[CPU + GPU Temperatures + Benchmarks] - XPS 15 [9560] Kaby Lake

Greetings all,

This thread is to focus on the CPU and GPU temps, along with Benchmarks with certain apps that will provide a consistent platform we can all work off of.

(3DMark - Fire Strike, Time Spy + wPrime v1.55 etc...)

-

Download Links:

HWiNFO64: Click here.

CPU Test (wPrime v1.55)
Make sure to: Advance Settings -> change your core count to 8 before running the tests. Please run both the 32M and 1024M tests w/ HWiNFO64 running and screen shot the temps, package power, multipliers etc...try to set it up similar to the info that I show on my screen shots.

GPU Test (3DMark): Click here.

-

My XPS is the 7700HQ / 4K UHD / 32GB RAM / 1TB Samsung 960 Pro / 97WHr Battery

• Repasted with Liquid Metal (Grizzly Conductonaut)
  
• Repadded with Fujipoly 17.0w m/k
• Light Heat Sink Mod/Alignment
• Tuning in progress...

Note: As much as I wanted to see the GTX 1060 in the XPS, having a light of flight time with the AW 13 (w/ the GTX 1060), I can say for sure that the 1060 won't survive in the XPS chassis. It makes sense why they went with the 1050 in this, although it'd be nice to have something bigger. - Anyhow, the heat sink/pipe in the XPS is barely big enough to handle the thermals of the 1050 as it is. The reason why I went with Liquid Metal was to see if it would actually make much of a difference in this small chassis. I wouldn't recommend LM, because it's a bit tricky to work with and from what I've seen, traditional paste on the GPU works just as well if not better than LM. As for the CPU, it could use all the help it can get so that's why I went with LM to see if it actually does better than traditional paste.

I've swapped out the Killer 1535 for an Intel 8260 WiFi card and did a fresh Windows 10 install, but before all of this I had ran some benches to get a solid base, before I had repasted the system. ( Repaste complete.)

Although this is not my primary rig, I will be using this as my daily driver along side my MBPr when I'm traveling so this is why I was eager to get my hands on a XPS again for gaming (Overwatch mainly and KF2) while on the go. Having previously owned the 9550 with the 960M, it (the 960M) was the reason I got rid of them both (I had two).

Although I do have my AW17R4 (GTX 1080), sometimes I didn't need a monster truck to go pick up a few groceries. Plus, who doesn't love having a variety of cars.

-

Please fill out the form here.

View the spread sheet here.

-

If you're looking to repaste:
XPS - Repaste Overview, by: @custom90gt

There really isn't much to it as it is very straight forward. If any of this is out of your comfort zone, then I can provide an in depth guide upon request, but the photos in @custom90gt's thread should be more than enough to get the gist of it.

-

More will be added so stay tuned and let the games begin! Let's see what the GTX 1050 can do in this thin chassis.

=====================================================================

Updates: 3/10/17

So it seems that traditional paste will be better for the GPU. I'll keep yall posted with any new findings. Boy is the (die) of the GPU so tiny lol...

While gaming the GPU bounces from 76C to 77C. So far... working to get this lower.

-

Update: 3/13/17

K guys, so I've been working for the past few days trying everything under the sun to tame the GPU temps.

The CPU temps after my liquid metal repaste are excellent. They are in the low to mid 60C's under full load.

During games ( stock) the GPU would max out at 80C to low 80C's.

After my liquid metal repaste, the GPU temps now lock at 76C under full load. Some improvement, but I want more lol... Especially having worked with numerous systems as you can see in my sig.

I know we're dealing with a small cooling system in a 14" chassis and one thing is for sure in that the thermal conductivity / capacity of the cooling system is definitely being reached when it comes to the GPU.

It may be a long shot, but my goal is to get the GPU temps in the lower 70C's and gosh even upper 60C's would be stellar, but I just don't see that happening without some major work or a miracle with the right combo of things...

I've even considered modifying the heat sink unit all together in adding bigger fins and heck even a bigger fan that can pump out more cfm's (maybe a fan from another Dell line of laptops that is somewhat close, but the XPS fans are so small so that's a bit of a stretch). I really think it's more of the heat pipes themselves and the exposed VRM's / FET's etc....than the actual fans... that's at least what I'm thinking... I may try to double stack the heat pipes, but then I'm thinking, would that be too much heat for the fans to be able to pump out?

These fans are (both of them) are rated at 5.2 cfm.



So after spending hours and hours over the past few days on this...I'm really going to crack down and find the culprit to get this XPS to reach its maximum cooling potential, whatever that may be. I just know that there is some headroom and that's what I'm after....

