up to about 140W and the battery voltage drops to just over 10V. A few seconds later the machine turns off and does not repsond to the power button until you plug it in. Maybe this sort of problem is the reason for the throttling? Power consumption seems to increase with heat - its only 125-130W when everything is still cold and rises over 10-20 seconds to 140W before it crashes.

 

Probably no Intel DPTF in linux?
Did you monitor the temperatures in linux? I guess they went higher?

Maybe the battery can't feed that power. Not sure if it is a good idea anyway.

 

CPU gets to its maximum ~98 and throttles, and the highest two ambient temperatures reach 90/95 (the higher one jumps up higher than that if I lift the laptop into the air, so I leave it on a flat surface). On windows the ambient temperatures only get to 70-85 before the throttling starts.

I can't actually see the GPU temperature in linux yet but I can only assume it gets to 78C and starts throttling - all that power must be going somewhere lol.

haha yes, I think the inability to use 140W on battery power is probably a good thing.

 

Interesting...thanks for sharing mate. Well noted. Kudos

 

So I worked out how to see the GPU temperature - and the 78C limit also doesn't apply on linux. GPU at 94C and still running 1920x1080 furmark at 53fps

 

Repasted my 9560 (7700hq) today and applied an undervolt of -125mv cpu / -80mv gpu. Originally applied too much paste and cpu was hitting 88deg in prime95 small fft in 10mins. Repasted and used the paste sparingly and its brought down to 81deg. Is this a good temperature to expect?

 

For small FFT's it is about right. I get about 78 @ 3.3ghz fans at 4000, it is limited by the cpu package sat at 45w, not the best test as it hits the power limit.

 

You can probably lower it by some degrees by undervolting a bit more and using an higher quality paste.
These are 10 mins of prime small fft on liquid metal but i think the metal lowers the temp by 5-6 degrees over paste meaning that you may have room for improvement using paste.

 

Although there have been a significant number of Kaby Lake instability reports with undervolts in the -130mv range...

 

Repaste and undervolt by -140mv

AIDA64 (no gpu)
http://imgur.com/a/OsNkT


AIDA64 (gpu): throttle at 8 minutes and unthrottle after 1 minutes, I don't know what happen.
http://imgur.com/a/WQuBI


Edit: XPS 9550 i5, disable Turbo, frame rate limiter at 70 degree. Room's temp ~ 31 degree

 

Those clocks are terrible!
And tick FPU only.

 

Excuse me Sir, can you explain more? I don't know how to fix it.

 

Better read the thermals thread then.

Sent from my SM-G920F using Tapatalk

 

The paste I used was Grizzly Kryonaut which I thought was a fairly popular one here. In fact I repasted quite a few times and low 80s was the best I got. Which liquid metal paste is recommended and is there any compatibility issues with using a liquid metal one?

 

Only the one with the slightest mistake allowing a ball to drop into the board may blow the motherboard with a warranty refusal as you have to damage the chip to apply it by roughing it up.

 

There are a few liquid metal threads here at nbr. Some are dedicated to LM and plenty of individual LM posts for the 9550 and 9560. Use the search button and read them for more info.

There are serious risks to poor application or use of a "drippy" product; i.e. poor cooling or catastrophic damage.

 

Can Sir send me link?

 

It's this thread.

Sent from my SM-G920F using Tapatalk

 

Are you kidding?!?

 

My xps i5 9550, undervolt -140mV, disable Turbo.
So, how can I undervolt GPU GTX960m?
I run nvidiaInspector.exe and limit frame rate at 80 degree.
When I'm playing game (assassin creed), frame jumps 10 - 110fps

 

To save us ALL repeating ourselves when you obviously haven't read ANY posts I suggest you take a few hours and play catch up

 

You can't undervolt the 960m

 

Dear Sirs,

I use TB15 dock (thunderbolt port, adapter 240w) to charge and some usb ports from dock.
PCH temp ~ 62 degree. How to fix PCH temp? I search forum but I can not solve this problem.

 

No fix, that temp is fine.

 

That TB15 dock has a lot of issues. Do a search as many were recalled by Dell. I think they were replaced with the TB16...

 

Dear Sirs,

Can you tell me where is PCH sensors in xps 9550?

 

Confirmed, TB15 sucks balls, the TB16 works flawlessly without any issues for me...

Sent from my SM-G935F using Tapatalk

 

i7-6700HQ, Gaming Load with GTX960M (Rainbow Six Siege for several hours).

Max CPU temp with stock voltage (CPU Package):88°C
Max CPU temp with 150mV undervolt: 77°C
Max Power with stock voltage (Package): 36,06W
Max Power with 150mV undervolt: 25,90W
Unmodified Geforce Graphics only dropped from 91°C to 89°C, will try to undervolt it too.

