XPS 15 9550 temperature observations (undervolt + repaste)

yeah I really stopped at -125mV on the GPU just because I got lazy...

 

lol you are talking about the intel gpu right?

 

Yeah, sadly without a modified gpu bios we are stuck with stock voltage there as far as I know

 

I have a GT72 with a 980m I use a modded vbios to overclock and overvolt it, but its too risky

 

Right, sadly Prema is likely the man for the job, but I'm sure he wants some sort of compensation to mod it...

My guess is the 960m has a good amount of headway like the skylake cpus

 

This is very interesting and useful thread, I had no idea it is possible to do this on this laptop.
It is great we can now make it quieter, cooler and extend battery life.
Would be interesting to share more experience about processor graphic (integrated intel) undervoltage.

Regarding the battery life, it is a bit disappointing on a 4k 9550 with 84Wh battery, it is way below rated battery life.
It is interesting that these days we talk a lot about car emissions and consumption scandals, car manufacturers are fined heavily if they lie about fuel consumption... If we would apply same logic to computers, somebody from Dell should go to prison for promising 15h battery life (I achieve 4-5h, without some heavy load).

 

Just wanted to add that with my i5 UV + GPU OC I'm getting CPU peaks of 70 and GPU peaks of 79 playing KF 2 at 1080p. It's running about 10-15C cooler than my Razer Blade 2015, though of course its GPU isn't as fast.

 

Just got my replacement 9550 i5 / 950pro the other day.

I did the re-paste (GC-Extreme) and undervolt myself. I'm at -125mV / 57/54/58/53 (prime testing >2 hours)... temperatures almost down 10c from stock.

I think i'll push to -150mV and stop. This can't be good for gaming even though prime has been stable. Thoughts on this?

 

Gaming will heat the GPU and CPU together. So the temps will be different, likely higher than P95 CPU alone in my own experience.

Yes, since the 9550 CPU and GPU share the same heatsink array, it works differently from a Desktop with its separate cooling setup (unless you are talking about custom water cooling).

 

For gaming it's better. It's a discrete gpu that isn't running with any less power, but will run cooler

 

Figured i'd stop at -150mV but i ended up pushing to -175mV. Temperatures have dropped to 53/51/52/49. Quite happy with results. This reminds me of OC'ing my i5 2500k to 4.5 Ghz (which im still running). Will let Prime95 run for a few hours and call it. There is noticeably less throttling once the temperatures came down. Fans are also working less hard.

 

You assume you used Intel ETU to do it? All you did was to set the "Core Voltage Offset" to -175mV, right?

 

I never found a way to make undervolts stick with XTU reliably.

 

Use Windows task scheduler.

https://www.reddit.com/r/Surface/comments/3xtq81/my_own_variation_of_xtu_undervolting_tasks_and/

 

Intel XTU, yea i set the core offset in steps, started with -100, -115, -125,-135 etc.. and i ran Prime95 for ~15 minutes per step up until -150, which i started to testing for several hours.

I stopped at -175mv, and tested prime for over 12 hours. Pretty happy with results.

I've rebooted my machine a few times and the settings have stayed the only time it didn't was due to a bluescreen, yea you know the one everyone gets once in a long while. (i think its related to WiFi card/drivers)

 

Did that-- despite editing the files correctly to -165 core/cache and -100 GPU, it would only change the values for the core/cache to -80mv. Couldn't figure out why, maybe the values for the components were wrong. Anyway, I just leave TS running to make sure it sticks.

 

can someone do a video on repasting? sorry me noob....I am upgrading from AW 13 to xps 15...

 

google & download the service manual and read it.

 

I gave up on the task scheduler, tried 3 different scripts posted, but they didnt seem to stick. The first one that was linked here did for a few reboots, but when i checked voltages through CPU-Z randomly it wasnt applying it correctly. I'll just manually run XTU until a better solution comes up. Really surprised Intel didn't bake such a feature into XTU

 

Chill out guys. We've been here for awhile now. Let's just try to help one another when possible

 

Play nice people. Remember the forum rules.

 

I'm not the one going around thinking I know everything, matter of fact I came in here asking for help...and people who doesnt do anything but being a keyboard warrior just come on and spitting out nonsense.

Anyway if anyone could make a video of how they do the paste would greatly appreciate it...if you dont want to do it or not gonna help, kindly keep your .02 cent to yourself.

 

I am far from a keyboard warrior, just a laptop engineer that thinks people who can't follow a service manual are probably best not breaking a £1600 laptop as you will void the warranty by damaging anything whilst inside it.
I was not trying to be sarcastic but either way it came across if anyone replies like you did will get it back. And as this is the internet I will chuck in my 2 pence when I see fit thank you.

Anyway, I doubt someones going to take the time to make you a repasting video as it is asking a lot of anyone. I am sure a google will find ones for other Dell laptops so you can use that as a baseline.

Download the service manual and read how to take the laptop apart, Dell doesn't say the best way to crack the case but do it from the vent side at the back as the clips are either side of the keyboard and at the front.
After you remove the heatsink any old repasting guide will do.

 

Speaking of the clips (both sides).. you might want to take extra care on those as it breaks very easily when taking the bottom cover off.

