Dell XPS 9560 VRM Thermal Insanity As Seen Through FLIR

Howdy folks, I wanted to cross post this from the XPS 9570 thread I intended this analysis for so that people with the XPS 9560 would be able to find it a bit easier.

Anyways, Today we take a look at the known poor VRM thermal design of the Dell XPS 9560 (effects XPS 9570 as well). We find that the CPU stress test while running hot, is at least under control. Then, switching over to the combined CPU/GPU stress test we find massive runaway/saturation issues for the VRM/MOSFET's. Installing thermal pads and optimizing airflow can help to mitigate this issue almost entirely.


Please let me know if you have anything you'd want to be seen through the FLIR as well (Picture or Video), I'm always interested in using my equipment for new purposes.

I also took this opportunity to do some thermal modifications outside of padding the VRM's.
-Installation of Liquid Metal on CPU/GPU
-Removal of copper air divider over bottom vents
-Installation of Aluminum Ducting Tape over Fan shrouds to optimize air flow through Bonded Fin Assembly.

UHD Image Gallery for part numbers or just a better perspective:
https://imgur.com/a/XCR3rD6

Link to original thread:
http://forum.notebookreview.com/thr...of-vrm-thermal-pads-in-xps-9570.826873/page-3

Thanks for looking!

 

I've always wondered if directly padding the VRMs would just lead to isolating the heat from the ambient sensor so I've always just padded the grey chokes to the bottom panel of the xps. Furthermore, in the CPU section of your youtube video there are 3 fets getting extremely hot, but what about the ones under the metal shield (labeled power on the underside) next to the cpu fan? (I feel like everyone only focuses on the xposed fets and not the ones under the shield.) Would it be possible to get some data on those as well?

Would it be possible to get some thermal images of your xps with prime+heaven running to see where the heat spots are on the bottom panel?

 

I actually have some images of the whole machine to see if there was anything else I'm missing, the only other part that really gets warm is the power MOSFET that's near the battery connector. This is going to be the main regulator for the general system. Although this one doesn't even get close to the levels we're seeing with the CPU/GPU VRM.


I'll run some tests and get you an image of the machine closed up under those tests. Give me a few minutes!

 

The area on the left side (near the cpu fan) where the metal shield is is obstructing the thermal camera's ability to measure the temperature of the components on the board is it not? Always been curious as to what the temps for the fets and chokes in that area are like.

edit. Didn't see that you were already on the run and collecting data at the bottom of your post. my bad

As for the heat generated in the area next to the display connector, I've padded both the choke and charge IC (i think that's what it is) and the PCH next to it and it had led to a noticeable decrease in palmrest temperatures under load.

 

Funny enough, you should see what the SSD does to that palm rest. From the FLIR it looks like it cooks right through.
So, with the back panel on I scratched my head a few times as to why it looks like the bottom panel is hotter with the VRM soak through than at load.
Then I realized that at idle the fans very rarely kick on, as the aluminum cross bars are over the inlets they must suck up most of the heat when under load.


*EDIT* I guess that is the main power regulator next to the battery? Must have had the SSD side and VRM mixed up. Also, ignore what shows in the display, that's an emissive reflection and not real readings.

 

Hi republicofsam, Ive sent your video to Frank Azor asking him if he can do anything about poor VRM cooling of xps laptops.

https://twitter.com/xtomci/status/1084736475965636610

I just hope they will have this fixed in the next xps laptops.

 

This is the current state of my XPS 9560 in terms of thermal cooling mods:

Reduce CPU+GPU+VRM temps
- Custom heatsink with 4 heatpipes and copper fins with the electrical tape "mod" to force more air through the fins and repasted with Gelid GC extreme
- Thermal pads on all the offending VRMS (1.5mm x 3-4 stack)
- Kapton tape on the copper sheet on the bottom panel where my taller heatsink makes contact with it (Did this to reduce heat transfer from the pipes to the bottom panel and heat up the VRM)
- Undervolted CPU by -0.126mV and GPU is locked at 1657mhz at 875mV

Reduce Left Palmrest temperatures
- Thermal pads on the thermal pads on the giant grey inductor near the battery connector (1.5mm x 2)
- Thermal pads on what i'm assuming is the charge controller since it heats up a lot while charging (1.5mm x 4)
- Thermal pads on the PCH (1.5mm x 4)

This combination of mods to my XPS has let play CSGO with the CPU @ 3.00GHZ around 70-75C, GPU hovering around 65C @1657MHZ and the VRMS hovering around 70C

 

what do you mean by this?

