XPS 15 9550 temperature observations (undervolt + repaste)

Let us know how it goes, I have a tube of Thermal Grizzly Conductonaut that is begging to be used, but I think I'm going to delid my 7700k and use it there...

Forgot to mention that you should use electrical tape instead of nail polish, it's not as durable but it's easily reversible.

 

I'm actually going to be using neither! lol jk. Actually I'm using a very small strip of thermal pad around the entire die of the GPU/CPU. That way, even if there's pumpout, the thermal pad will act as a barrier so nothing can drip out. In theory it should work. And if I cut the strips thin enough it shouldn't have any problem getting squished down by the heatsink... The strip will be more like a rectangular cutout for the die to fit inside... And I'm pretty good at using liquid metal paste...so I shouldn't have any risk of pump out...but just in case.

 

I was going to use some CLU on my 9550, but I decided to try swapping the 512gb PM951 to a Crucial MX300 SATA drive first to keep my fan noise down. Holy smokes, the fans are so quiet during normal tasks now. I just moved 350gb of data over from an external SSD and the MX300 never broke 40c when the PM951 would have been pegged at 64 and the fans blasting.

It still throttles when playing overwatch with no frame cap, but at least my fan noise problem is gone.

 

Wait...I'm confused... you swapped out the drive and your temps went down and noise? I understand typical platter noise but temps? lol I guess HDD's do generate heat too....

 

I swapped out an NVMe SSD for a M2 SATA SSD. I have the 84wh battery. No HDD. Part of the fan noise activity is a result of the SSD temps... not just GPU + CPU.

 

Fixed!

 

Anyone OC the GPU memory yet??? Results?

 

For real? everyone is battling the GPU/CPU from throttling and people still ask if we have overclocked it

In other news I watched my DIMM (aka the VRM sensor in this build of Aida64) hit 98c. I have to figure something out to reign it in without trying the chassis.

 

You can run GPU-Z on the "sensors" tab to visually see 3 gpu clocks and voltage sag real time in small charts

 

Yes! lol! I plan on going to extensive lengths to improve performance including thermal performance. For starters, as you know by now...liquid metal TIM. Thermal pads galore... Now where's my leaf blower....

 

Just wanted to mention I ran 3 basic heaven benchmarks back to back and received a score of 2741, 2731, and 2741 respectively.

I also wanted to add that my repaste was with arctic silver 5 (apparently this stuff isn't that good anymore?), however I am going to try a repaste with Kryonaut and give that a try.

GoNz0 I'm sure you already know this, but in addition to dropping the turbo boost etc., reducing the max CPU performance in power management to 80 has helped quite a bit (or just sub 100). But sadly still not enough for me in many situations


Edit: Just encase I didn't mention it previously, my CPU power has been at 80% with all of my previously listed benchmarks. Also Pressing you were correct - I7-7700HQ

Edit: Dyslexic moment: I initially wrote 2471 as my 3rd heaven benchmark, but meant 2741

 

Thanks Ginglymus for the update.

The big undervolt you and G0Blezz are running should give you MUCH more thermal headroom than GoNz0 has. Looking at the summary linked below, it seems like something backfired:

1. huge VRM thermal pads are not working as expected

[and/or]

2. GoNz0 did a better repaste (likely because he is professional computer engineer)

I suspect the huge VRM pads are not working as expected and might work to reduce the amount of pads there. Maybe start by just padding the GPU chokes?

Good new is that I think you should be able to exceced 2800 significantly if you can get this sorted.
(basis - GoNz0 hit 2800 so has upside if he kills vrm induced GPU throttling & undervolts CPU)

 

I hit about 280 earlier from a cold laptop, back was propped up as it does let in more air. Just a gelid repaste. my heatsink is nice and flat on both dies as the quick test fit covered both dies in a transparent film, the room was about 28 degrees and I have no other mods done. CPU seems to bounce around 80 and the start of the test is a voltage limit on the GPU and 2/3rds of the way through it does a temp limit. CPU sits at 34-3500 during the test. It went to temp limit at the halfway mark before.

