XPS 15 9550 temperature observations (undervolt + repaste)

This is why in the past I have posted about doing 2-3 test matings to see if both surfaces show a good spread, a good spread for this being a nice matt surface where you can just make out the laser etching of the die so you know if has spread thin enough to do it's job and not become an insulator. Dell make no reference to torque settings but then again neither did Toshiba, the guy who trained me made note to use threadlock (I use the stick type blue loctite medium) but failing that the nip it up enough that you can't turn it with finger and thumb.

If you really want to risk messing things up you can bend the legs up a bit but you should only do this if one isn't the same as the rest, another best to leave it alone as the risk of taking a corner off the CPU/GPU is 100 fold if you accidentally let go of the heatsink while fitting the screws.

I did do this but it was my call and my laptop, I would never suggest someone else do it a the risks are high.

Now I turn the screws a full 360 to catch whilst holding the heatsink down then rotate it to spread the paste, works very well. After that tighten the screws.

 

Certain high-performance thermal pads, like the Fujipoly 17 W/mk are very dense and not very "squeezible" even at 1.5mm.

Given the flimsy nature of the XPS heatsink, very fluffy pads seem to work best as they are so easy to fit and have a large fit tolerance. Proper mating between the vram and heatsink without distorting the heatsink seems to be more important than getting the absolute highest-performance themal pads. . .

That said, I spent a lot of time getting my 17 W/mk pads of varying thicknesses to fit perfectly, and enjoy superb thermals. It takes a lot of work to get proper mating on all 4 chips without distorting the flimsy heatsink and screwing up the GPU mating

 

Some of the better pads are dense but clay like and may be squeezed into thinner shape, but it takes higher pressures and a kitchen tool. I guess world-class conductivity is not needed for VRAM pads so fluffy should be fine.

 

Hi,
Heat sink's price ~ $20 usd for 20 items (international free ship). Default height ~3cm, you should cut ~ 1.5cm ~ 2cm.

Sorry my bad English.

 

Hi,

Thank you for your help~
I re-pasted by MOSFET heat sink.

My config throttle stop
Core CPU: -125mV
Cache CPU: -125mV
GPU: -105mV

I stressed for 15 minutes (AIDA64). Please see picture below.
My temp room ~ 31oC.
I use thunderbolt TB15 (240W).
The PCH temp will be 50oC if I charge by adapter.

http://imgur.com/a/rU5CU

 

.

 

This is my result:

CPU: -135mV
GPU: -100mV

all stable
I used also modified metod to direct all air to heatsinks.

http://imgur.com/a/6NQc2

 

Great mod until dust sticks to the exposed adhesive then it just blocks up the heatsink.

 

That's why I used paper as a dust stoper and a thermal insulation (normal 80*C tape). Just stick some paper to "sticky" part of tape.

 

Fair enough, I did look into it way back when people 1st tinkered with the idea but concluded the foam surrounding it did the same job so I left it alone

 

Yep. Foam SHOULD prevent air from escaping but for some reason... It just works . Now time to fix VRM temps.

 

Hey guys, I could use some help figuring out what is causing my GPU temps to be higher than what I did before.

So I used Arctic's thermal pads of 1.5mm thickness and used a variation of padding throughout this thread. Below is what I have done

http://imgur.com/a/osJ0Y

Note that below the heat sinks I have also changed the 4 pads with the 1.5mm ones and re-pasted.

I also have under-volted in throttlestop and I do have a cooling pad underneath the laptop.

However, since doing this my temperatures seems to have gone up and I'm not too sure why. In overwatch, I used to be able to idle in the menus with a GPU temperature at around ~60-75 degrees C and ingame it'll be ~75-85 degrees. However since adding the pads I've noticed that the game instantly goes to 89-91 degrees C in menu and in game with it sometimes dropping to 88 degrees.

