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

That's about what I get. 1798mhz at 0.975v. I think I can go lower but I'm erroring on the side of stability. OC your memory too. I get +250 and it really helps in benchmarks....

 

I'm upto 1839mhz which is around 0.987v...freaking awesome for the clock and voltage. Stock voltage is soo high it's not even funny. I wish I could determine though how much less power it uses...if I had a Killawatt meter I could roughly use that but anyway..no biggie. Temps are way lower. This laptop still amazes me. Coming from a Dell xps 14 with a gtx 420 it's a heck of an upgrade and about 1/2 to 2/3 of my gtx 970 desktop processor. Bf1 runs great with it.

 

Because more data is always welcome :

Just did @iunlock 's mod*
*with 1.5mm thermal pads as joint not 2mm, keeping the small foam pads around the fans, using "up to 105C" certified Scotch electrical tape to seal things up.

Firestrike results with CPU -120 / GPU -100 undervolt


Just awesome considering I had 95C° ambient before the mod (92 with heatspreaders on), even with undervolting so thank you @iunlock

15min Firestrike combined loop :

CPU: 70C
GPU: 73C
Ambient: 73C

So this completely unrealistic workload doesn't even bring the computer to its knees anymore, it now has plenty of headroom.

I feel like this mod makes the 9560 what it truly should have been.

 

I fix it for yoo If not, you would have been fried alive

 

He probably means the first "Ambient" sensor, which is the sensor for the VRM's.

I just did some gaming with my new settings, and I think I am going to stick with this undervolt/overclock combo:

CPU (7700HQ) undervolt: -135mV
GPU undervolt: 0.95v (~-125mV)
GPU overclock: +238mhz (1.784mhz)
GPU Memory overclock: +220mhz

So far it seems very stable in Valley benchmark (slightly higher scores all around than last test) and in actual gaming.

I also tested using Guild Wars 2, all settings on highest, no AA, 1920x1080, locked to 60FPS. In most area's I get a stable 60FPS with about 80% GPU usage, 30% CPU usage. With both hovering around 70 degrees C. I tested Lion's Arch as well, hovering around 35 FPS (large player hub, pretty stressful on the system). GPU usage went to ~90% while still hovering around 70 degrees C. CPU usage went to 50-60% while hovering around 75-80 degrees C. Both are not throttling, even though my VRM temps are around 87-92 degrees.

 

Great stuff and great job!

 

Ahah like MrBuzzkill said I'm talking about the "Ambient" sensor which is the vrm sensor. It gets pretty toasty because of the lack of airflow.

You deserve all the credit!

 

got that Heaven score after repaste, repadding of chokes, undervolt of CPU/iGPU/GPU

Edit... can't add image

 

Quick question...

I've been using afterburner as I've seen done here. I had it starting with windows but noticed my battery life wasn't improving, in fact it was worse than before I tried the mod. I was benchmarking and had throttlestopo open and I realise that without afterburner open, my laptop idles between 0.6w and 1.3w. Opening afterburner with the same settings as before, I am idling between 8.5w and 14.2w. I had no idea it was using this much resources and kinda defeats its purpose. Is there any way to have the settings applied at boot without having the afterburner app open constantly?

 

Open it, do your tweaks...close it...done...

 

It keeps the voltage then?
I use it on my games PC to overclock the hell out the video card but it's primarily open to control the GPU water cooler fan speeds via PWM as the temps sit so low the vBIOS want's to shut it down. Turning it off obviously stops the custom fan profile.

It would be nice if we could mod the vBIOS like standalone cards!

 

Has anyone done any power efficiency testing of the various CPU and GPU throttling mechanisms on the XPS 9560 in regards to gaming?

I'm looking to maximize the FPS per Watt while gaming and am curious how these settings affect this:

nVidia cPanel & Inspector throttling settings
Windows CPU and PCI throttling settings
Dell Power Manager throttling settings
XTU??
Afterburner??

My assumption is that some power throttling will help as it reduces power consumption during less computational intensive periods, but that maximized throttling might do worse because too much time and energy is wasted switching between power states.

Also I always use v-sync, so I assume I'd also want to reduce the clock and/or increase throttling until the GPU is stably generating frames just slightly above the refresh rate (60-70fps). Otherwise the GPU will waste energy waiting to generate new frames while at a higher clock speed.

