XPS 15 9550 temperature observations (undervolt + repaste)

I'm with a completely stock 9560, no repaste, no VRM pads, no voltage and any other software tweak.
The laptop sits on a passive plastic stand that raises it a bit from the desk, the room temperature is 21°.
The laptop screen is turned off while I'm using an external FullHD display connected with HDMI.
I've set Performance Mode in Dell Command Power Manager and Nvidia Control Panel, I've turned off V-Sync in Unigine Heaven and launched the benchmark with Basic preset in windowed mode after the loop was already running from about 15 minutes and got 2965, WTF?
The GPU has been apparently stable at 1911Mhz for the whole run (al least, that's what Unigine Heaven reports), the GPU temp peaked at 72°, most of the time was in high sixties.
I'm surprised by the GPU clock, never seen such high value in any 1050 review.

Reading almost the whole post my bench result seems too high. May be I'm wrong with some setting and this result is not comparable with what people is reporting here.
Can someone please post settings and requirements to have a comparable bench?

 

Your score is in line with everyone else's.

There's a summary made by pressing with some scores on page 80.

 

Edit - opssss....see GoNz0's comment on testing with laptop screen.

Also, a few posts up someone mentioned the Heaven Mhz clocks are wrong; look at the clock info from something like Aida64 or HWiNFO64 or GPU-Z.

 

try it again on the internal monitor as it has to process the signal motherboard side as the panel is pure LCD no processing, you are handing this off to the external so it may cause a higher benchmark.

 

It's a common misconception that Heaven reports correct clocks...same with Valley benchmark. They report completely incorrect core clock values. They're usually WAYY too high. Mine will report 2002Mhz when its only 1820MHz....fyi. Check GPUz monitor when running the benchmark to get your real clocks...

 

I've just repeated the bench using the internal display and I've got 2956. Using Hardware Info I've seen the GPU core clock steady at 1695.5 (yeah, the Heaven reported GPU clock is wrong) the VRM peaked at 92° for a moment, was steady at 90° for most of the time.

I haven't opened my XPS yet, may be there's already a decent pasting job in it? Who knows, perhaps the factory was running out of thermal paste! Considering the actual performance I'm getting at stock is among the best I'm not sure if a repaste/repad would be beneficial, what do you guys think?

 

lol highly doubt the paste is good... well...guaranteed its bad. All dell's come with about 10x the required thermal paste... you could probably scrape around the CPU and GPU and repaste your chips 2-3 over with it...lol

 

Stock paste application works decently on the gpu, it doesn't on the cpu though.

You can easily repaste both the cpu and gpu (use the spread method and apply slightly more on the gpu's upper left corner since it mates bad with the heatsink), reuse the stock thermal pads on the gpu memory and apply some thermal pads on the vrms or the chokes to draw the heat from them to the back cover; this will give you the best results for the least effort.

 

Spread method causes air bubbles in most cases and on a direct die is unnecessary due to such a small area. You should only use the grain of rice method and make it a long grain on a wide die.

Even on lidded chips I would only go as far as an X method to make sure it reaches the corners.

And yes I know liquid metal needs to use the spread method but that isn't your typical paste

 

In any case, if I were in your shoes I might consider running ROG RealBench stress test and see if there is any throttling. And maybe Prime95 small ffts. Then try some games you like. Some preliminary thermal results are on p. 80 and scattered throughout this thread for comparison.

Each CPU, GPU, heatsink, factory paste job is different, so If you see no issues...

 

Debatable, some prefer one method, some the other one; i prefer the spread one since even if you get air bubbles in most case you still cover a bigger area than with the drop method, as for the 9560 on the cpu die it may be true that even the drop method covers the whole area since it's small but on the gpu i'm sure it doesn't (tested even with a fairly big drop).

 

Tryed with Prime95 small fft. everything is stock on a 9560 i7, no repaste, no pads, no software tunings. The Laptop is on a passive plastic stand that raise it a bit from the desk, room temp is 21°.
I've set Dell Command Power Manager in performance mode.
After about 10 minutes the clock is steady at 2992Mhz with short jumps to 3092Mhz for 2 cores, 4 cores rarely.
CPU temp is about 80°, the fans are spinning at 4080rpm.
Seems good to me.

 

I think too much paste may look ugly but should't be disastrous with normal non-conductive stuff, as long as the sink is well aligned and the surplus gets squeezed out.

What's the real issue with the poor factory job? Bubbles? Too much gap? The vram pads don't seem to be too thick, otherwise repasting wouldn't work. Just the screws not tightened enough? It'd be odd that they'd pick a bad paste to save 50 cents, considering the small amount needed...

 

Paste doesn't have the thermal conductivity of copper, the thicker the paste the more difficult it is to transfer heat. While I don't know the thermal conductivity of the stock paste, I'm sure it's not great (and yes they would save 50 cents on paste if they could).

Recap - bad paste job = too much crappy thermal paste

 

Here's my Small fft prime95 test. Just ran it on my lap desk..lol

Granted its apples to oranges with the i5 vs i7 but this gives you an idea of the CLP along with added cooling... I believe the DIMM sensor is the ambient sensor by the VRM...shows pretty low temps for this stress test. No throttling...