Is it the 4 pads I've replaced under the heatsink causing this rise in temperature? I (stupidly) threw away the stock ones that were originally there so I can't know for sure.

Any help is appreciated.

 

Did you check that your 4 vram pads are making good contact with the heatsink? I think people only had an issue with 1 or 2 of the vram pads and left the other 2 or 3 stock. The problem with putting overly thick-dense pads is they will push the flimsy heatsink away from the GPU. That could cause higher temps. Carefully measured (fluffy lower performance) pads which are just thick enough seem to work best. 1.5mm for all 4 chips seems too thick.

How good is your repaste? Did you try your CPU / GPU repaste, tighten the heatsink fully, unscrew the heatsink to see how good the paste job was? It takes me 2 or 3 tries to get it just right.

Note the case bottom has very little cooling capacity. By putting so many thermal pads to the case bottom things get very hot, you are superheating the fan intake air, reducing the efficiency of your fans and radiators.

Is your m.2 drive nvme? If not, it probably does not need a thermal pad.

Do you need a thermal pad on the wifi card?

Covering the vrm with thick pads prevents what little air there is from cooling the chokes and mosfets. As the case bottom can not absorb much heat, the mosfets send heat to the motherboard...and eventually warm up the CPU and GPU.

If I were you, I would fix the vram padding, make sure you have an excellent repaste, and remove all the other pads you have installed. Look at the temps; perhaps then try some minimal padding in the vrm area (target say the chokes with small pads and see how results are after 1 hour of gaming). Try and see...

 

Put the old pads back in and stop messing about with a cooling unit pushed to it's limits, I fail to understand the needs to keep pissing about with vram pads as all you will do is draw the CPU heat into the RAM due to the better conductivity of the pads, it has the opposite effect.

 

Hello guys hope all is well, after doing a little bit of investigating this morning just wanted to mention something that you all probably know already: The relatively isolated heated area (during gameplay) on the left side of the laptop to the left of the shift and ctrl keys is another big culprit for thermal throttling. From a little googling action that seems to be the power delivery circuity? I couldn't find much about it, but I'm sure it is discussed on these forums somewhere.

 

Hi,
Long time no seen

Still havin my 10y old vostro 1500 with t7700
At that time i undervolted with RMclock.

Now i will get a new one soon (kaby lake) and would like to use Throttle Stop.

Now my question:
For RM clock each multipliers voltage could be adjusted individually.
For Throttlestop you can only apply "-0.x mV"
Does this mean only maximum multiplier is undervoltet or each multiplier by 0.x mV?
To be honest setting each multipliers voltage would be beter because more accurate, but I guess this is not possible anymore for newer CPUs?

Thx for your reply

 

Sadly it's all offset (-X mv), but I do remember spending a lot of time with RMclock figuring out which multiplier would run at which voltage. It was quite a bit of fun.

 

Thx for your reply.
But this is also kind of strange.
If applying e.g. -0.1 mV, this will apply for all multiplier, but stress testing you all do is as far as I see only for max clock (with e.g. prime95) , means max multiplier?
At RMclock you could tick the multiplier which should be used for stress testing.
To make sure each multipliers voltage runs stable.
Hmmm really wondering.

 

What are you wondering? CPU power control isn't as simple as it once was so offset is what we can control. You can still specify the multiplier it runs at so if you want to test the others at that offset you can. I haven't had an issue with offset.

 

i see.
I just think some potential in undervolting/stability got lost in just setting an offset. Because stable minimum offset voltage was different for each multiplier (at least for older CPUs).
Anyway, if this is only possibility of undervolting I appreciate it anyway, since it was not possible for most (all?) CPUs between nehalem and skylake.

 

Reading through this thread, I'm finding it hard to find a definitive thermal pad placement for the 9560 7700-HQ.

