XPS 15 9550 temperature observations (undervolt + repaste)

Philaphlous - looks like you are the first to get ROG RealBench stress test to run without throttling. And your Heaven Basic score is 2960.

Did you do undervolt as well?

Are you running the i5 7300HQ?

Provide any details that would be relevant for the p. 80 summary including temps and tweaks. Thanks!

 

Firestrike - 5597
Unigine Heaven (basic, ran after looping for about 10 minutes) - 2372

Firestrike Max CPU 75C / GPU 76C
Uningine Heaven Max CPU 85C, Max GPU 78C.

(i7/16/UHD. undervolted -120, vrm pads, pads on ends of pipes)

I got TDP throttling on the CPU after a few minutes of Unigine heaven


Fresh unigine basic nets me 2894. Didn't TDP throttle (though I clamped at 35W with TS)

 

How did you get around TDP throttling when running the bench for a long time?

 

The i5 likely was a key factor (the other 4 laptops summarized on p. 80 were i7). The liquid metal may have helped also...

 

Yeah the i5 is the main reason; by comparing the scores the gpu memory overclock doesn't help.

Also, there must be something wrong with your setup eason because unigine heaven never throttled on mine, what were the average temps? (you reported the max temps)

 

These are my graphs from the test. I ran it about 10 minutes in a loop before benchmarking. Note it was a TDP throttle (reported by throttlestop) that caused the dip to 800mhz in the "ring clock" graph.

I've got some 2mm-thick pads with holes for the chokes cut out over the VRM area. It does not touch the case.

 

Nice detailed graphs.

Gpu temps look high but the behaviour shown in the graphs is normal, i'd check the heatsink contact patch on the gpu.

Cpu temps look a bit higher than normal for a unigine heaven run but they're not generally high; it's strange that it throttled when it barely reached 83°, i think that's vrms temp fault.

edit: just read the tdp throttle part, but i don't get how can it tdp throttle if it's designed and set to run at a 45w limit and temperatures were not too high?

 

The GPU can use 55W at turbo, and who knows about the GPU. It's a 130W adapter, so there probably isn't much headroom for running all the components off the mains like that.

You think the GPU might not be mating well with its heatsink? Given the thickness of the thermal pads on the RAM, do you think that it could be an issue with too little paste (rather than too much)?

paste is gelid GC extreme btw

 

If those pads on the vram are >1mm arctic ones then i'm quite positive that the heatsink isn't making good contact with the gpu.

Also, i'd try spreading the paste rather than using the rice sized drop method.

My setup is

• 140mv undervolt
• grizzly hydronaut repaste (thin spread)
• 0.5mm vram arctic pads
• 3mm thick (2 1.5mm stacked) pads on the vrms to make them reach the choke's height and then a 1.5mm thick pad on both the chokes and the vrms' pads (a single piece), this setup doesn't touch the heatsink (it does touch the back cover)
• 2 small (3mm x 20 mm) slices of 3mm thick pads (2 1.5mm stacked) on the gpu heatpipes (the 2 black curvy things)

Some things will change when i get the k5-pro and the conductonaut.

I did your same test (8-10mins heaven + benchmark) and these are the results.

 

I'm using the stock VRAM pads, as well as a 115mv undervolt. I'm not sure how much that 25mv will make (probably not more than 1C).

Are your pads on top of the GPU touching the case? If not, what is the purpose of having the pads on there? Thanks!

 

Stock vram pads shouldn't be an issue for gpu-heatsink mating but then there must be something else, if you compare our tests there's a bit of a difference.

Yes, they're touching the case.

 

Ah OK. I suspect that's keeping temps down a bit by mating the top of the sink to the case. I did that on my 9550, but it sounded like people disapproved of that on this model. I will add pads and run the test again when I get home. If there's still a large discrepancy then I'll do a repaste with a spatula

 

Just got 2916 with the new BIOS.

 

Check out my results...

Edit: I must have crappy vram or the vram is undervolted compared to the desktop variant... I can only get around +260. Tried +300 and it crashed. UGH!

i5-7300HQ version
This is with a +200 memory overclock



Kinda surprised how hot it gets but definitely no throttling. The GPU only downclocks a little. 1721MHz is the final clock after the benchmark...

Couple things I've noticed...
1. Seems there's really bad memory leaks somewhere on my OS... I was sitting at 65% used lastnight with like a 3.9GB cache memory..weird.
2.

 

What drivers are you using? I get a BSOD every time I open nvidia inspector

 

Ok, putting pads on top of the GPU brought my max temp to 74 in uningine. CPU hit 80. 2850.

 

I actually just added 2 4mm thermal pads to the heatsink where it mates the CPU and GPU. I think it helped... I have a full realbench in with 0 throttling...


I've got pics but its nothing fancy...just 2 thermal pads above the heatsink where the GPU and CPU are...

I think I'm at a -0.110 undervolt on the CPU...Haven't tried more yet. Pretty impressed with the undervolt thus far.

 

Does it not put pressure on the mobo when you do that?

Putting a little pad on top of the heat sink so it touches the case helps a lot. CPU temps dropped 10C from 83C max to 73C. GPU still 77C max running uningine looped

 

Stock GPU - http://www.3dmark.com/3dm/18231490?