Fire Strike Combined Test LOOP is running and everything is at full tilt...


---

Update: 3/14/17

Here you guys go! Now that you know where the monsters are....CCHHAARRGGEEE!!!

After running through varies tests, I'm getting really close to finding the right combo.

Note: It's not about lining the bottom randomly with thermal pads, as that will only block air flow and make things actually warmer.




Special shout out to @pressing, for being just a cool guy and fun to bounce ideas around with... all them late night discord brainstorming sessions haha... Great stuff mate.

---

Update: 4/6/17

The moment some of you have been waiting for.

Thanks for being so patient. I just had to make sure to be thorough with some extensive testing to ensure that my theory worked/works.

Here it is!

------------------------------------------------------------------------------------------------------------------------------

>>>THE XPS FiX<<<

I'll try to keep it (TL.DR) Free as much as possible.

Issue: GPU Throttling due to the FET's being exposed and not being cooled properly.


The XPS could use bigger fins as well and a larger heat sink area.
-

Notice that there are:

CPU, Ambient, Ambient, Temp3, Ambient, Ambient, Ambient and DIMM?

The first, "Ambient," is the temp sensor near the FET's.

105C. Ouch.



-

Those FET's under full GPU load were in the upper 90C's. Just crazy. Why did they leave it exposed I wonder? Those should always have proper cooling...


-

Those are the ones that get very hot and these little aluminium heat spreaders can be found pretty much anywhere.


-

Here's the MacGyver moment that I had after staring at this thing for hours upon hours (accumulative throughout the past few weeks) wondering how the heck to tackle this thing...

One thing for sure is that this lining makes temps worse. I've tested this over and over again to be sure under many controlled tests.

At least there is a use for this shield if flipped over haha... read on and you'll get the joke...


-

You want that area as level as possible....because...


-

This is how I had checked the clearance with using my LED flash light.


-

Get that area level as possible to ensure the best possible contact later once you...


-

Ok so here is that copper shield flipped over haha.

The black side will actually serve a purpose here in shielding the heat from saturating the bottom lid.

Therefore, the theory (well now proven) was to isolate the heat transfer through its own isolated channel / medium (this copper sheet.) - And...it worked / works...


-

I've used a little double sided tape to keep the copper sheet in place for when I was measuring the clearance.


-

So that area where I had mentioned above to make level as possible? Here's why...

I've marked the places that would need slight pressure from the lid to push the copper sheet down, only on those targeted areas.

So the theory was to have the middle make contact, while creating a slight arch to the ends (left and right) from the center area where the heat would dissipate from the fan air (heat travels from hot -> cold). Hope that makes sense.


-

That copper sheet flipped over fits almost perfectly. It was meant to me...


-

Okay so now the fun part. If you have some extra thermal pads laying around, great. If not, no biggy. I've used thermal pads because they are squishy and also seals well against the copper side of the sheet as it has some adhesive on that side. These are 2.0mm pads that worked perfectly for what I needed it to do.

The theory (proven) here is to create a channel of air from the fan to blow toward the center. I have isolated half of the area (width) of the fan area to dedicate that for the air channel.

Also, since the thermal pads are sticky and flexible, when you place the copper sheet on top of this, it'll create a nice seal.


-

Here is a shot showing both air channels.

Note: At first I was going to use a small tube and cut a whole on the side of the fan blade; and from there route the tube toward the center to have the air blow out through the rear in hopes to cool the FET's. However, after a lot of tests, the air channel method works very well and probably better.


-

Apply the copper sheet on top and align everything to get it ready for taping...


-

It's important to seal off the areas that really need to be air tight, otherwise it'll lose its efficiency. I've trial and error-ed this many times with numerous scenarios and yes, it has to be near perfect for this to be effective.


-

And there you have it folks... Crazy I know, but I can't tell you how many all nighters I've pulled to think of how to tackle this with the limited space we have to work with on this thing...



Enjoy...oh and...does it work?

-

Results?

That first, "Ambient" sensor / FET' Temp .... now 69C from 105C.

No more GPU throttling.

CPU Avg. Temps: Mid 60C's
GPU Avg. Temps: Mid-High 60C's

My unit (XPS 15 9560 - 7700HQ - GTX 1050) has been repasted with Grizzly Conductonaut, Repadded, HS Aligned / Mod, and Tuned.



Yup. Very well...

Update:
Here's what a random reviewer said about using these heat sinks on his FET's.

"Used these to cool the MOSFETS on my mATX motherboard, before they were jumping up to 120C now down to about 85C. Prevents the motherboard from throttling the CPU down due to excess heat."

So around ~35C drop in temp for him...and guess what, about ~30C+ drop in temp for me so totally spot on.