 

So I have the same issue with Pwr showing up, but my ambient readings are way way lower. You fixed this entirely by getting more cooling on VRMs?

 

To answer myself, yes, it's 100% a temperature issue. I took a blow dryer on cool and held it against the entire VRM section and the GPU happily runs at 1.7GHZ the entire time during Unigine Valley; 5 seconds after I stop cooling the VRM area, Power limit throttling kicks in.

Now just need to figure out which ones are the culprit for GPU.

Edit: So I did some testing and i'm 99% sure it's the VRAM overheating that causes this.

I have an IR gun and when I put a blow dryer on cold air up against the VRAM for a minute or two, GPU will happily run at 1.7GHz and the Performance Limit in HWiNFO jumps to "reliability voltage" - which I understand to be the max voltage it gives the gpu to run at max clock.

Every single Ambient sensor onder DELL in HWiNFO drops below 45c but still gpu does power limit throttling, but the moment i put the fan on the VRAM (specifically the ones closest to the back of the chassis) ambient temps do not matter - it'll stop power throttling. Ambient temps can go up to 50+ and it still happily runs the GPU at max clock.

So question is what to do, if anything. the stock thermal pads make good contact because the heatsink above the hottest vram chip (below the QR sticker) gets up to 60C easily.

Any ideas?

 

Got a 7700HQ arriving tomorrow. Will enable SpeedShift, do a clean install of Windows and drivers, do a simple undervolt, careful repaste with Conductonaut, and some (very) quick vent modding with Kapton tape.

If the VRAM chips are making good contact with the heatsink, I think I will leave the factory pads on for now.

 

My 2nd phase also had similar setup

Thermal Manangement #2

 

Received the 7700HQ with a Samsung 961 SSD. Benchmarks are much better than those of my 6300HQ with Samsung 850 EVO. This 9560 really overwhelms the XPS' cooling. Did no physical modding to the 9560.

Did a clean install and several tweaks to minimize latency. However, I was dissappointed that overall latency is slightly better on my 6300HQ. I expected improvement moving to a faster CPU, nvme drive, and a gen2 board. Improving latency for virtual instruments was the main reason to upgrade.

Given weaker latency performance, high heat, and broken left HDMI-USB ports, I returned the 9560 and am happy with my 9550.

 

What did you use to attach the little Heastsinks? I ordered some here in Germany but they just use double sided tape which I'm assuming hinders heat transfer between MOSFET and heat sink.
On the other hand I fear that if I use a thermal pad to attach the heatsink it might fall off.

Any suggestions?

 

If using the little sinks "standalone", google up thermal adhesive, but mind it may be difficult to remove afterwards, and that the sinks alone don't help enough. Doing the full mod, you can use regular thermal paste, and make sure you put in thermal pads in so that they keep everything in place after the chassis is screwed back together.

 

It's double-sided thermal tape and for your application will be fine. Glue them and kiss goodbye to any warranty.

 

There is virtually no airflow in the VRM area so I don't see the alu heatsinks making any difference unless you can increase airflow in that area significantly.

1. Thermal pads generally have very little adhesion so are not really an option the attach underneath the alu heatsinks. Putting the thermal pads between the case bottom and alu heatsinks as a type of springy fastner is a bit dicey as the case bottom is flexible and the motherboard is a bit flexible. That will put a lot of mechanical stress on the very small and very weak soldier joints holding the mosfets and/or chokes. Also if the case bottom loosens more than you expect, you may have a bunch of alu heatsinks jingling around and shorting out the VRM.

2. The double-sided tape might adhere well but if you are moving the laptop, I would expect movement, heat cycles, and exposure to air would cause the tape to eventually fail. When the alu heatsink is floating around your laptop, expect a catastrophic failure. Also, the mosfets and chokes are only attached to the motherboard with very little soldier, so don't assume you will be able to remove the double-sided tape without catastrophic damage.

3. Thermal epoxy properly applied is probably the most robust adhesion. From a practical perspective, I don't think that would be removable if you had warranty issues.

 

yeah that was my thought as well. I upgraded to the 9560 because that one has warranty for another 2 years so I'm definitely not messing with that.