VRM's steadily clock up to 99c, after one run the VRM's will probably be too hot to run the same score twice.

 

Has anyone tried just padding the chokes vs putting pads over the whole forest?

 

I tested quite a few configurations with pads on almost everything and think I now got a sweet spot for me. I just ran Heaven Benchmark twice and got 2834 on the second run (5min after finishing the first). No throttling or power Limit. CPU was around 76°C (with max. 81°C) and ambient about 80°. GPU was at 70°C and running at 1680Mhz the whole time.

My Setup: undervolted by 120mV, repasted with Kryonaut, pads on all chokes (GPU and CPU) and the chipset. And one additional pad on the heatsink over the GPU.

Before I had them nearly everywhere. Even lifted that small black metal shielding over some more chokes on the board. I put them all away as they got to much heat to the CPU/ambient.

I will test later with Realbench and Watch Dogs 2 which seems to drive the CPU/GPU a lot.

But it seems as padding the chokes, repasting and the undervolt did it for me (by now )

 

Good to know!!!! This is great! One thing you might want to consider... you might get a slight increase in thermal dissipation through the thermal pads if you use a small amount of thermal paste on the actual components prior to adding the thermal pads... This helps reduce and additional air gaps that might reduce thermal dissipation...just fyi! But then again that might make your mobo messy with all the paste everywhere..lol

Sound to me like you definitely don't want to use a thermal pad between the heatpipes of the cooler to the bottom of the case... you want to let the heatpipes draw the heat away to the cooling fins... I almost wonder if insulating the heatpipes might actually reduce the overall ambient temp near the VRM... maybe a few strips of electrical tape around the heatpipes...?

 

Can you take a picture showing the pads you've got on the chokes? And the black metal shielding, you mean the plastic tape?

 

Thanks for your detailed posts - indeed you are a rockstar!

Rockstar - GoNz0 mentioned his GPU was voltage limiting at beginning of Heaven. Did you have that issue? You could see that realtime in the GPU-Z program in the sensors screen.

G0blezz - Now with 3 Heaven samples in the 2800 range, the 2294 indicates your systems has some issues with his config or GPU...

Updated summary sheet is linked below.

http://forum.notebookreview.com/thr...ndervolt-repaste.785963/page-80#post-10463488

 

Did some Tests and it´s getting funny. Maybe someone could explain
In comparison to my last post, I now took the pad on my heatpipe above the GPU away. Heaven Benchmark is still the same but now my ambient temp is a little lower (ambient at 78°C max).
Funny thing is after 15min Watch Dogs 2 the PL1 Power Limits to 15W. GPU is still running at 1670Mhz and CPU is down to 2000Mhz and shortly after that at 2800Mhz and after about 10min later at 3Ghz. HWiNFO reports "Ring Power Limit exeeded" (and all cores too, of course), "Ring Max VR Voltage PL4" and "Package Level RAPL PBM Pl1" all true ("yes"). Temps were about 78°C CPU and 84°C ambient, GPU at about 70°C
There seems no to be thermal issues but something triggers the throttling. But what?

Here´s my setup:

Spoiler: Padding

I removed all pads from the heatpipe as they are driving the heat from the chokes to the CPU and my temps got worse. After removing temps dropped about a delta of 6°C. SO the idea of isolating the heatpipe from the ambient is interesting and may work, but how to get the heat away
The metal shielding is right on the left side of the right fan. I took it of and padded it underneath and then to the bottom case. Temps got worse in ambient so I removed it.

 

The laptop BIOS has to juggle the wattage of CPU and GPU so it can go a bit overboard at times trying to bring things into line again, I assume it will voltage limit as part of it's thermal configs when all the sensors start getting on the high side.

Of course ambiant was higher when you were using the chassis as a heatsink, it will be warming whatever sensor is near the chassis base.