 

There's a Chinese seller on taobao selling a custom heatsink for both the 9560 and 9670 that has a second set of heat pipes (identical to the original ones) stacked on top of the original ones with the aluminum heatsink fins replaced with full copper ones for better heat extraction.

If you feel that the heatsink is a bit too extreme for your tastes, they also have a normal heatsink with just the copper fins

You can see the heatsink stack in my picture if you look closely

 

How do you think those custom heatsinks are working? Is the exhaust hot as hell?

 

I'd say the custom heatsink works extremely well if you look at what kind of temps I get playing CSGO with my current XPS config. Exhaust temps are the same or lower

The kapton tape is also essential as it prevents the heatsink from making good thermal contact with the copper sheet on the bottom panel and heating the VRMs

Another thing I do is remove the copper oxide layer that forms on the heatsink by polishing the layer off with toothpaste to help conductivity. (Just don't go too hard and make the surface uneven)

 

Do you have a link to those heatsinks? I don't see the double stack you mentioned.

 

I'm sorry. My reply is not, strictly speaking, helpful... at least from a technical standpoint. I just...

...HOW DID YOU NOT PUT THE 'PREDATOR' SOUNDTRACK UNDER THIS VIDEO?

 

Yeah for sure. If you're looking for the double stack on my image you can look closely at where the heatpipes leave the CPU side of the heatsink and go towards the finstack

Otherwise, you can look at these glorious modded heatsinks on this taobao link where they also supposedly have upgraded fans (although I couldn't see the difference) and modded heatsinks for the 9560 and both models of the 9570

Also, my heatsink came UNWARPED unlike all of the normal factory heatsinks.

 

Hah I wish I could figure out what is going on in that store, even google translate doesn't help me. I'd love to get my hands on one and do some comparison testing to see how much it helps.

 

Lmk if you need help with ordering. Not sure how you're going to get it shipped outside of china tho

 

Hah, that's too bad they don't offer international shipping. Oh well.

 

@htaige @custom90gt WTF! I am mind blown.
For certain things, you gotta love the Chinese creativity.
Let us know if you find a way to purchase it.

Dell sent me a replacement Heatsink because mine was a little warped. I still have the original one
It's a tantalizing idea to try this hack, but I am really afraid I could screw things up.

@htaige what kind of temps/frequencies do you get using prime95?

 

I wouldn't recommend modifying the heatsink yourself as the modded heatsink from the link above was about 30-35USD which is the price of a new stock XPS 15 heatsink.

With my current setup, I can maintain full 3.4GHZ turbo on all 4 cores with temps fluctuating from 70-80C because the fan curves are all kinds of messed up with this much cooling. The laptop will ramp up to 5000RPM for a bit but then go back down to around 3000RPM when the temps climb up to 80C and then ramp back up until it cools down to around 75C. (repeats over and over)

 

> The laptop will ramp up to 5000RPM for a bit but then go back down to around 3000RPM when the temps climb up to 80C and then ramp back up until it cools down to around 75C. (repeats over and over)
This goes on with no clock downthrottling and with no throttling flags showing up (HWinfo64 Sensors)?

Yeah I'd be interested how this compares to iunlock's mod.
Could you perhaps also try running Prime95 and Unigine Heaven at once? (Do the VRMs heat up)

The copper fins, are they bigger than stock?
I assume the image posted by htaige on the previous page is with the stock heatpipe?

 

I mostly used throttlestop to monitor throttling flags, but as far as I can tell there's no clock throttling going on. I can look more into this if you want.

The picture on the previous page contains the CUSTOM HEATSINK INSTALLED and while it visually looks identical to the stock one. I can assure you that it weighs significantly more has 4 heat pipes and copper fins that are the same size as stock.

I think prime95 + heaven might be a bit too much for my machine as the VRMs already reach 70c while the GPU is under full load at 1657mhz @ 0.875mV and CPU @3.00ghz drawing about 15-21W as a package. I can also look into this if you'd like. However as stated before, temperature for the CPU and GPU in this state are 70-75ish and sub 70C respectively

 

Please do, it'd be good to know if this is really just due to faster cooling response - Intel DPTF throttling actions sometimes look similar.

Tnx, I wasn't sure... I also don't understand what you mean by that it came unwarped?