Maybe I am overthinking this anyway. I do get Power Limit Throttling when I run p95 small ffts test but it never goes below 3200 mhz so I am probably not losing any real life performance. I haven't tested any real games yet, though.

As far as I understand power limit throttling comes from overheating mosfets so they can't provide enough power. Do you by chance know what temperature in the ambient sensors in HWinfo initiates power limit throttling?

And one more question. How do I know if one of the VRAM modules doesn't make complete contact with the heatsink? Will it cause the entire GPU clock to throttle down or is there a specific heat sensor?

Thanks for answering my questions by the way. Have a great weekend.

 

The main cause of pl1 throttling is the chip hitting its rating of 45w, it is after all a 65w cpu set to 45w.

Sent from my SM-G920F using Tapatalk

 

This is all well documented here in the forums.

There are several factors that will trigger power limit throttling. For example hitting power limits lol. There is not much you can do there (although you could try Intel XTU to try to trick the system but that is dangerous and the Kaby Lake runs so hard I don't think it will help too much. It worked perfectly on my much less aggressive 6300HQ).

Also hitting around 76-77*C in the local ambient sensor will trigger the power limit flag.

On the 9550, one or two VRAM chips had an issue where the thermal pads were not properly contacting the heatsink. Not a reported problem for the 9560.

 

For those thinking about doing the mod by iUnlock: Do it! I did it about 4 weeks ago and it really does help. Before I was quite happy with my setup but was tempted to try the mod out one day. Then I bought a spare bottom cover and gave myself a go It really works as he described it and if you just follow his steps your fine. The bottom doesn´t get hot as before (I padded my chokes) and the temps are great.

Thank you, iUnlock for all the effort you put in!

 

So interestingly enough I redid my iunlock mod and I don't get any throttling in realbench. Not sure if it's because of a bios update or anything, I should have tried before I redid it. Really I just added a small heatsink to another IC and then put thermalpad on all of the chokes. Ambient temps look good, highest is 68C. I'm going to try to undervolt my GPU and run it again since clocks are dropping to 1GHz on the GPU while running at 78C...

Hah spoke too soon, CPU throttled after about 14 mins of realbench.

 

Great topic, very helpful!
I've collected most of the stuff for mod - just need to get some thermal tape for heat sinks.

Only difference is
1) Made custom heat sinks, because I could get 11x6.5x6.5mm sinks, had to saw them smaller and deeper. Some of them are probably still too high (about 4.5mm)


2) Bought some copper tape, in that way you don't need to cut the original copper or buy another bottom cover - also cheaper (possibly).


----

Todays quick run with the laptop, stock paste and setup @about 21C room temp

cpu -125mv
igpu -100mv

ngpu -35mv
1911mhz core (goes 1923mhz and then down to 1911 under heavy load, for some reason doesnt want to stick - although temps didnt hit over 68C)
1866mhx ram

I tried to put ~1950mhz on core, but it crashed heaven bench in few secs with same voltage (that's still under stock). Mem I didnt bother to test out, possibly a bit more in it.
Now I have to find out an optimal. It worked fine with -100mv and ~1820mhz.

Firestrike from 5500 > 6046

https://www.3dmark.com/3dm/22816891

 

Awesome! How are you undervolting the nVidia GPU? Afterburner and nVidia inspector won't let me touch anything beside the clocks.

 

Hi @iunlock,

I saw an archived thread on reddit from around 8 months ago advertising Team Lhz's repaste services for Alienware laptops (and in the thread you mentioned other laptops too) - is this still going on? @Mobius 1 suggested I contact you on this board as I would be sending this unit from Australia (I will handle shipping both ways).

I have a Dell XPS 9560 that is hitting 98 degrees celcius when gaming, and I saw your repaste guide on it (with some strange air-flow corridors made out of thermal pads); it just looks way out of my league when it comes to laptop repairs. Do you still provide this kind of service? If so, would it be possible to get a quote including materials cost?

Apologies for the public post - I have not yet made the prerequisite 5 quality posts to qualify for private messaging, and I thought this would be preferable to me making 5 nonsense posts.

Thanks mate, looking forward to your reply

 

Why not start with a simple repaste and see how you get on.