-

As promised I haven't forgotten about yall...

Right now I've been heavily focused on the new AW 17R4 (Kaby Lake) so if you want to check out what's happening, feel free to peek in to check out the latest.


Th umbs up if you like this. Oh wait, this isn't YT LOL...j/k

Cheers and Happy Modding...

 

CPU TEMPS:

~18C drop in Temps after repaste.

Note: I have a feeling that the heat sink is not adequate to keep the GPU cool. I'll be looking into this closely, but do note that traditional paste may be better for the GPU since the (die) is so small.

BEFORE repaste, bone stoke out of the box:


AFTER repaste with Liquid Metal (Grizzly Conductonaut): High 50C to Mid 60C's

 

Gaming Temps: Overwatch

~10C drop in CPU Temps after repaste.

Note: I have a feeling that the heat sink is not adequate to keep the GPU cool. I'll be looking into this closely, but do note that traditional paste may be better for the GPU since the (die) is so small.

While in a game the GPU temps bounce between 76C and 77C. Hmm....I'll keep you posted.

BEFORE repaste, bone stoke out of the box:


AFTER repaste with Liquid Metal (Grizzly Conductonaut): Mid 60C's

 

Benches

Fire Strike: 6047
Fun Run...just getting warmed up...




Time Spy: 1976



wPrime v1.55: Full run with CPU at only 60C.

 

Guide

 

Q&A

 

I think we should add a section of what type of modification's we have done on the spreadsheet (i.e. pads sinking the VRMs to the case, CPU/GPU sinked to case, etc)

 

I agree. I'll get one up shortly. Would you like to help with some ideas for the form that I'll be putting up?

I was thinking to keep it simple to use Fire Strike (regular) for the GPU bench and wPrime v1.55 for the CPU bench as this app is very thorough and realistic to real world CPU usage, unlike those crazy ones that are over the top and not very practical for real world data.

I'll get a link up shortly so that we can all pitch in.

 

Looking good how far do you think you can push this thing?

Sent from my SM-G935F using Tapatalk

 

Thanks.

With the XPS 15 Kaby, @Pete Light clocked in 6069 so far on Fire Strike so hoping to stretch it into the low 6000's.

What's crazy is that the highest recorded 7700HQ/GTX 1050 score recorded is: 12751 (http://www.3dmark.com/fs/11865064)

Crazy right? That can't be the XPS could it? It's a Dell though...lol...12K is just wild.

 

Interesting. 12K would be crazy, probably never going to happen. That must have an Inspiron 15 7000 on LN2 or something

Sent from my SM-G935F using Tapatalk

 

I know right? That's the first thing that I had thought haha...

BTW just broke 6K in Fire Strike:

 

Clocks? do they hold up during continuous stress without any throttling?

 

I'm running a Fire Strike loop with the graphs on hwinfo64. As soon as I have all the data I'll update the OP(s).

The new XPS reminds me a lot of the AW13 with the GTX 1060. It's just so small and compact that it can barely handle cooling the chips....although the CPU temps are great for what it is, the GPU I'll be working on more extensively throughout the day.

 

Your CPU volts seem really high... I know the 7700HQ is a beast seeing that physics score... definitely beats mine..
http://www.3dmark.com/fs/11907972

I can hold steady 1860MHz on the GTX 1050 during gaming and benchmarks. Also my CPU undervolt of -0.145 reduces my CPU package power by a full 10watts on the CPU alone... you need XTU and some good undervolting. I see like 0.950v under load in games and something like 0.895v in prime95...

 

Hi,
got the 9550 i7, initially pretty miserable but now okay repasted with Cryonaut. I had thoughts about Conductonaut but seeing things out of alignment wasn't reassuring.
Mind explaining a bit more about heat sink alignment you did?
I'm curious why LM wouldn't be better for the GPU. Wouldn't LM help dissipate the generated heat at lower temperature? Alignment problem (LM not filling the gap everywhere)?
Unlike you, most gamers seem to focus on making the GPU run as fast as possible, perhaps even throttling down the CPU (iGPU) a bit intentionally to avoid joint thermal long-term overload.
For work, some folks would be better off with getting some more out of the CPU, which might be feasible when the GPU is idle.

 

There's a secondary GTX 1080 there

 

I'm just getting started and haven't even got to tuning yet. There's always a bit of head room on these CPU's as Intel over volts the lights out of them to create a buffer for stability. Once I get this thing running at its peak, the results will be updated in the OP(s).

 

If using traditional paste the only paste that I would recommend is either Grizzly Kyronaut and/or Gelid Extreme.