I guess it would probably throttle at one point, considering that you don't improve VRM cooling like iunlock. Though, better cooling via heatpipes might also change things for the VRMs. That's why it'd be nice to see load tests as similar to those iunlock did for comparison.

Pity these custom heatsinks don't come with larger heatspreaders that would also cover the VRMs (with holes for the higher components, coils etc).

 

Can you run Prime95 Small ffts and tell me what temperature the CPU 'attachs' to and what your cpu package power and clock speeds are, are they fluctuating? What happens if you lift the laptop of the desk?

 

PRIME 95 BLEND TEST WITH DPTF ENABLED
As you can see, temperatures for this result are messed up thanks to DPTF interfering for no apparent reason as all ambient temperatures are under the threshold 77C. This may be a result of DIMM throttling, but I have experienced with gaming and throttling where the DIMM temps easily exceed 60C.

You can also see that the fans ramp up and down during the blend test, as mentioned before, and that my temps to lie somewhere between 70-80C.

I DO NOT RECOMMEND LIFTING YOUR XPS because doing so provides too much unrestricted airflow through the vents on the bottom of the laptop. Although this may be beneficial to core temperatures for the CPU + GPU, it tends to negatively affect ambient temperatures as the now unrestricted flow of air through the bottom means that less air is entering and flowing over the VRM area through the vent near the display hinge.

Fortunately the functionality of DPTF is extremely easy to enable and disable with throttle stop as you can just enable and disable the new DISABLE AND LOCK TURBO POWER LIMITS under FIVR in the newest 8.7.06 version. I will run the tests requested by the other users when I have time to do so and post results.

As for what I mean by not warped out of the box, I mean that my heatsink makes relatively flat contact with the CPU die as the core to core temperatures are usually 2-3 C within each other, unlike the 5-10C you might find on a stock dell heatsink.

 

DPTF actions are undocumented an it is royal pain somewhere not simply seing what exactly makes power limit throttle. But I recall stable Prime95 performance was reported several times with the 9560. Not sure if something was changed in BIOS or DPTF since. Check iunlock's thread, and the 9560 thread a few months earlier.

This is indeed a simple new option, though I recall some managed to kill DPTF in the past (raising the power limits is new).
But perhaps consider working around better so that you wouldn't have to do this. The loaded GPU eats more watts than the loaded CPU, both together even more.

The default airflow through that grille is practically negligible, so if you want some, you must produce it yourself (see iunlock's mod). I do recall seeing funny effects of increasing fan speeds sometimes, but raising the rear was mostly helpful.
Though with your heatsink it may be a possible to consider taking VRM heat to the heatpipes somehow.

 

Performance I've had with Prime95 used to be stable until this past summer where it would set PL1 down to 7W immediately after starting a blend test. I've been using throttlestop to prevent this from happening and have only recently enabled DPTF once again. Would really appreciate it if there was documentation on what Dell's implementation of it is.

Interestingly enough, this PL1 throttling does not occur while gaming and the power hungry GPU is loaded 100%. I use the throttlestop method because I don't have to worry about windows update from reinstalling DPTF and whatnot. I have seen the posts were users have nuked DPTF

I am going to try A04 version of the diver and see how it goes later. (Used to be on A09)

I have looked into iUnlock's mod and although the current airflow through the hinge vent is pretty minimal, I do see a couple C difference between raised and not raised. I can also see the ambient temperature drop while running Prime95 when the fan decides to ramp up from 3000RPM to 5000RPM and the ambient temperature go back up when it ramps back down to 3000RPM

Furthermore, directing airflow as shown in iUnlock's mod as shown in his thread is not possible with the modified heatsink as the extra height from the 2 extraheatpipes leaves no room for the air to escape through the gap he created.

 

thanks for posting.

It looks good. However, I am not sure how good your results are because I am not familiar with how the 9560 with a quadcore chip behaves with the stock cooler.
You do have a drop to 7W towards the end. Are you sure you can fix that with throttlestop's new feature?

 

Throttlestop's new feature basically prevents anything running within windows from modifying the PL1 and PL2 power limits. This I have tested before and it runs Prime95 with 0 issues after enabling it. I was quite disappointed to see that DPTF's erratic behavior still remains.

 

Thanks, @htaige. When you have some time, you can use that new throttlestop feature and post new results.
You might run the same benchmarks as iunlock so we can compare with him (like wprime or times spy).

Also, have a look and the other temperatures. The Ambient sensor in the middle of the VRMs gets pretty hot.

 

Pretty sure that iUnlock also has Prime95 results where he quoted 60C with liquid metal.