 

I'd rather pay someone experienced to get it right first time then mess around with it myself. If @iunlock can't do it, then I guess I'll have to try it myself - but really leaving that as a last resort.

 

Ctrl + F, this will show you the voltage and core clock chart.
If you put for example 1800mhz @0.95v and straight line after it (as on my image), it will not go over 0.95v.

 

Just check a youtube video on repasting, it is really easy.

 

Easy, but on second thought I wouldn't let my gf try doing it... not easy for all.

 

It's not just repasting though - it's also applying heatsinks to the FETs and creating a wind tunnel using thermal pads, and ensuring all this retains an even surface.

I work 12-14 hour days, I really don't want to spend 3-4 hours figuring this stuff out only to find I've ****ed it up.

Here's hoping one of the Team Lhz members (and most likely @iunlock) are willing to sell their expertise.

 

Not sure why you think you need to go to that extreme when 90% of the time a repaste is all that is needed. Mines fine with a repaste.

 

First, I would advise you to monitor the Ambient temperature.
If just your CPU is overheating and Ambient (the one that iunlock hit 105C) is not over 80-90, you might just get away with a repaste.

Also lower cpu/gpu temps and voltages will result lower ambient temp.

Secondly, lower the voltages on cpu, igpu and ngpu - if you have not done it already.

Thirdly, do the repaste and then again monitor the temps.
Even if you've never done it, its so darn easy these days - not like some fifteen-twenty years ago when you had to clip some akward clips over you athlon xp cooler, hoping it wont crack the fragile cpu (no ihs or anything these days).

* Look through a short tutorial video
* Too much, or too little paste is not good, but a bit more is better than too little
* Remember how tight the screws are. Too loose or too tight is also not good - overtighten and you might brake something.

That's it. Should take no more than 15-30 minutes (or a bit more, if you take it slowly).

Ofcourse if you live in the hot regions of AU, then you'll probably need all the cooling power possible

Good luck!

----

hwinfo64,
You see

: CPU
: Ambient (thats the hot one, you need to look for, also seen on iunlock screenshots)
: Ambient
: Temp3

 

Most people don't need to do that. I repasted, undervolted and I can run at 3.4GHz all day long, not throttling whatsoever.

 

OK, so I looked into this method, but I think you are misinterpretting how the Frequency/Voltage graph feature works. This doesn't alter the voltage directly, but instead adjusts the clock depending on the provided voltage. So you really aren't "undervolting" in the traditional sense by using this method. For example, in nVidia inspector, if I set an overclock of +240, I peak at [email protected]. But if I set an underclock of -100MHz I get a peak of [email protected]. So in a sense this is undervolting because when I was overclocked, at ~1544MHz I'm most likely ~0.850V, but when underclocked, at 1544MHz my voltage is much higher at 1.050V. So you are only controlling the clock, but as a result it ends up altering the voltage provided at each clock as well.

This actually helped me understand why my GPU wasn't running much hotter at different settings. The overclocked setting was running at a higher clock but also a lower voltage and vice versa, so the thermal impact of both cancels out.

 

Well said and yes I encourage everyone to try and repaste it themselves first. The only time it'd be wise to seek help is when using LM.

Traditional paste is pretty fail proof, but it still requires certain details in the process to avoid doing a ghetto job.

::iunlock::

 

Why not? There are usually reasons why people want to do what they want to do lol. Either way it's their choice. Just because something works for you doesn't mean everyone's usage habits are the same.

Post some screen shots of hwinfo64 with you running fire strike combined test (loop) + wprime v1.55 in the background. Let's see how you're temps really are...



::iunlock::

 

It will be damn

 

Why, simply because we don't all need to void a warranty because we read someone else did as it isn't necessarily the answer.
And I don't need to go installing benchmarks for your benefit when I know my laptop if fine for my purpose with a simple repaste.

 

I think you can't assume that everybody's happy with your purpose. I guess many folks buy the XPS 15 seeing the 7700hq and 1050 listed in the specs, assuming they'd be able to run normally at their nominal clocks. Because there isn't anything saying the contrary in the specs. Otherwise they could go for an 1kg ultrabook with an U CPU and just iGPU...

I'm not a gamer, my laptop is near idle 99% of the time and I carry it around a lot so I like the compact form factor and low weight. But 1% of the time I do benchmarks of computational algorithms that may take a while, and throttling is highly undersired.