Checking my charts... hit 65C on the GPU and 61C on the CPU during this test... Surprisingly not overly hard. It's basically not enough time for saturation of the heatsink to occur before the test is over which is good. I'll rerun it with the memory overclocked... I could probably overclock the GPU core on this 1..hehe

Edit: +260 memory +40 core - 5659
http://www.3dmark.com/3dm/18231677?

 

Pics of padding plz! Did you use CLU?

 

https://www.flickr.com/photos/30149337@N04/albums/72157680615207736

Before you judge... just keep in mind that the thermal pads are the same height as the VRM ones... and these are like 3-4w/mK pads...they're super squishy and flatten really easily.. the whole idea is to create a constant pressure down on the heatsink and VRM...and everything else I put a thermal pad on..lol It may look like they're WAYYY too thick but they actually aren't... I had 0 trouble putting the bottom back on and it only took a little pressure to get everything down tight...

Just ran 5 loops of Valley Extreme HD preset. Totally stable...no throttling... I'd consider my work here done! woot woot! Best part is when I'm on battery..fans NEVER turn on...

Hehe...1797MHz on the GPU stable so far...hehe...

 

Wow, you built up quite a bit! Congrats for your temps, but you should check if your keyboard sticks out. Put a ruler across to check if ist touching the glass when lid is closed!

 

Lol good idea. I don't think it is but I'll check.


Sent from my iPhone using Tapatalk

 

Someone needs to make a Google Doc of all the undervolt for Kaby lake... Or maybe I should


Sent from my iPhone using Tapatalk

 

Have at it, if you link it here I'll put a link on the first page.

 

You can start with the 9560 summary I have at p. 80 (which is not updated and not 100% correct). Once Custom90GT references your summary, then I will delete the obsolete p. 80 summary.

There is a google sheets summary in this thread for the 9550 so that might give you ideas for formatting...

 

Or we can use yours Pressing, whoever wants to take that up...

 

Thanks but I only have the 9550; at this point several users know the 9560 pretty well so they will be able to summarize issues better than I can.

Also, it would also be better as a first post link so it does not get burried at p. 80. You might note in bold on first page 9560 discussion starts on p. 7x, but thermal issues are similar in 9550 & 9560 so thread is relevant for all users.

You might also ask moderator to change title of thread to remove word "9550". I did that with the SpeedShift thread as the 9550 & 9560 issues are virtually identical and no need to loose all that good learning...

 

I will do that this coming weekend, I've got so many midterms this week I won't have time till then. I'll try to revamp the first post too, if you think of anything you think I should add to it, you can post here or PM me.

 

Holy cow... this GPU is awesome... Was hoping for a little more boost in performance over stock but I'll take it... Just run stability tests so far but its looking good. I think like 1844MHz stable on the GPU core...still not hitting anything over 70C.


Valley and Heaven record core clocks wrong so its actually like 1844MHz in this pic...Memory is correct though.

 

I'm still getting instant crashes when I open nvidia inspector. :/ plus my nvidia control panel keeps disappearing.

 

Your drivers are borked... you might want to do a clean install of the latest beta. I guess there were issues with the WHQL version with the GTX 1050 and 1050ti cards for some reason... there's a hotfix on techpowerup's website. I'm running the hotfix with no issues. I got 1 crash with Nvidia inspector but its all good now.

 

Hey guys, been very busy with grad school but just wanted to mention that I feel as that I have minimized, at least for now, the tweaks I have to make to successfully pass real bench without PL1 throttling. I haven't really changed anything since my last post, but just wanted to reiterate what tweaks I'm currently using. Below is the HWinfo64 full monitor display, as well as temp graphs for 15 and 30 minute realbench, and Heaven basic test. I do have some form of mild throttling (possibly thermal related) which you will be able to spot. I haven't really looked into it, but it has very mild implications compared to a PL1 drop.

Here are my current tweaks:

-Padded VRM + RAM (Even though the ram padding may be pointless, I just don't want to take off the back for a while)
-Intel Graphics power options: Max Battery Life (this dropped my CPU graphic cores by at least 10 C during high loads)
-CPU Usage: 90%
TurboBoost disabled
Edit: I suppose it is still important to mention I'm using a laptop cooler

***This tweak setup is for realbench stress test. For the games I actually play on occasion(Overwatch, WoW), I set my CPU back to 100% with no issues***

That is it (besides repasting, and clean installation). I don't have XTU installed. I personally would rather just adjust my max cpu % rather than using XTU - just because it is one less third party program I have to use, and changing max CPU usage is similar to just clamping the CPU if I am not mistaken. However, if I used XTU, or decreased my max CPU a little bit more, I would easily pass realbench 30 minute with no throttling of any kind.