Glad to see you have seen sense and starting removing your plethora of cooling accessories, Dell didn't do a great job but it isn't bad considering the ultrabook size of this, only minimal tweaks can help such as a repaste. Why not try removing some of the copper shield covering the middle of the vents to see if convection kicks in and wicks the heat away from the VRM area.

 

Does make sense but then it could be hard to resolve that problem...
The idea of removing the copper sounds interesting but it may not be reversible. Did somebody tried this already?

 

People have discussed cutting or removing the copper-mylar sticker on the case bottom but one has reported doing that here.

Some of these tweaks are a balancing act and will improve one benchmark and hurt another, like you are seeing between Heaven and Watch Dogs. MSI real bench stress test is a decent balance between CPU and GPU.

Watch Dogs seems more CPU intensive. The 84*C vrm temps are likely driving your PL1 issues. I think Dell triggers PL1 with vrm around 80*C. Also the mosfets are getting inefficient around 84*C so voltage sag-choppiness will be another trigger.

1 - You may need to get vrm temp down a bit
2 - I had low vrm temps. So got around PL throttling with Intel XTU unlocking TDP (increasing turbo watts to 66w & time to max - both for short/long).
3 - Keep a close eye on thermals because you don't want a meltdown

 

Only way to really tell what is causing the throttling is to get the infrared thermal gun out or FLIR camera and check temperatures under load... Without having the ability to do this it's going to be a guessing game... Gonz0 noted the case temp sensor by the top of the VRM bank which is great..but we also need to identify each component of the board and how its being effected under load...and where the throttling point is... then finding a solution. I'm planning on recording temps under load with the bottom open...hopefully I'll be able to create a detailed diagram of each component and the heat under load...then find a solution to reduce temps...


Oh and the VRM...mosfets... I looked up the Texas Instruments mosfets similar to the ones used in the dell...they're rated upto 150C...so I don't think we have much to worry about with meltdowns... I would think copper traces would fail before a Mosfet or choke...

 

That sticker can easily be stuck back down, it should stay sticky but if not some 3M 9088 tape will stick it back down.

 

Aida64 will be fixed in the next release, current release is 14th feb in the beta section so any build after that.

 

True.

But the mosfets are the only components that really heat up (and are very sensitive to heat above say 80*C).

The power IC and caps just don't get hot.

The chokes heat up a bit but are quite robust.

The idea of heatsinking the chokes was because they are bigger closer to the case bottom, so much easier to pad than the mosfets. Also that would not restrict the minimal air circulation that exists in the vrm area. And fact that case bottom is a poor heat sink so really can not handle much heat...

 

I think a thin copper plate stuck to the chokes with thermal epoxy will be the only realistic solution without cutting holes in the chassis.

 

Fixed your post. lol

 

I thought about that but the thermal epoxy is permanant and given the warranty issues with these things decided against it. Also you would still need some thermal pads between the copper and the case bottom to help absorb case flex or risk knocking chokes off the motherboard.

As a proxy, I used some super 17 W/mk thermal pads between the 6 chokes and case bottom. The case bottom got super hot in 5 minutes and the keyboard was uncomfortable to use. Really cooled the vrm area a lot however! (well for a few minutes anyways as case bottom got saturated with heat, it no longer was a good heatsink)

I think two guys used cheap thermal pads (6W/mk?) on 6 chokes and got a good balance for gaming. They are still using that as far as I know and it is probably the best all-around vrm solution to date...

 

= China china china!