 

Ideally the

On my 6300HQ the first sensor labeled Ambient in HWiNFO64 is the vrm sensor. That fist sensor is also your hottest Ambient sensor.

You can confirm with a new version of Aida64 (which has the correced Kaby Lake sensor labels)...

 

I relabeled the sensors on mine. The VRM do get quite hot..however, they operate upto 120C so its not really anything to worry about...IMHO. Plus I'm not planning on running my rig on full tilt without vsync so I'm not too worried about it. I'm debating redoing the entire VRM portion of the thermal pads but honestly I might not do that since the top near the keyboard already gets pretty warm after this test....it might cool the VRM to the frame but it may in turn produce more updraft into the keyboard... Now keep in mind I was running +140 on the core and +260 on the memory... that's 1860MHz during this heaven test...


I did however get a 3108 score in heaven basic..lol

 

No disrespect but that just shows inexperience doing it as I get excellent coverage using the grain method on both dies knowing I am not leaving air bubbles that will expand when the dies are hot pushing paste out the sides

As I mentioned the debatable part comes in with lidded CPU's but it is not needed with direct die cooling as it is such a small area. Even with lidded the X method is proven not to introduce air bubbles while having the desired result.

Obviously it is your laptop and your choice but I don't want people thinking it is the best way to go about it as in this instance it isn't unnless you are using paste that specifically states the spread method.

I am only going on 20 odd years repasting and 10 water cooling. What do I know

 

I think the main issue is the paste is thick so doesn't spread enough due to the low mounting force or you would be able to make out the writing on the chips when you take the heatsinks off, instead it is still a blob of paste you can't see through. I expect this is why everyone seems to get better temps with a DIY repaste.

 

I guess those 30 years experience didn't teach you humbleness; to me it looks like your contribution to the 9560 cooling matter is constituted mostly by criticism (and not in the form of polite advices) to other people's work and tests.

 

I never head out to be rude so apologies if I have managed to offend another member on here.

Me, I just use as fewer words as possible to get my point across. Any opinion can be classed as criticism (and usually is by the person it is aimed at, especially when repeated) although in my case it is a generally held view shared by the majority of overclockers and some 9550/9560 owners I expect .

You should google pasting methods and watch the youtube video's on it. Here is a great example why you shouldn't use the spread method with your average paste.



While it will not be as bad on a small die it will still have a lot of air bubbles that expand each time the die heats up and we all know what happens when air expands when trapped, it will push out whatever is in the way. In our case the very thing we need to stay in contact with the heatsink.

 

Everyone has his way of expressing himself so don't be offended. GoNz0 contibutes a lot to this forum and ends up repeating himself a LOT so I see this as a bit of comic relief.

There are a lot of opinions around repasting technique.

As an example, my engineering buddy ran a supercomputer with thousands of consumer graphics cards. Needless to say he had full-time staff replacing & repasting CPUs & GPUs. He personally repasted thousands of GPUs and CPUs. He uses the card spread method and has the stats to back it up as reliable compared to other methods he tried (temps and time to failure). But he also told me spread method was not appropriate for me because it took people ~100+ tries for him and his staff to get reliably good results... He recommended the small rice method for regular laptop people.

I also found GoNz0's recommendation of repasting, closing everything up, then checking the quality of paste to optimize application to be very helpful.

 

I bet your m8 would be a great plasterer!

 

On this video der8auer shows most application methods and says that he prefers the spread one and he's not the only one, i'm totally open to new and better techniques (possibly backed up by test results) but i don't think that saying that my (or yours) method is the best one and everyone else's is the wrong one is the right way.

Also, my conductonaut is taking ages to arrive and i think i'll just use the liquid ultra in the meanwhile both to test it and test the k-5 pro to replace the pads.

 

But he is using paste on the IHS not direct die. The video I linked earlier shows it will cause air bubbles and it is backed up in your video as well, now that 1mm air bubble (that will no doubt be several) on a direct die could cause a problem as the chance in the grain of rice method not covering such a small area is minimal.

That video you linked has put me off ever using the liquid metal as you cannot apply it without introducing bubbles.

 

And i agreed with you on this aspect, what i didn't agree on is your teacher attitude since i can assure you that the rice method on my gpu didn't cover the whole die (and i used a pretty big drop of paste).

 

Well that's the trainer in me, my past employer said I had a direct and bullish nature making me perfect to train people in the field in the shortest time possible


I made the grain method cover 2/3rd's the length of the die to get full coverage. Just trial and error.

 

Playing Overwatch without vsync I am getting PL1 throttling on all cores. Is this the VRM throttlign?

 

I thought PL1 was just the CPU limiting itself to it's TDP?

 

thats what I thought, too

HwINFO lists my PL1 power limit as 18.750 W. That might be an issue. Shouldn't that be 45W? How did that happen...

edit: when I restarted, it was back to 45W. Is that what happens when the VRM temp is hit?

 

I think so but I did have a theory I didn't test that the iGPU counted towards the 45w?
Even though it is using the nvidia card it is still going via the iGPU so it may be the PL1 trigger?