I've had better GPU temps on my gaming laptops (AW17R4's, 15R3's and AW13's) with traditional paste over liquid metal, however, with the GPU (die) being so small on the 1050, I'm still experimenting with a few tweaks to get the most cooling. I'll keep a log of updates in the OP.

I'm curious what machine that is...a Dell __________.

 

Check out my Flickr album for cooling mods: https://www.flickr.com/photos/30149337@N04/albums/72157680615207736



Been using it for the past 2 weeks now. Liquid metal paste is awesome! This is coollaboratoy Liquid pro. I actually like pro better than ultra. I've used 2 tubes of CLU and just part of 1 tube of CLP. I'd say you'd get 3-4 applications of CLP with a full 1.5G tube on both the GPU and CPU of the XPS 15...
My temps are awesome. In Prime95 any test, I max out at 62C with the fans at 2500rpm. With the CPU and GPU going things get much hotter around 70C but still that's really good for 0 throttling in this laptop...

 

Do you have any more screen shots of hwinfo64 (showing all four cores, power limits, etc...) that you can post? Similar to how I have mine set up in the OP.

Which config do you have? 7700HQ?

I have everything padded below with some crazy mods...testing still in progress and I'll update the OP later today.

 

But why would that be so, considering that the thermal conductivity of LM is so much higher? Does the CPU heat up the GPU (higher temperature on the same heatpipe)? Or is it because the dosing of LM is pretty sensitive an one is likely to apply less than needed to avoid a spill? Air bubble? Some have posted photos showing a rather thick paste layer remaining on one side of a chip due to misalignment.

 

I'll input my experiences with LM...
1. There's a misconception out there that LM doesn't work well as a gap filler. It's actually the absolute best gap filler you can get. Liquid metal pastes are based on 2 main elements, gallium and indium. When you physically combine gallium and indium you create an alloy which has a very low melting point...Gallium alone melts around 29C...however with indium its much lower. So that's why you see liquid metal staying a liquid. I guess Coollaboratory has some other metals mixed in but my guess is it's because the gallium and indium aren't totally pure metals. I believe the chemical name of the alloy is Galinstan (Gallium + Indium + Tin). Since the "paste" is actually metal...its thermal conductivity is much much higher than a normal paste. Wiki has galinstan listed as a 16.5w/mK thermal conductivity which is about half what coollaboratory claims...but I digress. I'm no chemical engineer but it sure sounds fun...

The challenge with the liquid metal paste in a laptop is you have to get the application perfect... too much and you run the risk of it dripping out from normal laptop handling and too little and if your heatsink flexes and has a gap..it might not fill it...in my rig, there's a slight gap caused by the thermal pads on the VRAM. Now I corrected this with adding thermal pads between the heatpipes and frame of the laptop to continually push down on the heatsink... then I also removed the pads and added paste instead since the gap tolerance was less than what a 0.5mm pad could do...so paste was the next best option.

Liquid metal if applied correction can show incredible temperature improvements. I'd highly recommend it for people that are incredibly experienced with electronics and understand that this paste could completely destroy your laptop if it drips on a motherboard lead or you get some on the alu heatsink area...

Some people are concerned that the liquid metal paste will infiltrate the transistors in the actual die of the CPU or GPU...to date that hasn't happened for everyone...I was concerned with this also since I want a liquid metal paste to last me 5+ years in a laptop...the die is sealed with some chemicals and basically it shouldn't have an issue with it...if the die wasn't sealed a metal heatsink would short all the transistors...

I do get thermal migration from the GPU or CPU to either one... so if I'm benching just my GPU and very little CPU usage...i'll eventually see my CPU rise, not to the same temp as the GPU but it'll definitely be higher...and same with the GPU....since the heatpipes are together and the CPU/GPU use the same heatsink its bound to happen. This is also kinda a fatal flaw for gaming... since games use a fair amount of power from both the GPU and CPU....running both will heat up the laptop more than any single benchmark would. For instance, I've seen 72C on the GPU and 70C on the CPU in BF1 after an hour of gaming... conversely, running just prime95 I max out around 62C after an hour... way lower but that's because there's no power needing dissipated from the GPU...

 

With traditional paste, it has a more even contact with the heat sink and unlike the CPU where it has four cores (areas which are sensitive with a lot less room for error) the gpu (die) has a much greater tolerance; where having all the contact you can get plays a bigger role than the thermal conductivity rating itself.

Here are the thermal conductivity numbers of traditional paste vs liquid metal:

Liquid Metals:
Thermal Grizzly Conductonaut - 73 W/mk
Coollaboratory Liquid Ultra - 38.4 W/mk
Coollaboratory Liquid Pro - 32.6 W/mk

Traditional Paste:
Gelid GC Extreme: 8.5 W/mk
Grizzly Kyronaut: 12.5 W/mk
ICD: 4.5 W/mk (Not recommended. Rubbish in my books.)