Would really appreciate it if anyone has any experience or insight on the placement shown in this article: http://www.ultrabookreview.com/14875-fix-throttling-xps-15/

 

The placement in that article is "as good as it gets", but the issue is that it doesn't solve the VRM area overheating issue. Such pads to the bottom cover actually make things worse by overheating the case and intake air if they're too conductive. If they're low-grade, they don't help enough. One reportedly successful mod by iunlock in another thread is much more involved, adds "fin" surface in the VRM area and redirects a bit of air from the vents overthere and then out through the grille. Some other thin & high-spec laptop designs make use of copper plates or heatpipes from the VRM mosfets towards the cooling fins near the fans.

 

Do not bother, surely you have read it is a waste of time in 99% of cases as it overheats the case or the C/GPU no longer fully contacts the heatsink.
Every answer you need is in this thread, no "definitive" guide as it just doesn't work out well.
Repaste the CPU/GPU and leave it at that.

 

Worked for me

 

I had good luck with iunlock's mod:
http://forum.notebookreview.com/thr...ures-benchmarks-xps-15-9560-kaby-lake.802345/

 

Just undervolted for the first time ever. I have the i7 6700k with the 4k screen and my core package temps when using Prime95 hit 77c, and my Intel HD GPU temp was 61c with the GTX 960 GPU temp at 54c.

After undervolting the CPU (and cache) by -150mv and GPU by -100mv my core package temps when using Prime95 was 62c, and my Intel HD GPU temp 55c with the GTX 960 GPU temp at 40c.

I'm pleased with that! Admittedly though, both stress tests were only 10 minutes each, so I'll stress test it for longer to be sure.

Then I'll do a repaste and compare.

Thanks for sharing all this info!

 

Ok, so I did a longer stress test (about an hour) and my package core temps did rise 2 degrees to 64c.

Now today I did a repaste using some MX-4 I had leftover here, and I'm glad I did it because when I removed the heatsink the original thermal paste was applied really thick and had typically squeezed out the sides.

I did a trial run first and although I only applied a grain of rice size of paste that too proved slightly too much, so I was careful to apply even less the next time.

Now however I notice my core package temps have risen to 67c (with only a degree or two difference between each core). This begs the question does MX-4 need time to bed in or could two degrees higher ambient room temperature account for the 3 degree rise in core package temps?

Either way, 67c is still and big improvement on the 77c my initial 10 minute stress test showed.

 

MX4 doesn't need bedding in but due to it being quite thick you need to press down on the heatsink with the battery disconnected after a suitable benchmark to warm the paste up so you can spread it out better.
Performance pastes are not really developed for low pressure laptop heatsinks so some "tweaking" may be needed.

 

That makes a lot of sense Gonzo because I don't suppose there is that much pressure from those sprung metal strips holding the heatsink down. I did give it a gentle push after screwing it back down, but when the paste was cold. Heating it first before pushing it down would obviously be better at spreading the paste thinly and efficiently. Thanks for the tip!

 

2 stage process m8, you did the 1st part right but you need to pop back in once things are hot. If you decide to loosen the screws to give it a bit of wiggle room you will need to keep a finger on top of each chip to stop it lifting as that can cause air bubbles.

 

Again, that make perfect sense too. Another top tip. Cheers!

 

After some "tweaking" my core package temps on Prime95 are at 59c - massively down from at least 77c (even on a short test))

I'm very impressed! No more fans on idle really does it for me. Cheers guys! I'm only a novice so achieving this is a very good thing!