 

Yep, a very warm laptop so they flex as much as possible. Also running a benchmark to heat the paste up just before removal never hurts.

 

Hah things got a little heated in here I guess.

@xlawx - I do not have time to make a video, but the process really isn't too difficult. Remove the bottom cover, take out the 6 screws from the heatsink, clean heatsink and cpu/gpu with rubbing alcohol, apply a grain of rice sized dot of thermal paste in the center and then put everything back together.

GoNz0 is correct in that it's not tough and that the service manual would show you all of the screws/steps to taking apart the laptop.

If you have any questions you can always post here or PM me and I'd be happy to answer them.

 

Hi, I had the same problems and messed around with some script from reddit. Mine works fine now (checking with HW64) and keeps undervolted for many weeks now. I posted it in another thread and already sent it to some members but don´t know if it works for them too. PM me and I sent you my script for taskscheduler.

 

What type of DIMM temp you observe under heavy load ?
I see 70-80C... after 2-3 min...look way too much... unreal... maybe the sensor is not onboard at the dimm but under and near the heatpipe...

 

Same here, I did wonder as I fitted thermal pads on the last laptop and nothing changed. the base plate was just warm where the dimms sat.

 

I got 50° after running Prime95 for about 10min. CPU are about 70°.

Edit: But I got 87° at the first ambient sensor. Seems too much...

 

I wouldnt mind trying, but i am afraid of applying too much or too little...I heard some horror story of that happening.

 

1. run a test and check temps
2. open laptop
3. clean old paste off with rubbing alcohol
4. put new paste on
5. run bench again. if the temps are higher than before, YOU DONE ****ED UP IT UP
Just don't use too little. It's hard to **** up that badly.

 

4.5 take it off again and check it has covered the CPU & GPU without loads spilt over the sides then clean up the parts and repeat .
I always do that if unsure.

Sent from my SM-G920F using Tapatalk

 

I wouldn't do that because I heard it can trap air. Not sure if there's any truth to it though.

 

Hence me saying clean up the parts and do it again once you know you're doing it right m8
.

 

Yeah GoNz0's right, you can do a trial fit.

Put a small bit of thermal paste down, screw down the heatsink, take it off and see how it looks. It should be enough to spread over the surface of the core, but not so much that a ton comes out the sides.

 

Just got my annual bonus. 1 step closer to getting the 9550. Got my clu ready for the cpu. I'll probably pick up some fujipoly thermal pads for the vram...

 

Try SpeedFan.
Remember to enable "DELL support" in the option tab.

 

Try SpeedFan.
Remember to enable "DELL support" in the option tab.

 

wonder what you guys use....I just bought some Thermal Grizzly Kryonaut. Anyway, in the video i see people doing the spread, but some doing the pea size...just curious what you guys do.

 

Go with the dot method.

 

What he said, always dot on the cpu's without a heat spreader.

 

not doubting you, but why is that? it doesnt seem like the paste cover the whole cpu.

 

That cpu has a heatspreader, the CPU & GPU in the Dell doesn't so a dot will cover with ease (as I have noticed every time I repaste a chip with no spreader on it.)

 

Mainly I was showing that the spread technique does a great job at getting air bubbles and nothing else really. The actual core under that heatspreader in the video is much smaller so it is covered, and like GoNzO mentioned there isn't one on the 9550 so the dot method works great. You don't need a ton, these cores are pretty small. Do a test mount and see how the spread is.

 

grain of rice size does the trick.

 

I do not agree at all.

For CPUs with IHS, dot method is really good since the TIM is going to completely cover the part that generates most of the heat (just above the die).
But for bare dies, dot method will not completely cover the entire die unless you apply too much TIM (and it will probably spill out the sides), so covering the entire die should be the way to go.

Anyways, some analysis showed 0ºC difference between using either the dot, line or covering methods; so you are fine whatever method is used.
However, I got a 4ºC improvement on a delidded 3770k from using the dot method to completely and evenly covering the entire die. The difference was a little bit bigger when benchmarking with Dry Ice.


Thanks for the post! I have yet to receive my XPS 15 but I will post my results after repasting both CPU and GPU with MX-4, changing the thermal pads to some Phobya 7W/mk ones I have, and undervolting the vCore.

 

You don't have to agree but considering even AS5 isn't conductive enough to cause an issue for the average person who has to ask putting a dot on is probably the safest method.
After all, no one's going to slap as much on as Dell.

 

The problem is not short-circuiting something (which will not happen with AS5), the problem is dies these days are usually not square-shaped and the amount of TIM required to cover the entire die might be overkill depending on the shape. It is true that you will not be using as much as Dell, but it is a waste of TIM and a mess to clean afterwards (spilled out the sides). I would go with the line method instead of the dot one if I don't have the time to go with the smooth spread method.

Anyways, the i7-6700HQ die (picture below) looks more square-shaped than the average die, so the dot method should do the trick as well.


But this is not the case for desktop CPUs (picture below). Dot method will either not cover the entire die or be overkill.

 

Here's a trick, make a thin line down the center. Do a couple of test fits. Works great.

Spreading causes air bubbles.