 

Quick question for anyone: Playing around a little bit today, and after a repaste (still arctic silver 5, waiting for kyronaut to come in the mail) I had much higher CPU Graphics temps. For example, after a real bench stress test my max core cpu temp was 78, max GPU temp was 77, while the CPU graphics cores were at 93 C. My question is: Does the majority of the integrated graphics heat manifest within the CPU itself, or can that heat manifest at the VRAM next to the GPU? Hopefully this makes sense

(essentially trying to figure out if the integrated graphics temp increase is due to a poor repaste or from the VRAM - I moved one pad as it was shifted, and had a cut in the middle so i'm trying to figure out which is the likely cause)

 

If you are new to repasting then do 3-4 test runs aand make sure the dies have a nice thin even layer before repeating a final time. You may have bent the heatsink removing it so it may need tweaking. Also pads that are thicker than the gap will stop it mating.

 

Have you considered using TS to clamp the CPU TDP at 35W?

 

After pressing mentioning the mosfets, this does make sense I put some cheaper pads on the mosfets in the heatsink area as well.

Thanks for suggesting! I did set it to TDP=35W and let Realbench Stresstest run. There was no PL1,2,3 Powerlimit. CPU sitting at 3,3-3,4Ghz and CPU temps in the mid 80s. Ambient about 80°C and GPU at 74°C/1680Mhz.
After about 5min it got down to 1350Mhz (standard) Voltage drops to 0,8V. So the GPU is limited in powerconsumption as pressing mentioned. I will test Heaven and Watch Dogs later with this setting.

But I´m really curious about the thermal behaviour of the system under load


Edit: I ran heaven benchmark with TDP clamped at 35W and got 2822. No limits or throttling and temps all below 80°C.

Is it possible to set TDP with XTU?

I think I stick with this setup as I am getting bothered opening the laptop


EDIT: The TDP clamp seems to stick after several reboots without starting throttlestop!
Thanks pressing for your hint!

 

If you are using thermal pads on the mosfets, I would remove all the other thermal pads; you are heatsinking the only problematic parts in the vrm. All the other vrm heatsinks are just hindering mosfet cooling. That might allow you to remove the TDP clamp and see higher performance. Post some pictures when you have a chance.

For reference, running the RealBench StressTest on a 9550 with 6300HQ, I get no throttling. I can run the 15 minute StressTest several times in a row with no changes:

- CPU max temps ~72*C
- GPU max temps ~78*C
- vrm max temps ~75*C (no pads in vrm area).

I don't see a TDP adjustment in XTU for SkyLake processors.

 

I thought that too and took the pads on the chokes away. I don´t remeber exactly the temps but decided to put some cheaper pads back on. Trying out Watch Dogs with the cheap pads on all chokes and mosfets seems to work for me now. The botton keeps colder and the game ran well without limiting (just setting TDP to 35W). CPU stuck at 3,3-3,4Gh and GPU at 1670Mhz.

Me neither...I really like Throttlestop but get my undervolting done with a script and XTU since my 9550...maybe I have to Switch.

Edit: The TDP clamp seems to stick after several reboots so I could stick to XTU for now...

 

Do you pad all the way to the case?

 

Yes, I found some 5mm pads that fit well on the mosfets. I´ll take a picture next time I open it up.

 

With the 17W/mk thermal pads just on the chokes, my keyboard and case bottoms were scorching after 5 minutes. Unusable.

The lower rated thermal pads seem to be more practical for vrm cooling

 

Hi all !

After reading the feedbacks and intervention that some of you did on the new 9560, I'm currently planning mine

It seems that the solution from GoNz0 is the best ones if you don't want to undervolt anything ?

So just to be sure of the intervention, you repaste both CPU and GPU with GC Elite, and then applied some thermal pads on some hotpoints ?

Do you have some pictures to share the important parts to cover ?

Last question, someone mention a little black adhesive protecting the heatsink and was wondering if it could be link to the warranty, any intel on this ?

Your repaste doesn't modify anything on the warranty ?

Thanks a lot

 

Repasting doesn't invalidate the warranty, damaging it during does obviously.
the sticker is not a warranty sticker and can be lifted on one site to gain access to the screw. I didn't need to apply thermal pads as increasing the conductivity of the GPU Ram makes the ram get hotter as it will take the heat from the CPU as it passes by.