PRIME 95 SMALL FFT

So these are Prime95 results at the small FFT settings. Temperatures have stabilized around 80-82C, but as you can see, the fans hardly exceed 4000RPM and are starting to pulse between 4000-5000 RPM almost as if the machine wants to maintain a CPU temperature around 80C.

I decided against using wPrime as a workload using it only resulted in 21W of CPU power usage compared to the much higher power usage of 42W using Prime95 at small FFT.

I don't know why people are telling me to look at ambient temperatures even though I have stated that all of them remain below the set point of 77C. (I will add that with Prime95 only, NONE ambient temperature sensors exceed 70C.

I would also like to add that these temps appear to be pretty reasonable as the fans are not running at max speed 100% of the time and that I have not tried using liquid metal as a thermal compound.


Prime95 Blend + Heaven
Using Prime95 blend + Heaven causes DPTF to kick in and throttle the processor. (I think DPTF doesn't like the fact that the CPU has high power usage while the dGPU is active and running.

I could disable DPTF, but I need this laptop for college and would prefer if a mosfet didn't let the magic smoke out while Prime95 with full GPU load. Sorry guys

I am however willing to post results to more conventional workloads that are more in line to what the machine will see in everyday/enthusiast usage.

 

Thanks for your tests,
I wish I could try that custom heatsink.

 

Which CPU do you have? what is the TDP?

 

Prime95 Small FFTs torture test.

I ran this benchmark because I wanted to show how the computer will PL1 throttle on a desk, and then raised.

I started the test with the laptop hard against the desk.

You can see it maintains CPU PP (CPU Package Power) of 56w for the first 6 minutes. After that it appears DTFP takes control and aims to maintain a lower CPU temperature. PL1 fluctuates between 56w - 45w. After a minute or two it tries to maintain a high CPU PP on average again, and allows higher CPU temperatures.

You can see where I raised the computer as CPU PP and PL1 IMMEDIATELY dropped to 40w evening out to 45w, if you allow this to run longer it will appear it is trying to maintain a CPU temperature around 80c. I believe this is because it thinks it is on your lap and doesn't want to become to hot, I think this behaviour must be written into the BIOS at minimum and further controlled by DTPF. I will explain why soon.

When lifting the computer certain temperatures do rise (hwinfo64, dell ec, second in line, maximum of 74c) ambient will rise. I have seen this temperature upto 120c before so I don't believe it is the culprit for PL1 throttling. We should call this lifted throttling.

I have removed intel DTFP to see if the removes lifted throttling, it doesn't. instead of PL1 to 40w, The CPU PP will jump between 30w and 65w, it's very erratic.

Removing intel DTFP does remove the first throttling we seen. But it is not an improvement overall due to the lifted throttling behaviour becoming worse. It seems there is no way to remove the lifted throttling to see how the VRM temperatures behave as it is possibly written into the BIOS. What triggers the laptop to know it has been lifted? I don't think it is any temperatures in DELL EC, as I have seen them higher with the laptop on the desk. As soon as the laptop lifted it immediately PL1 throttled, if it was a temperature trigger it would happen at a certain temperature each time.

 

@J99 the xps 15 does have accelerometers. I have to test your findings.
I wouldn't mind having the CPU throttled when the laptop is moving.

 

Sometimes it is immediately and sometimes it is not, I would guess it has multiple trigger points and sometimes it might ignore some until another is reached.

 

Could you please use the FLIR to see if bridging the VRMs to the heat pipe with thermal pads is effective as some claim?

 

Are you sure you're not temperature limited? 97C is a really high temperature and it doesn't seem that your laptop is really able to maintain sub 90C without throttling. I would recommend a repaste and a healthy undervolt before you continue with any testing.

Also the high deviation in the core to core temperatures in your results also suggests that your heatsink is WARPED. Like really really warped and does not have even contact and pressure on your CPU.

My XPS has the i7-7700HQ, and was able to maintain full turbo with around 85C running just Prime95 by itself. And as you can see from my most recent results with the custom heatsink, temperatures are even better but DPTF keeps interfering.

To prevent PL1 throttling, I highly recommend that you use the Disable and Lock Turbo Power Limits settings under FIVR so that nothing is able to modify PL1 limits.