Real Bench 15 minute stress test
Temp graphs: http://imageshack.com/a/img924/9050/E49Srg.png
All sensors: http://imageshack.com/a/img922/1946/FzkcvR.png



Real Bench 30 minute stress test
Temp graphs: http://imageshack.com/a/img923/139/yYX1mX.png
All sensors: http://imageshack.com/a/img922/7868/Yrrtyn.png



Heaven Basic (3 consecutive runs - no throttling of any kind)
Temp graphs: http://imageshack.com/a/img923/1098/ue2ufd.png
All sensors: http://imageshack.com/a/img923/5516/fqhSoh.png
Scores: 2916, 2930, 2935

 

Are you still running an undervolt of CPU?

I wonder what is causing GPU throttling?

 

No undervolt currently. I guess I am not 100% sure it is the GPU, but again I haven't really looked into it. I had a hunch it was GPU when I did the tests - but not I cannot really remember why. Will have to take a closer look here soon.

 

Some people don't like to run an undervolt. But, it has some real "free" performance advantages in such a small laptop.

Mainly, you can reduce CPU temps by ~5*C. Perhaps, more importantly with the XPS, that reduction in power demand can:

- reduce vrm temps (= less throttling)
- increase vrm power headroom (via lower voltage demand AND increased efficency of cooler mosfets)

 

I don't know if you can throttle your CPU and then claim you are passing a test "without throttling" :/ nor do I believe a laptop cooler is legit

 

Running without turbo boost means your cpu multiplier is at 25 the whole time (as it shows in the graphs) while it should be at 34, it's normal that it doesn't throttle; you're missing out on roughly 43% of your cpu's performances (33% from turbo boost + 10% because you set the cpu at 90% power).

I'd say try at least to enable turbo boost and tweak around with that.

Also, cpu undervolting is not the same as limiting the cpu power since it just gives the cpu the exact amount of power it needs; you're missing out on some precious degrees of freshness without downsides.

 

I'm getting a slight stutter/frameskip every 1-2 seconds playing games. It's very noticeable with vsync. Is anyone else seeing this? I'm running on the latest hotfix drivers. edit: caused by f.lux it seems! Strange!

 

I know, it's like the people who fit the gearbox & diff from a 1.6 litre to a 2.8l and brag it's fast off the mark when the top ends 90mph instead of 155

 

GoNz0, what were your temps on your heaven bench? I got 2894 with max temp of 78 for both CPU/GPU. Not sure if I should keep repasting.

Edit: I should mention that ambient temp was 25c in the room

 

Yeah I was heading for 80 or maybe over but I don't worry about temps that high running a benchmark. In games I keep vsync on or lock the game to 60fps as anything over is a waste of power and just causes more heat.

 

As mentioned, PL1 throttle silly, since that seems to be the most decremental to performance. We all know you don't like laptop coolers

 

Its interesting there's throttling going on... with undervolt I'm using only around 19-24w depending on the program. Now I did get throttling on the stock machine when I first tried realbench but I find it surprising it would happen now with all my cooling efforts...maybe the GPU and CPU draw just exceed the limit...

 

Pressing thank you for the constructive response as always, it never fails . I suppose it wouldn't hurt to just find the stable under volt and just leave it on. I wasn't aware that it does not affect performance until custom mentioned it a little while back.

As far as games are concerned, having TurboBoost enabled only improved frame rates by a very small margin, and the disproportional performance increase to thermal output is just not worth it in my opinion. I only use it for school, and the occasionally gaming (very occasional sadly). If I was using it more often for CPU intensive tasks then I would definitely look into it.

 

Make a TS profile that disables turbo at 80C.

 

Did an NV inspector modest OC of +100 core +200 mem, got 5900 firestrike. Could probably break 6000 I reckon but... they're just numbers

 

I've found my max memory overclock... +280. Anything over that will cause lots of issues and lock the GPU down in a lower state so I know its unstable.. Kinda sucks but its better than nothing...I like the memory overclocking because it doesn't really create additional heat/power.....they use so little power as it is.. +280 is really nothing for them. They must be at a lower voltage than the desktop cards because according to other benchmarks with the exact same type of Micron memory, they easily hit 8GHz... +280 won't crash, +290 does, and I think to be safe, +260-270 will be completely stable for my needs...

Still unsure on the GPU core overclocking... Depending on the game I can push it more but I do see a slight increase in temps... I'm keeping it under 70C so I assume that's really nothing much to worry about. And I'm not even getting close to a VRM thermal limit.

 

Interesting, I'm sure they lower the voltage supplied to the ram, my desktop 1060 easily hits +850mhz on the ram. We need an unlocked bios, lol...

 

Yea. I was dreaming before I actually got my 9560 that the memory was full voltage. lol

I've had good success with the GPU at +100. Wonder how far it'll go. I'm sure it'll either downclock during benchmarks and null the overclock or I'll finally hit the 56w power draw limit...
A desktop 1050 pulls around 70-75w, take the fans out of the mix and we're talking about 60-65w tops... we should be able to get in the 1800's easily if temps allow us... not sure about the 1900's..lol

Checking temps with the thermal infrared gun... I see almost 50C on the keyboard but honestly, I have no problem typing on it...it actually doesn't feel that hot so maybe its between the keys... and 42C on the bottom of the case. Once I'm in a more stable situation I hope to get a cooling pad...that should end up helping temps some... I dock my laptop as it is right now for gaming so a cooling pad would be the logical next step.