 

Just done undervolt + first repaste actity

Default
Undervolt : Non
CPU : 91c
Ambient : 68c
Fan : 4890 RPM

Spoiler

Default_Prime_SmallFFT

Undervolt (1)
Undervolt : CPU-0.14 iGPU-0.10
CPU : 83c
Ambient : 65c
Fan : 4070 RPM

Spoiler

CPU0.14v_GCU0.10_Prime_SmallFFT

Undervolt (2)
Undervolt : CPU-0.15
CPU : 80c
Ambient : 65c
Fan : -RPM

Spoiler

CPU-0.15v_Prime_SmallFFT

Finally decide 0.14 for some safety factor

Repaste + Repaste
Undervolt : CPU-0.14 iGPU-0.10
Paste : Cooler Master Mastergel Maker
CPU : 74c
Ambient : 62c
Fan : 3220RPM

Spoiler

Repaste_CPU0.14v_GCU0.10_Prime_SmallFFT

Repaste + Repaste
Undervolt : CPU-0.14 iGPU-0.10
Paste : Artic MX4
CPU : -c
Ambient : -c
Fan : -RPM

Spoiler

TBC

Detail Step
1. Preparation

Spoiler

1. Repaste_Prepare

2. Internal

Spoiler

2.0 Internal
2.1 Internal

3. SSD Thermal Pad

Spoiler

3.0 SSD Add Thermal Pad
- Confirm thickness needed = 2.0mm
3.1 SSD Thermal Pad
3.2 SSD Thermal Final

4. Paste

Spoiler

4. Thermal Solution

5. Before & After Pasting

Spoiler

5.0 Paste Original
5.1 Paste Clean

6. VRAM & Paste

Spoiler

Confirm Dell using 1.0mm thickness for all
6.0 VRAM Pad Thickness
Realign
6.1 VRAM Pad Align
6.2 VRAM Pad Add

7. VRM

Spoiler

Confirm 4.0mm distance to touch chassis
7.0 VRM Distance
Preliminary try, like to observe few day on temps
7.1 VRM Pad Add

 

Great photos and thanks for detailing the step by step.

It seems you have the Kaby Lake 9560 7300HQ based on your photos. There are a lot of posts here indicating that Kaby Lake is not quite so amenable to undervolting as the Skylake was as it is running much higher clocks in essentially the same package. For the 7700HQ I think people were closer to 120mv to prevent issues but I don't know about the i5 so consider running some serious stability testing or be prepared for potential crashes from aggressive undervolting.

edit - Oh you might try running the ROG Stress Test for 15 minutes as it is a good workout on the CPU and GPU at the same time. It kills the 7700HQ but a well tuned 7300HQ should be able to do it without throttling...

 

Yes, I'm using i5 7300HQ as per sig. Will give it a try for ROG later, so far few day undervolt CPU -0.14 had stable with Prime + Game NFS.
May keep as it now.

Got a question for iGPU undervolt, so far try -0.10, its seen still had more room to go, but does this really benefit to battery life??
(3DMark stress test is focus on dGPU, i not sure how to stress iGPU here)

Surprise the standby time now is 8h+ Previously around 5h+ (56wh battery)
Battery Life

Update : ROG Realbench
CPU : 83c then throttle at 2.5MHz at 74c
Ambient : 79c then throttle at 2.5MHz at 73c
GPU : 75c then throttle at 1.3Mhz at 70c
ROG_Realbench
At least the Undervolt is stable so far, so I will leave it at the moment.

 

Good. You should be able to kill the throttling in the future. This is an interesting thread to follow

http://forum.notebookreview.com/thr...ures-benchmarks-xps-15-9560-kaby-lake.802345/

 

iGPU undervolt helps marginally. If you look at HWInfo when on your desktop the iGPU doesn't fully ramp up at all so essentially it barely impacts anything at all. I think mine is set to -0.150 or something crazy. lol

 

@tachlio

Thanks for posting the useful photos.

Which thermal pads did you use? Also the blobs of blutak, where they used just for gauging the height required for the thermal pads but then removed before replacing the bottom cover?

Thanks.

 

I'm using Henkel Gap Pad VO
http://www.bergquistcompany.com/thermal_materials/gap_pad/gap-pad-VOUltraSoft_properties.htm
- Low Density & Thermal Conductivity
- A highly compliant material that is ideal for fragile component leads.

