XPS 15 9550 temperature observations (undervolt + repaste)

Making good progress on RealBench & Heaven. Thanks for the very detailed posts. A few questions and thoughts:

1- Is it necessary to cripple the CPU to pass benchmarks? (disabled Turbo, disabled Hyperthreading, 80% max CPU power options).

===> As next step, I would work on getting these enabled to enjoy full benefit of your i7

2- Does disabling Turbo in BIOS make XTU Turbo (watt & time) adjustments null?

3- In the vrm area I still think long-term temps would be improved if you just had a few thermal pads to the case bottom:

--- only critical components are heatsinked to feeble bottom case for max impact

--- allows a the little stagnant air inside some chance of circulating to keep other vrm components at reasonable temps (e.g. the power IC (vrm brain) and all the tiny caps that the thermal pads miss)

4- I have not seen anyone reporting temp issues with RAM on the 9550 or 9560 (you can ignore posts where the VRM temp sensor was mislabled DIMM temp sensor). With those thermal pads you do run the risk of pulling heat from the bottom case to the motherboard. Or heating the RAM more than unpadded. I would consider removing those pads, leaving the feeble bottom case for critical heat sinking

5- Your nvme SSD probably should have thermal pad as it runs hot. Can't tell if you still have one now. My SATA m.2 drive runs cool and is probably one reason I have no throttling, as it is all at the margin

6- With your current temps and benchmarks I would assume your paste job is good. I would not bother changing thermal paste at this time. I would not touch vram pads

 

1 + 2 ) So I was able to uninstall XTU, and raise CPU to 100% with no throttling. This is likely due to my intel graphics power options change that I will talk about below. I apologize, I actually had hyperthreading enabled but I have removed it from my previous post. I tried enabling turbo boost, which was a massive fail and throttled within the first minute (Edit: CPU was in the 90s within seconds). Hmm that is a good question regarding disabled turbo boost/time/wattage, mine has been disabled since the start of this endeavor haha so it likely had no effect the entire time.

3+4) I will likely keep the pads where I have them due to my current results, however it would be beneficial to have data on padding less of the VRM with new/clean pads (I know rockstar had success with it). I did previously try padding just the chokes and didn't see much of a difference, but the pads were relatively dirty. Next time I do open my case I may remove the ram pads. I also did not have any crazy temperature readings for the RAM, it was just when I touched them they seemed as hot as the heatsink and so I figured I would give it a try - which may have not provided any benefit.

5) Yeah sorry haha it was stuck to the case.

6) Yeah I will keep it as is for now.


So I did a realbench test with My Intel graphics on Max performance, balanced, and max battery life with some insane temperature results (CPU Graphic cores) which will be apparent in this picture:

http://imageshack.com/a/img924/5576/FJcHQu.png

Intel graphics power options set on:
1) Max performance - as you can see massive temperature spike into the 90s until throttle
2) Balanced - same as above, massive temp spike into the 90s until throttle
3) Max battery life - no temperature spike and no throttle, temp stays in 70s - very nice

Again this is with CPU at 100% in power options, and no XTU.

Ultimately I believe my VRM pads are still necessary during gaming loads. Ram pads may not be necessary, but it seems like changing the intel graphics power options is probably 80% of the success.

Edit: Heaven scores are now pretty consistent around 2974

 

Thank you for your tests.

Changing the integrated graphic settings to max battery life reduced temps during the stress test by 2-3°, this made it throttle at 12-13 mins instead of 7-8 mins.

I'm still not sure if this setting has any impact on usage peformances, if it doesn't then it's a great tip.

Honestly i think disabling turbo boost and limiting the cpu is not a good route, we need to find a way to make it work full power; if it can't be done (and i'm starting to think that it can't) than the value of the cpu drops since it could have been better to buy an i5.

Also, don't hate me if i say this but you stated that running a stress test with all the default power options enabled (turbo boost, cpu power and intel graphics maximum performance) with an external cooler made your cpu temps jump to 90+ degrees within seconds; this means that something is wrong because with everything enabled, without an external cooler, mine stays at an average of 84-85° for the full duration (a bit less during the first 2-3 mins, a bit more later) and starts to throttle within 7-8 mins.