 

Lol that's because it was running thunderbolt with a 1080...

 

Yea that's what annoys me in that these eGPU's via an AMP doesn't have their own category.

I hope they change that soon.

 

Nice topic, I would be interested to see some 3DMark 11 scores with i5-7300HQ and i7-7700HQ. Both the final score and the Phisics score.
Also if someone would do the same tests with DDR4 SO-DIMM DDR4 3000 would be much appreciated.

 

You buy the ram, I'll test it for you...lol jk

my 3d mark score is posted in the main OP...I have the i5.

 

unlock,

Here's my HWINFO gaming temps:


Only gamed with the GPU at 1820MHz during my session but nevertheless...still overclocks great!

 

Great temps, I just delidded and used Conductonaut on my i7 7700k and my temps went down 30C, I'm a big fan now...

 

Under heavy load?

Sent from my SM-G935F using Tapatalk

 

Under several hours of Prime95 28.10

 

If you see something that is too good, simply move on No need to investigate further

 

That's why we call it...Liquid Magic. ~30C drop in temps is not abnormal ....crazy right?

Not sure if I'm understanding what you mean here. What I was saying is that those who use amps should have their own category and not mingled in with laptops.

 

Why bother? It becomes more and more laptop models with AMPS. Futuremark software engineers may be forced to work overtime And can as well screw up the 3DM database, like they screwed up 3DMark Fire Strike

 

Ah yes... well they could just add a sub category. Ex. When you do an advance search and type in GTX 1050, it'll also pop up with GTX 1050 (Notebook). They could at one that says, GTX 1050 (External AMP) or something along those lines.

 

Futuremark is always falling behind. They need all kinds of hardware microcodes, and it is not always they are updated. Think we should leave it as it is Hence... If you see something that is too good, simply move on

 

Yea those results are manually purged daily to update the results. Crazy right? Also, my emphasis on adding the respected categories is to keep the boards true and accurate as Amps don't belong in there.

Ah well haha...live goes on...

 

Agree with you totally. Most Amps are going to end up being CPU bottlenecked sooner or later anyway, unless they really speed up the rollout of TB4/some other x16 port.

Sent from my SM-G935F using Tapatalk

 

But I saw 3DMark firestrike score from you, does not say anything to me. I'd need 3DMark11 scores...

 

aint

nobody

using 3dmark2011

 

The betters stayed with 11. Only one score, not like 3-4-5 benches and scores in the new 2013 version. Nobody can follow compare those scores. According to Notebookcheck Dell XPS 9560 with i5 7300 and Geforce 1050 does 8201 points. Was so hard to give Atom Ant an useful answer?

 

Can you elaborate on that please?
I do not see the correlation between CPU bottlenecking and eGPU connection speed.

 

Basically, even a 6700HQ/7700HQ bottlenecks (to a certain extent) an internal 1070. With an eGPU, the lower bandwidth of TB3 vs. PCIe x16 becomes a problem as cards in the eGPU enclosure get faster. A 1080Ti eGPU would be bottlenecked by a 7700HQ, for example. @Papusan can explain further.

Sent from my SM-G935F using Tapatalk

 

I do agree on fast GPUs getting bottlenecked by a mobile CPU and I also agree that faster GPUs can be hindered by eGPU connection speeds (tough at this time there
does not seem to be an indication for that ). What I do not see what the CPU speed has to do with eGPU connection bandwidth.

Maybe I am reading too much into it and you just had a brain fart with the sentence:

But it really does interest me if you found any correlation between CPU speed and eGPU connection speed.

 

With Gtx1070. And new games will come. 6700K run stock. And the 6700hq is fully LOCKED!! As we know We all know the situation with the RazerBook.

 

Firstly, thanks for the picture I really appreciate the information.

Though I wonder if I express myself wrong here.
The core of my question was how a faster eGPU connection speed would ease the problem of the CPU bottlenecking in a game.

 

If anything I would think that a faster connection speed would increase CPU bottlenecking (by eliminating other bottlenecks).

 

Oops... Mostly a brain fade, but also the faster the eGPU is and the faster the port is, the closer it will come to being CPU bottlenecked, as shown (Admittedly with internal GPUs) by @Papusan

Sent from my SM-G935F using Tapatalk

 

Brother @Papusan - don't even start on the RBP - we know it is badly thought out and designed - that will set me off on a tirade quite quickly.

Sent from my SM-G935F using Tapatalk