Yes, the plasticine is working like gauge, and I'm stick 2x2mm pad to compare as I dont had caliper on hand.

Spoiler

7.0 VRM Distance

It will remove after measure
Check final condition

Spoiler

7.1 VRM Pad Add

 

I found XTU was delayed startup, it need around 1 min from cool start to effective.

Anyone try to change it? or how to make it launch faster?

XTU_Service_Delayed

Together, I found after PL1 throttled, it took so long to recover back... does this impact performance?

PL1 Throtte Recovery

 

Below is a superb & brief intro to thermals in electronics.

Shows how engineers do back-of-the envelope heat calculations in layman's terms and why our thermal paste upgrades are so effective:

 

Rocking 1733MHz at a 0.900v undervolt.... pretty awesome really! Temps WAYYY down with that much of an undervolt. So far everything seems stable!

 

Mind posting a picture of your settings?

 

Actually ended up crashing..so I bumped it up a little to 0.925v. Still awesome considering that's about stock clock...

Here ya go! I think valley topped out at around 62C

 

Great too see so much potential for this laptop.

Sent from my SM-G920F using Tapatalk

 

Thanks to CLP liquid metal paste and some hacked together thermal pads...lol

 

thanks for this great post! i was able to play 3 overwatch games without throttling (bordering 87C though). i did a repaste and repaded. undervolted 101.6 mV. works nice.

 

That is interesting. Mind sharing why you are putting pads alongside the vram there? Is there something on the other side of the board generating heat?

 

Quoting from the owners thread:

But a PL1 should show up on HWiNFO right? And it doesn't as shown in the screenshot. As for causing problems by messing about, I've built PCs since the 386 era and my first overclock was the dual celeron 366 in 1999 so i have some experience there, but to be sure I applied a bit of force during unigine valley benchmark on the heatsink in the corner of the hottest vram and it made no real difference so I don't think it's bent; all while GPU is at ~49c.

I could call Dell, but what to say? "Hi, during benchmarks my VRAM heats up so badly that the GPU hits a power limit throttle". You think they'll do anything based on that input?

It does have a huge impact, but it's not a little air; it's a massive amount of it. It's louder than a vacuum cleaner so it's a ton of air.

I ordered some 0.5mm 11W/mk pads, and the K5 Pro is a gooey malleable type of paste that does not liquify and sits at 5.3W/mk i think. So I'll give those a shot. Still, it's a weird case. I would love to do some testing without the heat pipe using separate heatsinks so I can check individual temperatures of VRAM and GPU more accurately but obviously there's no way of doing that without a lot of MacGyvering, so that's probably off the table.

Once the new thermal pad+paste is in I will test some more and see if I can further isolate the airflow away from VRM and only on the VRAM; tricky for sure but I'll try.

I don't think there's a specific VRAM temperature sensor on the board is there?

You are talking about these mosfets, correct?


What's weird is that in my case, all of the mosfets in green were well below 50c but meanwhile the VRAM under the QR (shown here as 66) was >60c.

I'm not sure what's being shown here in the red dots with yellow numbers though, is that the mainboard temp? Because I could not locate those.

 

For the benefit of others as I'm sure you know, the MOSFETS are the little black rectangles which tend to be next to the CHOKES which are the towering silver cubes. Both get hot but the former are causing the real heat issues people see.

As the heatsink shares heat among the CPU, GPU & VRAM, I can envision plenty of cases where the CPU and GPU will tend to heat up the VRAM above VRM temperatures. With all the thermal and power gates the rules are not so clear but maybe lower power situations with both the CPU and GPU humming along. Not sure it indicates an issue.

Still overall you seem to be having some strange problems. . .

iUnlock told me what those red dots with yellow numbers are but I don't remember if that was in a chat or is posted here. If I recall correctly, he was just pointing out precisely where the temps were being measured.

Also there are tips here to check to see if your heatsink is bent. First step is just to put it on a very flat surface and look at it.