 

You point that only at me when realistically you should be pointing that comment at iunlock as well, I am not assuming everyone is happy with my purpose, I am merely saying you don't have to dive into this laptop with an angle grinder 1st off when a simple repaste may suffice. YMMV, not every laptop is the same. The minority post here when in reality you would probably find 99% have no issue to fix.

At least a repaste is unlikely to void your warranty, if that doesn't work more invasive methods can be tried if the owner decides.

One loose heatsink that never had any place in the laptop and you killed it along with your warranty when the damage is spotted.

 

It's okay mate. Just ignore him...

You're right. Seeing the specs listed for the XPS and having it not perform to specs is very undesirable.

While a lot of people use it for regular work tasks, there are a lot of content creators and light gamers that opt for the XPS due the dGPU.

It's perfectly normal to have the 'want' in modding a system for it to work at its best ( as possible), however, some people just don't understand that.

 

To be fair to Gonzo, you are not reading his posts correctly. Gonzo never said repasting voided the warranty, he said that modding your entire laptop with added heatsinks, cutting up the copper, adding ducts with thermal pads everywhere, does void the warranty.

You DO NOT need to go for extreme modding from the get go. You try repasting first, if it solves all your problems: then you stop there. If it doesn't, and only then, you might want to take a look at extreme modding.

I think it is very unreasonable (and arrogant) to suggest that people will only get good results if you do the extreme mods, while like Gonzo said, in 95% of the cases, a repaste will suffice.

 

FFS, read my previous post.

 

@MrBuzzkill, that is not true. There is no physical modifications of cutting up the heat sink and adding heat spreaders is not a physical alteration of the original component, therefore, this has nothing to do with voiding the warranty.

The same goes for repasting the laptop, which can involve the user replacing the stock thermal pads etc...it is no different.

Dell themselves have tweeted on this and the enthusiast community has all heard it from the horses mouth on many occasions.

This is not a secret about the warranty etc... That is why I'm making this very clear so that people don't get confused over this warranty subject.

I agree that one does not need to go to extreme modding to achieve a positive result, however, in a thin chassis like the XPS that already has many handicaps with it having very small fans (cfm output), it is already at a disadvantage.

I would assume that if anyone is wanting to repaste their system in the first place, that they are wanting to do so to achieve positive results. There are a lot who also use the XPS for gaming, content creating, rendering etc... and in that case they will benefit more from addressing the core issues of what causes the throttling in this particular laptop. I've map out in detail and provided all the areas to address and you'll see that the culprit isn't just with the heat sink, rather the vrm's etc...

A properly done repaste alone is better than nothing, indeed. However, there is more to it than that for those that do anything CPU/GPU intensive for a long duration of time.

 

I am pointing at you because you are a bit in denial regarding the well-known thermal issues of XPS 15. I agree that only a minority has a real issue with this, but we do exist

I surely agree that loose heatsinks can cause very much trouble to folks that needn't bother with that anyway. But I also think that your attitude caused thread pollution here, which I hope some mod will tend to.

This is something I don't agree with. Some of us do have work that causes very similar workloads as stress-testing. In particular, computational algorithms which make use of both the CPU and the GPU at once. Also certain video processing programs which try to make good use of the GPU. Not to mention certain games that in addition to loading the GPU also have work for the CPU. Then the point of stress-test benchmarks is mainly running things that other users can repeat so that the results are comparable. Until you show understanding for these issues I don't see a point that you continue posting in this thread, though you surely do help folks a lot in various other matters.

I can't agree that the computer was "not designed to take that punisment" is thermal limitations are not listed among the specs on Dell page. And I think VRM throttling, assuming good paste, wouldn't be needed, if they redirected the resources spent on throttling development to heatsink improvement so that homemade plumbing wouldn't be needed. After the mod, the laptop likely runs cooler at nominal clocks than stock throttled.

Loose sinks, liquid metal etc are surely risks but I think adequate warnings have been posted in this and other relevant threads. It is certainly a good idea to firstly try just undervolting the CPU and the GPU, then if needed repaste and add simple pads for the VRM mosfets while at it, and only consider going further if the outcome is not satisfactory. Even repasting is not a job for just anybody.