A question, how do i raise the power limit back to 45w after it throttled down? Is restarting the only option?

 

Any clue on the approximative page ?

 

Better buy an I5 then. It will still do it though.

No but it isn't rocket science, find a repasting guide and keep practicing until it looks right and do it once more.

 

From what i understand pressing's i5 doesn't throttle.

The thing is that after all the tests i think i have a decent setup (by the way i'm quite positive that a couple of 2.5mm slices of thermal pads on the gpu side of the heatsink lowered gpu temp by 2-3°) but i'd still need to lower cpu temps by at least 7-8 degrees to avoid throttling; setting the integrated gpu to maximum battery life lowers it by 2-3°, i'm not sure where to get the remaining 5°, maybe moving to alaska.

Of course most of these tests are just for the sake of it since i think most people won't constantly use 100% of the cpu power and 100% of the gpu power at the same time in the real world, but since we share the same passion you know better than me that it's more of a "being able to do it" thing than a real need.

 

Why does changing the iGPU have any effect on the temps when it shouldn't be in use anyway? Very strange.

 

I think the integrated gpu still runs the system and by setting it to maximum battery life it underclocks thus the lower temperatures.

The gain is not much but it's something, i didn't extensively test it but the cpu ran 2-3° hotter on the second test i did setting it to maximum performance (after letting it cool down).

It's easy to test by the way.

 

Really interesting tweak. I will definitely try that.

 

Well, in principle I agree. But then I guess you should place that Conductonaut order, use VRM padding with a laptop cooler, and consider a copper band over the VRMs towards the vents (the ASUS ROG GL502 seems to use an additional heatpipe for this, see http://images.anandtech.com/doci/10885/Interior.jpg). Question is, is the CPU at full power worth the effort.

 

Man it looks like based on some of these graphs that liquid metal is going to be a great addition... moving heat away as fast and efficient as possible from the CPU and GPU area is a must...especially if we're using the case to dissipate the heat.

Stay tuned! I should have some pics up of my progress on Thursday/Friday...

 

I went another road to get my system running the way I like it. My Setup is putting 5mm pads on all mosfets (I think they were about 7W/K), undervolted by 120mV, put TDP clamp to 35W (thanks Eason!) and di a repaste with Kryonaut. As I mentioned I did put them on almost every piece just to get to know that it didn´t work perfectly .

In Heaven benchmark I get a consistent 2850 value without throttling in any way GPU stays at 1670-1680Mhz. Watch Dogs 2 runs fine without throttling too. GPU sits at max. 70°C, CPU at about 75°C, ambient in the mid 70s. Realbenchs stresstest is not completing without throttling. The temps are ok but there seems to be a power limit somewhere.

I just put the Intel tweak and have to see if something changes.

Heres a pic of the pads:

Spoiler: Pic of pads

Putting pads on the heatpipes doesn´t work for this laptop as the temps are getting higher (at least for me).

Ginglymus : Your heaven score is quite impressive! Did you watch your GPU clock speed? It seems that this Benchmark is less depending on CPU as yours is quite crippled

Does anybody know how the 1050ti clocks in turbo in real life? I find it quite impressive that our 1050 stay at 1680Mhz for quite a while and don´t want to have thTi in the 9560 regardind the temps/powerdemand of the cards.

Hopefully I got everything...if not feel free to ask

 

Firestrike score?

Have you tried taking the pads off the mosfets and instead putting pads between the CPU/GPU spreader and the case and compared benchmark scores, with all else equal?

 

FYi... from what I've seen with other laptop mods... a Liquid Metal thermal paste should net you a 5C drop in temps across the board... So.... 10C drop from undervolt... 10C+ drop with liquid metal paste... I'll have actual numbers later this week when I get mine tomorrow night...