Ah huge success, i am really happy

 

Hi,

I saw a Mastergel Maker kit that comes with a grease cleanser so it looked like a nice little kit to use for a repaste. I asked the vendor if the kit was compatible with laptops, and they said it does. I just wanted to get some more thoughts on whether the Mastergel Maker with Nano Particles is compatible with the dell xps 15 9550 heatsink. I've read that there are many types of thermal pastes, I wonder if this one would be safe to use. ( it supposedly got some "HIGH-TECH NANO DIAMOND PARTICLE" in it according to the CM website. http://www.coolermaster.com/cooling/thermal-compound/mastergel-maker/ )

Thank you.

 

Find out if it is suitable low low pressure applications, if not don't bother.

 

Why not use some of the tried-and-true pastes mentioned repeatedly in this thread?

 

It's being sold nearby, so it was just convenient. So I decided to give it a try with the mastergel maker. After some research, it's suppose to be on a similar level
with the MX-4 mentioned on this thread. So I opened up the heatsink and found a lot of stock thermal paste. So I cleaned all that off and repasted with
the mastergel maker paste. I tried to follow the "grain of rice" amt. but I might have just squeezed on just slightly more than that.

My temperatures with the stock paste with -144mv UV was Idle: ~38C and Max: ~70C
After the repaste I was getting Idle: ~40C and Max: ~ 73C

So I my temperatures actually increased slightly to my surprise. But experience wise, the fans were on way less while idle and at much shorter durations after some
heavy gaming. The fans used to be on for very long durations after gaming, sometimes it just won't turn off even when temperature readings were already in the ~45C.
I also accidently touched some of the thermal pads during the repaste which might have dirtied it. Should I do another repaste + thermal pad change as well?
I forgot to take photos of how much paste I put in, It was my first time and I was all focused on getting everything done correctly. But for sure, it's way less than how much
was used with the stock paste job. The 2 missing thermal pads on the gpu ram is stuck on the heatsink.

Thank you.

Image of stock paste.

 

It's very easy to bend the heat sink and make your temps worse

 

My lappy

http://imgur.com/a/en65N


My room's temp: ~30 degree

 

Where are your metal heat sinks from and what is their depth?

 

No extra heat sinks, no thermal pads, idle. New thermal paste but it actually made CPU temp go up about 3°C.

 

In what conditions? Compare when running the same benchmark (like Prime95 for the CPU, Heaven for the GPU, or both thogether).

A higher temperature may mean that you have too much gap between the chips and the sink. You can try tightening the screws (carefully, screw threads may get stripped). If not better, remove the heatsink and see how much paste remained and where, check if the VRAM pads are too thick or the heatsink is bent.

 

Room temperature 26C, sink in perfect condition. I think there's too much paste but I'm happy with temperatures anyway. I'll measure temps under load.

 

I have to butt in here, that is a really really bad idea, this isn't some large PC heatsink where you have sprung screwdowns you can tighten so far to get the desired pressure, this is a flimsy affair with nuts that will happily pull through the motherboard (a few posts about it) leading to a full board swap and a voided warranty if you leave marks doing so. It is so easy to cam out tightening these and gouging the board.

It is a low pressure heatsink, no excessive force should be used, just nipped up enough to stop them vibrating out if you don't add more threadlock.
If it doesn't fit properly after changing pads then you ballsed it up and need to put the originals back in if the ones you have are too hard.

 

Just want to second GoNz0. Review the mating of the heatsink. Putting "better" pads on the VRAM could cause it not mating properly as the Chips differ in height and Need some squishy pads ike the Dell ones. And seriously be careful w/ the screws. On my old 9550 one did pull through the motherboard (I saw it just before returning it )

 

Sure, but you don't know if he perhaps just didn't tighten the screws enough do you? A torque screwdriver would've been a good idea for this purpose, but mostly there is none around, and I don't know if the torque is specified at all.

On youtube there are great videos of paste spreading under a piece of glass. With a real heatsink you don't see anything. You can just try to feel if it sits nicely (doesn't wiggle or lift back), squeeze it with your hand so that the paste spreads reasonably before you start tightening the screws. "Thin" pastes get squeezed out of the gap nicely without much pressure. Factory paste seems the "thick" kind that one sometimes sees pre-applied on replacement heatsinks, that probably requires a lot of pressure to get out, and I guess that's why factory paste is bad so often. Too thick VRAM pads are also kinda squeezible but would require too much pressure so that the screw threads would give in first.