 

I have tried this method with the recommended Arctic 0.5mm thin thermal pads, but only saw an increase in Ambient Temperatures. I don't think thermal pads are very effective in x and y direction and only really conduct their rated power in the z direction. Would still be great to see some thermal images tho

 

Not sure how one would do that, FLIR mainly shows surface temperatures, not heat flux. Maybe temperature difference of the heatpipes once the pad sheet is applied would be visible, but this should also be visible indirectly in CPU and GPU temperatures. Comparison to padding to the backplate would be tough because the latter needs the backplate mounted to work.

 

Well assuming mosfets spike all the way to 110 degrees, shouldnt that make a noticable change in the copper pipe's thermal profile?

Ive read a couple of posts of people claiming that bridging the fets to the pipe resulted in their temperature being kept within the power throttling margins.

As for the upgraded heatsink mentioned recently:

https://www.ebay.com/itm/new-for-DE...918241?hash=item4677944c21:g:nNIAAOSwI6RZvgRB

Ive ordered one here as they ship internationally. Will be interesting to see if changing the fins to full copper does help with keeping the vrm area a bit cooler.

 

Copper fins should help due to the higher thermal conductivity, but I think this is the cheaper version of the modded heatsink that I got from my taobao link. Would be cool to know if this one has the quad hear pipes as well

 

I was having this idea of solving the vrm thermal issue for good:

Applying thermal paste on the mosfets.

Taking a thin copper sheet and cutting out holes for the taller components.

Covering the copper sheet with insulating tape from both sides.

Cutting out the tape where contact surfaces will be established.

Removing a patch of insulation covering on the heat pipe and making contact trough a thermal interface.

Seems doable but I have doubts about the ability to retain it all securely.

 

Thermal paste + a highly conductive copper sheet probably doesn't have enough thermal mass to move a significant amount of heat from the MOSFETs. Not to mention how risky the entire operation is as there is no good way to secure the sheet to the MOSFETs or the heatsink. Any minor shift in the sheet could result a short occurring across the individual inductors and rendering your xps a doorstop.

Best bet would be a new heatsink bracket that has a vrm cooler integrated into it as no thermal paste (aside from liquid metal) has a thermal conductivity equivalent to that of a solid metal piece. Which I think is what someone is doing on the Level1Techs forum.

 

Yeah I saw that thread but he's making one for the 9570 which has a different layout of the fets i reckon.


How about securing an insulated copper heat pipe inside the underside lid and using thermal pads under pressure of the enclosure to link the mosfets to the main pipe? Main issue is making sure the pipe doesent transfer heat to the lid itself, and the thickness of the pipe that can be crammed in there...

what was the gap length between the lid and the components anyways?

 

Yes it was reported that the heatpipe temperature rises, which shows in the CPU and the GPU warmer by a degree or two. With FLIR one could this see relatively quickly how much the temperature changes. But for proper evaluation one should run load tests for at least half an hour, compare to the other simple method of padding to the backplate, and preferably make a well documented post here. Folks reported doing this on dell reddit, but the reports were a bit below the documentation standards around here.

 

I think this is possible, though wouldn't be a simple method anymore. The main suspicious thing about the thermal pad "bridge" is if it can conduct enough heat side-wise, considering that the pads have an order of magnitude or two lower thermal conductivity than copper. But as you say, cutting copper, fixing in place in insulating is not as trivial.

 

Doesent seem like there is a simple method to get those vrm temps in check at this point. The heat has to somewhere and it seems that dell crammed a 1050 inside just to show it in specs while expecting the customer to use it like a facebook machine... Bridging with a thermal pad seems interesting but questionable at the same time. Id love to see a thermal image of a thermal bridge during combined stress test.

 

If it works well, I'd be happy doing a drawing for the 9560, if you're happy DIYing it yourself (9550 has a different mount layout, because there were less GPU memory chips). Looks like there are some taller components that would mean the part cooling the VRM would only be able to connect back to the CPU area, instead of spreading more evenly.

I have quite a lot of people interested in the project. Might consider getting a quote from a Chinese manufacturer for a large batch, as there seems to be a good amount of interest. Would also be a good opportunity to refine it a bit, and possibly use taller heatpipes.

 

@_Zavar : Let us know how your project goes. I would be interested in one for my 9570 if it works.

 

Sorry for bumping this thread. Almost done with my heatsink.

https://forum.level1techs.com/t/project-make-the-xps-15-less-on-fire-custom-heatsink-mod/135326/50

@htaige, with you heatsink that has the two sets of heatpipes, does it make contact with the bottom panel? Mine works out at pretty much the same height, and it looks like the screws I'm using are the only problem.