 

Yes, I did and it didn´t work. The heat got to the GPU and ambient got quite hot because of the case not getting the heat away. I had done this on my 9550 and was one of the people who didn´t have problems with this mod. The 7700 seems to run hotter.

Firestrike will follow as I don´t have it installed yet.

 

I´ve messed around with a lot of stuff on desktops and laptops but never used liquid metal Looking forward for your results

 

Did Firestrike 1.1 and got 5401. Don´t know if it´s good...does somebody have some 9560 to compare here? But again. No powerlimit/throttling. GPU max 72°C, CPU about 74°c with Spikes to 80°C, ambient below 74°C.


Edit: Just googled a bit and it seems that someone had Firestrike score up to 5800 on a 1050 Laptop. So maybe something is missing here. Maybe the TDP clamp? Maybe it´s simply a different machine and the XPS is at this value...

 

Yeah it's fine, mine did roughly 5564. Not sure on the temps but as long as it doesn't throttle it's fine.

Not sure on the version, downloaded it through steam.

 

I believe the 1050ti desktop is around the 1700's. Full clock on the 1050 should be in the 1700's as well..

 

Can't get my i5 6300HQ to throttle in games or benchmarks run consecutively a few times (Heaven, Prime95, ROG StressTest, etc.). As noted, I haven't run FurMark or simulataneous tests which can permenently damage laptops.

I think the i7 is a good performance upgrade if you can resolve the throttling, disabling turbo, clamping wattage, running at 80% power challenges. I think those were all sorted in the 9550.

If I had the 7700HQ in a 9560, I would be looking at CLU. Just make sure you carefully research the dangers of using liquid metal inside a computer (make sure mating surfaces are hyper-clean, do not allow contact with alu surfaces or any electronic components, find a good high-temp electrical tape to potentially mitigate risk of running liquid, find a non-runny liquid metal...)

Good luck and please post details...

 

sem - That is a cool solution one could make quite easily! Would you run new heat pipe to CPU or GPU side or both?

 

I can see a member going in the business of professionally producing modded heatsinks to actively cool the VRM..lol I might have the perfect opportunity since I might be in the market in the near future...

 

http://uk.farnell.com/wakefield-solutions/121716-rev1/heat-pipe-flat-2-5mm-dia-copper/dp/2676230

This has got me thinking. a very thin I guess 1-2mm copper plate with thermal epoxy to the heatpipe with thermal epoxy to the area with the fins. Just being next to the fins would do a better job than any other attempt!

 

Some ideas...

1. One side of the new thermal pipe
-- epoxy to the area near exhaust fins
-- CPU or GPU side?
-- need to bond clean copper to clean copper

2. Other side of the new thermal pipe
-- epoxy rectangular copper plate (1mm?) over vrm area
-- below copper plate, just mate mosfets with high performance thermal pads (both CPU & GPU mosfets?)
-- optional, could epoxy some copper slugs between rectangular copper plate and thermal pads (this might increase risk of short circuit if things move around however)

** Need to make sure no chance of shorting out components with new copper gear

** Need to make sure new thermal pipe & plate are solidly mounted so will not move around. I think that would require a second epoxy point for the new thermal pipe

** Solder could be used instead of thermal epoxy but that probably requires a blow torch and the heat might damage the magic inside the cooling pipes.

 

When my gpu die wasn't being cooled down by the heatsink (because it didn't make contact) my cpu ran much cooler.

When i fixed the issue almost the same exact amount of degrees got shared between the two chips (90° gpu and 70° cpu became 75° gpu and 85° cpu), to me this means that bringing more heat (vrms' heat in this case) to the heatsink without increasing its cooling power can't be a good thing.

I think the most beneficial upgrade would be to the heatsink cooling power but i don't think that's possible.

Imho the next, more realistic, test would be passing more heat to the heatsink (by using liquid metal), and see if the heatsink can dissipate it; if it can't there's not much else that can be done.

I have a tube of coolaboratory's liquid ultra but i'll wait on some squishy thermal pads (for vram) to be delivered before thinking about jumping in again.