 

It might. Just confirmed that I got it to throttle down to 9v by just running overwatch capped at 120 fps. Is this TDP or VRM throttling?

 

I assume it is based on my tests.

This is what I think happens. When the mosfets heat up, they get incredibly inefficeient on and exp scale. So they no longer produce enough power. I don't see exact spec sheets but similar mosfets might begin issues at 80*C or so. Either the vrm temp sensor adjusts TDP down and/or components sense some shortage of power and PL throttling begins...

 

Right. I've got 3mm of pads on mine but maybe another layer on top as they show in the pics...

 

A big pad over the top of the vrm is stopping any airflow whatsoever from getting around other vrm components. I'm not so sure that is a good idea, as the other vrm chips, chokes, resistors likely are less sensitive to heat but still suffer from heat...

Also, these chips use the board for some passive cooling so I don't think big pads help there either...

 

Yeah so now I see why it's juggling. Before I mated the top of the gpu to the bottom case so it would cool itself at the expense of the CPU. Now I think that nothing touching the case except those vrms would be the best way of cooling them.

 

The case bottom is a poor heatsink so gets saturated very quickly so you don't have the luxury of heatsinking everything there. Figuring out the "highest" value item to heatsink is a challenge.

GoNz0's idea of copper pipes from the vrm to the fan bases is the best but most difficult to implement...

 

I thought that the conductonaut would have taken some more time to be delivered (at least that's what amazon said) but it arrived today instead.

Observations:
- the k-5 pro paste does indeed solve the gpu-heatsink mating issue: i did a test application first by applying the same amount of paste to each module and noticed that the top right and bottom left vram modules totally squished the paste (as in < 0.1 mm space) while the top one and the left one barely touched the heatsink, the gpu was fully covered by the heatsink; on the final application (removing it is a bit annoying btw) i applied more paste on the modules that barely touched the heatsink and solved the issue once and for all. Highly recommended.

- the conductonaut: aim it somewhere safe for the first push, since it's slightly hard to press the first push shoots the liquid metal everywhere (fortunately i did and nothing bad happened). Covered the sensible parts with electrical tape and spread it both on the chips and on the heatsink; the cpu die is too smooth and it's impossible to spread it, i used the liquid ultra sponge to scratch it a little and then it worked.

Into the results now:
By the first tests i gained (as in temps are lower by) roughly 3 to 6 degrees on the kryonaut; temps during the realbench stress test didn't change much and it can't be passed anyways, it throttles at 5 to 10 minutes.
What i noticed is that cpu temps during the stress test climb slower but steadily, to me this means that the heatsink reached its cooling power limit and the chip transmits heat faster than it can be dissipated.

tl:dr k-5 pro is really good for the heatsink mating issue, liquid metal helps with temps but it doesn't solve thermal throttling (it happens on highly unlikely scenarios like the stress test but still).

 

9560 #2 arrives tomorrow so I will have a tinker with #1 to figure out what can fit before I return it.

You can use speedfan to set the fans to max to aid a realistic test as the fans may spin up slower/faster masking results.

Also the heatsink may be a little bent out of shape, either by the ham fisted person fitting the heatsink in China or it can happen if you try and remove it by anything other than fingers on both heat plates at the same time to break the initial sticking caused by Dells paste so it may be worth checking that the next time you are removing it. I managed it on my 9530 a couple of years ago and it was easy to straighten as the pipes are so soft. This 9560 has a really good fit on the pads.

 

The heatsink is not bent, it's new and it's definitely straight but it still doesn't have the same contact on the cpu and gpu, the cpu is really tight but the gpu isn't; i didn't test it with the stock pads but with any pad that is not silicone based (even 0.5mm ones) it doesn't mate well. K-5 paste solves this issue.

 

New has nothing to do with it, they paid the lowest bidder to make the heatsinks so there is a ton of slop in them. I've had maybe 1 that I would consider good (out of the 20 or so I've looked at).

 

Most of the times you can see if it's bent by looking at it; heatpipes are butter soft, it would take 1 second to make it straight.

 

if you can judge flatness by eyeing it, you're a better man than I. I had to compare it to my calipers. Also butter soft is true, also super easy to pinch/ruin...

 

Not had one yet that will sit square on my metal ruler....

Sent from my SM-G920F using Tapatalk

 

Has anyone tried removing the black plastic off the aluminum vent to free airflow?

 

Do you mean the plastic on the bottom case covering the center part of the vents?

Several have discussed but I don't see any posts of anyone doing it.

 

Yeah. If it can reattach, no harm in trying...

 

Potentially good vrm cooling if you:

-removed the plastic on the bottom case covering the center part of the vents

-removed some of the foamy pad on the back grille (by the f7 key and under the Dell screen sign)

-used a laptop cooler fan under the laptop to push air past the vrm and out the back

 

Just to compare what's your vrms temps during a stress test? My highest ambient sensor reaches 85 max and a lower average.

 

That plastic thing has turned out to be copper... so I can't really remove it.

I've put pads on the VRM to the case, but I still get PL throttling after about 5 minutes of looping the uningine test.