 

i'm back with some results (some are good others are bad).

The gpu wasn't evenly covered by the heatsink, as you can see from the thermal paste in this picture.
What i did to mostly solve it: using 0.5mm thermal pads on gpu memory instead of 1.5mm ones helped, then i put 2 small slices of 1.5mm thermal pads on top of each other(for a total of 3mm) on the heatsink where the gpu part is to push the heatsink closer to the gpu a bit.

I did try with another heatsink to see if mine was bent and the issue was the same so i suggest you to put a really thin layer of paste on the gpu die and carefully check if the heatsink evenly covers it.

This helped me lower gpu temps from 77-78° average to 68-69°, now the gpu doesn't throttle and i managed to get an unigine heaven basic score of: 2968
Great success.

Bad news on the cpu side instead, in my previous posts i stated that i could easily complete a realbench 15 min stress test without throttling; well i wasn't lying but i could only do it because the heatsink wasn't covering the gpu well and the cpu had much better cooling power, now that the gpu is covered the cpu needs to share the cooling power and i can't complete a stress test without throttling anymore (it starts to throttle at around 8 mins).

Did anyone complete a test without throttling?

 

Forgive me for being harsh but!.

WTF?

The GPU made little to bugger all contact, I thought you did test pastes?

You really need to put that back to stock pad wise and see how you get on with a proper repaste.

 

I did multiple test pastes, that's how i learned of the issue

Stock pads are even thicker than 0.5mm (but squishier).

Also as i stated i tried it with a brand new heatsink and stock pads and it didn't make much more contact (it looked like it made more contact only because the stock heatsink comes with like 50g of paste attached to it), that's why i suggest people to look out for it.

 

Gosh that GPU contact is basically non-existent.... I'm planning on doing it right and taking my time...especially with liquid metal paste it's gotta be right...
Give me a week and I'll be able to report back with proper paste and mounting results...

 

Looks like good progress; best Heaven score shows your GPU is performing fine

- The risk of vram thermal pads being too thick and/or too dense is that they prevent proper heatsink contact to the GPU. I think that is what happened to you. Really fluffy vram pads + trial and error to get proper thickness reduces the risk of that problem. That complexity is likely the reason GoNz0 recommends keeping stock vram pads unless there is a defect. There were in a large percentage of 9550s with heatsinks not fitting properly over vram hence the thermal pad mod; no defects demonstrated in the 9560 heatsink fitment to my knowledge.

- GoNz0 summarized his tedious repasting process which worked wonders for me. You can search the post on this thread. Basically, clean the CPU, GPU, heatsinks. Repaste and tighten down the heatsink screws. Remove heatsink screws to assess how well paste was applied. Clean and repaste and make other adjustments as needed; adjustments could be more/less paste, different application method, reduce vram pad sizes, carefully straightening out heatsink (danger!), etc. Redo until you see good, very thin distribution. Plenty of photos here and on the interwebz. His steps make process almost foolproof.

[EDIT] I also run the laptop after each test repaste to see how thermals are. Paying particular attention to temp at each core. If one or more cores are significantly hotter than others that might indicate repaste is uneven or heatsink is bent or pressure on screws is not same...

==> I have not yet seen a 9560 RealBench StressTest without throttling. It would be helpful if a few 9560 owner could pass the 15 minute test merely because it would provide a simple goal to assess CPU performance and CPU repaste quality...

 

Thanks for your answer.

Indeed I misread the message about your modifications.

So you don't think apply thermal pads on either RAM or VRM chips could bring down even more the temperature ?

 

You will/may/probably/in theory should increase the chip temps using better thermal pads as the heatsink heat will travel back. Nothing wrong with mine after extended testing, pissing about with various pads/thickness/ "W/mk"

If the pads fit, great, leave them alone, even the dodgy looking one with a corner cut is fine so long as most of it makes contact.