 

Yea! I'm going to use a normal paste at first till my new pads arrive... once I get my new pads then I'll properly be able to pad up the VRAM and then I can do the CLP right the first time without worrying about the VRAM later and having to repaste both.... that'd be a huge pain... I'm hoping this can be a one and done solution... I'm sure it'll take quite a few attempts to get the VRM just right but I'm hoping to take the cooler off just a few times...

 

As a general matter, hotter copper fins should transfer heat to the air much more efficiently. That could be a secondary benefit from using liquid metal TIM

You see this same dynamic with forced hot water heating systems in old homes. Room radiators are much more efficient with very hot water. But, there are some balancing acts. Max water temp is a system might be 80*C to steer clear of dangerous pressures of boiling water.
_____

There are a few liquid metal shootouts in the NBR forums showing promising results. Never saw any liquid metal data for the 9550.

"iunlock" owned a 9550 and worte a liquid metal shootout guide here on NBR; not sure if he used liquid metal on the 9550. "xlawx" posted briefly that he used liquid metal but provided no details other than saying it worked great. Both users appear to have sold their xps'.

Without going through this thread manually, below may be the best summary of 9550 thermal paste used & results (google spreadsheet):

http://forum.notebookreview.com/thr...ndervolt-repaste.785963/page-54#post-10345187

 

I like the idea of braising the pipe to a plate, they do this anyway so as long as you don't keep the heat on for so long it pops the pipe (highly unlikely) or melts the solder off the end that is crimped (best not to so anything with heat on that end and keep it wrapped in something wet) then I can't see it being a problem.

Baldrick, I have a cunning plan.

 

Not enough space for a heatpipe ;( But an 1mm copper band might work, with the original copper stuck to the case bottom removed, especially if the case bottom can be raised a little bit with spacers on the fans and washers below the two central screws. I'd paste it to the case bottom. On both sides. I'd try to keep a thin gap from the heatpipes for air, with thinner spacers. The case bottom in that area seems to be cooled by the fan exhaust in thermal images. Sealing of the fan exhaust might need to be adjusted.

 

I got a 5601, so definitely similar to both of your results. This was with Turbo Boost enabled, no XTU, and CPU at 100% (iGPU was Max battery, will have to try with it on max performance).

 

Aww I'm dying to get my XPS 15. I look forward to taking baseline results, then comparing UV, repaste, and pads.

 

Laptop is 3 miles away! DSLR is charged! Thermal paste and pads are ready! Torque screwdriver is set! Flickr account is back up! Ready!!!!

 

Lol, glad you are excited, let us know the results. Did you get your liquid metal paste?

 

There's something that I don't get about the benchmark.

Some of you undervolt their GPU, and your performance increased vs standard settings.

Undervolting the GPU doesn't make it underperforming ?
Or is it because with the undervolt, the temperature are lower and the benchs can go further ?

 

Undervolting will lower the temps and let it boost more, but I don't know if they can lower the voltage on the 1050 - I thought that was controlled via bios

 

Got my Liquid Pro and Liquid Ultra... Although I don't think i have enough ultra to use... looks like I'm a druggie with the syringes in a bag..lol

I'm 99% sure you don't have any GPU voltage control. Unless the bios allows access to it, which I'm almost positive you can't unless you do a total system bios mod...and know the exact GPU voltage controller on the motherboard. Voltage should be locked with the GPU. I think its around 1.05v, correct me if I'm wrong.. I'll know tonight... But according to Pascal architecture...that should be plenty volts for even decent overclocks while keeping the power consumption down...

 

my bad my bad, I missclick, I was talking about the CPU voltage control

 

So your question ultimately focuses around why Intel left such a voltage gap that we can undervolt our CPU's and still crank out the same performance??? I think that's what you're getting at. All chip manufacturer's over estimate the voltage needed to reach a certain MHz or GHz because not every chip is made the same. It's very apparent with desktop CPU overclocking on the different voltages required to reach a certain overclock... For Instance... my old GTX970... Stock volts was 1.186v on load. This boosted to a stock clock of like 1332MHz on the GPU. I could actually get 1475MHz without any voltage change... And here's the best part... I could undervolt the GPU down to 1.088v a whopping -100mv and I could still overclock it to 1400Mhz... pretty crazy right? That's great for my chip but Joe Schmo down the street with the same GTX970 might need 1.2v just to get to 1400Mhz... so it's just the luck of the draw... or as most people call it, "silicon lottery"... which is also a company btw...

My XPS that arrives today might be a good CPU undervolting chip or it might not..I won't know till I try. And same goes for the GPU... I may have a huge overclocking headroom...hopefully... or it's going to breakdown on a +50Mhz overclock... just have to wait and see what I get... Although averages do help in revealing what I should be able to do regardless of having the actual unit in hand... Like most 7GHz GPU VRAM can OC to 8GHz quite easily..... I'm hoping for a 10% clock lift...we shall see on this one!

 

I'm curious if prima could do a modded bios for our laptops with the GPU undervolted...

 

Well yes that was my question indeed.

So by all your experiment, we could say that thoses chips run at the same performance with -100mV, with less heat so better heatflow so better perfs ?

 

Yeah the voltage has no bearing on the performance of the chip, only stability. Intel has a margin of safety built in so they don't have to worry if the cpu will be stable or not. The goal for us is to run at the lowest possible voltage and still be stable (hence the need to run a stability test like Prime95).

 

@Prema how much money would it take to make it worth while?

 

Never seen a hacked BIOS due to it being encrypted? (iirc)

 

Funny one dell! Very funny!
At least the paste wasn't fully dried yet...


At least the CPU cleaned off nicely... Didn't have time to do the GPU real well...And probably the worst thermal pad job you'll ever see....I was in a huge time crunch...so no time to pretty anything up yet...


I've run into numerous issues with my OS right now so I'm trying to sort that out before I can start to get some numbers on this thing...

I know the GPU clocks to around 1730MHz and MSI afterburner does some really weird stuff with Nvidia optimous... lol I went to OC the memory (+100) and it went to +49575813 instead...lol wut...Good thing I didn't hit apply...

So a few things:
The GDDR5 is Micron D9SXD grade VRAM. It's actually quite dated memory according to the model number.

 

I don't trust any OC software except nvidia inspector. I've had two laptop GPUs die suspiciously while using afterburner. It sets the clocks to 65535/32767 randomly.

 

I'll give that a try. Linky to the most current version plzzz++++!!!!
Edit: NVMD I got it...

Ok... So I didn't do a few things that I need to do...

DDU my PC then reinstall Nvidia drivers...
Reset windows
Use Nvidia Inspector instead

Think I'm going to leave the laptop cooling as is right now with the VRM. New thermal pad should be here tomorrow or Saturday and I'll properly do the VRM cooling... Hopefully rest of the VRAM pads will be in or I'll use thermal paste instead in prep for CLP.

 

If you're using inspector btw, set up task scheduler so that your clock shortcuts run every time the laptop is plugged in + turned on from a low power state. That way when you're plugged in it will run the OC and it won't if you're on batt.

 

So you guys got me into repasting with CLU...I couldn´t resist and did it. Temps are now quite a bit lower and GPU doesn´t throttle down in Firestrike(CPU max 65° GPU max 65° Ambient max 72°). I had the TDP set to 35W. I put electrical tape around the cores and the use of the "paste" wasn´t that difficult. I tried it before on some dead laptop mobo CPU to get used to it.

 

We require pictures or else it didn't happen...

 

Cool, why did you limit tdp to 35w? Try running stock.

Try running realbench's stress test and monitor cpu temp and power limit with hwinfo, if it passes without throttling and cpu temp is fine than i'd say it's a success (with the graphs, to check if there's any strange behaviour).

For the gpu most of us tested with unigine heaven set to basic, throttling wasn't an issue for anyone but you may still have better temps try running it and monitor gpu temp and clocks (with the graphs, to check if there's any strange behaviour).