 

Well, I finally defeated Realbench, and I am completely happy with my current setup. I am not only able to run real bench without throttling, but I can run a 30 minute stress test without throttling with a max sensor temp reading of 72 C. Before this, the furthest I had progressed into Realbench over the last 3 or so weeks was ~600 seconds before throttling. I should also mention I am using a laptop cooler during this benchmark.

HWInfo64 Graphs: http://imageshack.com/a/img924/159/Awohoi.png
HWInfo Monitor/Sensor display: http://imageshack.com/a/img922/5653/kCFfHK.png

I made 3 changes, with the latter 2 being completely new:
1) I re-padded my VRM with larger/wider pieces rather than piecing in like a puzzle (same 6W/m arctic thermal pads). This was needed after me opening my case 300 times, or at least it felt like that many times! The pads were exposed and touched too much during this whole diagnostic process.

VRM Picture 1: http://imageshack.com/a/img923/857/ErQbV1.jpg - This is 2 layers of Arctic thermal pads 6 w/mk, each 1mm. Total of 2mm - which seems to be just about flush with the top of the chokes.
VRM Picture 2 http://imageshack.com/a/img922/5884/Tw46VD.jpg - This is an additional 2 layers of the same thermal pads just different dimensions (1mm each) for a total of 2 mm from the top of the chokes to the case bottom.


The 2 following changes were completely new, and I think are what really allowed me to make progress:
2) Changing Intel(R) graphic settings - --> Maximum Battery Life rather than maximum performance in power options. This change dropped my CPU integrated graphics temp by a lot, unfortunately I don't have on average how much, but it was substantial. After doing this I also received my highest Heaven benchmark of 2813. I believe my Integrated graphics cores were really one of my main thermal issues.

3) Padding my physical memory/RAM (shown above in VRM picture 2). Despite the temp not reading very high, I think it was typically in the 60s under load, I just happened to touch it at one point with the bottom off and it felt incredibly hot. The temp sensor may have been accurate, but I decided to pad them anyways. This substantially increased the time it took for me to throttle on real bench. Each square was 2x1mm pads stacked for a total of 2mm.

Ultimately i'm pretty excited about this. One thing I want to mention that i've observed, is that temperature sensors are only indicative of thermal throttling to a point. For example, there were many times where I would thermal throttle with all of my the temps in the mid low 70s. There were other times where It didnt throttle until some of my temps were high 80s. The reason I mention this, is that just because a specific temperature sensor is not high, there might be another spot on the motherboard that is much hotter, but it isn't shown. The hardware senses that it is hot, and so it will throttle, because it is not at the same location as the temperature sensors that are reading in HWInfo for example.

Here are my current setup/tweaks on my XPS 9560 I7-7700HQ 32g RAM w/ 4k panel:
*CPU Usage in power options: 80% (min and max)
*Intel Graphics Settings in power options: Maximum Battery Life
*XTU: Turbo boost time 96 seconds, Turbo boost max 70W (I'm sure this would work exactly the same at 65w or even lower - who knows)
*XTU: -100Mv undervolt for CPU,Cache, and Graphics.
*Dell Command Power Manager: Ultra Performance
*Turbo Boost: disabled in BIOS
*Thermal padding on VRM and RAM.
*Clean installation
*Repasted my CPU and GPU. (I just received Kyronaut in the mail, however im a littel bit skeptical to mess with anything right now so I think I may wait and keep the Arctic Silver 5 on it for now)
*Stock VRM Pads (although I was considering replacing with Fujipoly pads, but I think I will wait on that idea for now)
*Laptop Cooler: https://www.amazon.com/Zalman-Minim...ncoding=UTF8&refRID=KP9GJCSXT2RRC66MH6YW&th=1

That is all I can think of at the moment, and I will add more if I can remember anything else.

Edit: once I made the change to the intel graphics power options, I was able to uninstall XTU, and raise my CPU back up to 100% without throttling.

 

Ok then, I'll go with the repaste