Google Spreadsheet - XPS 9560 Undervolt/Temps/Scores

Hey guys. Feel free to navigate your way over to the Google Doc Spreadsheet that I put together and include all your stats and scores.

Let me know if you're encountering issues with the sheet.

https://docs.google.com/spreadsheets/d/1C5sKc99Qse9EzyYy3vadf7NzA-PI4IJfOIabxQsPl04/edit?usp=sharing

In the future I'm planning on adding additional information but this is a good start for us 9560 owners for now.

 

I don't have edit rights.

 

Fixed. This is my first go around doing a public google doc...

 

Hey man what was your CPU clock frequency during the tests? Did it vary or stay at 3.8GHz throughout like it shud? Mine only goes up to 3.4 but it stays there 95% of the time

Sent from my SM-G935F using Tapatalk

 

I never go up to 3.8 during benches. 3.4 as I assume it's using all 4 cores.

 

So is that the max it can go to, 3.4GHz when on all four cores? So it only goddess to 3.8 on core 0?

Btw just ran firestrike with GPU OC and got 6000+ with 6747 graphics score

Sent from my SM-G935F using Tapatalk

 

Yes. Single core max is 3.8, dual core max is 3.6, quad core max is 3.4

http://www.tomshardware.com/news/intel-kaby-lake-cpu-series,33278.html

 

What OC clocks? I went back to dell stock drivers after the .77 hotfix drivers were giving me BSODs.

 

Oh that sucks lol. My 6820HK goes up to 4.4 on all four lol

Sent from my SM-G935F using Tapatalk

 

What pads if any did you guys use during your repaste?

Sent from my SM-G935F using Tapatalk

 

Updated the spreadsheet with my new scores. +100 on the GPU seems to be a good overclock setting for me. Not too hot too!

 

Guys, little bump here. Please keep updating the google sheet with your scores and stats regarding undervolts and temps. Good info thus far!

Attempting even lower undervolts! Heck... -0.150 on the CPU...do it!

 

When you're undervolting, are you doing both the CPU and CPU cache?! I'm using Throttlestop and doing both is the only way to bring temps down.

edit - ok, l've finally found that this is what should be done..

Can i ask if anyone undervolts the Intel GPU too and if it's worth it?!

thanks

 

Are you guys resetting the power and TDP limits using XTU before you change the clock frequencies? Your changes won't have much of an effect if you don't

Sent from my SM-G935F using Tapatalk

 

I keep meaning to update the spreadsheet with my results. Will do that soon

Sent from my SM-G935F using Tapatalk

 

Used AIDA64 stress test for my results.

i5 model, undervolt -140mV CPU / -100mV GPU.
Max temps for CPU: 79C, Average 73C

Don't have temps for GPU since I forgot to add the sensor to the test (didn't know it didn't automatically record it).

 

Looks like -0.145 undervolt is my max and 1911MHz on the GPU will kinda crash 3dmark benchmark... It doesn't fully crash but it doesn't complete and says cancelled by the user. My guess is its a built in protection...? So +160 on the GPU Core and +270 on the memory seems to be the overclocking potential of the GTX 1050 on my laptop...

That's like at least a 10% step up in performance over the stock GPU and 0 throttling...

Right now I'm pretty comfortable with a +80 on the GPU core which brings it to around 1811MHz when gaming...seems to be a good blend of better performance and temps stay in check...GPU memory OC doesn't add much heat so I'll OC that to +260.

 

Let the games begin...

 

Other people have been able to PM me here so I dunno why it'd block you...
What needs changed? If you really insist create a new tab and copy over the data to your new format... but I can certainly add additional columns for info.

 

No worries.

 

Is info from the XTU stress test valuable? If so, I'll add to the doc

 

Some tend to avoid XTU because of certain issues.
For stress testing I used Heaven + Prime95 at once, at least 15 min for PL throttling to stabilize.
Though I think it would be better to use something with more static GPU load, no load disturbances due to scene changes, for cleaner signals. Furmark is, but is not recommended because it supposedly fries laptops. Better ideas?

 

Definitely don't use prime95, that will fry your laptop just as bad as anything else!

Sent from my SM-G935F using Tapatalk

 

Well then my 9550 is already deep fried Also with the small FFT option which runs even a bit hotter than default.
I don't suggest Furmark because I've seen many warnings that it fries laptops. Though I've been running it for a minute, and after repaste it runs unthrottled. Before repaste the GPU would get throttled within seconds. So it did with Heaven. The local CPU and GPU throttling both seem to act quite fast and smooth (unlike power limit "VRM" throttling where the time constants are longer, there is switching/oscillation etc.).

I'm also occasionally loading the CPU for work when doing simulations, that's not much different than Prime stuff. And I might start torturing the GPU too with CUDA. I got into this whole benchmark business because the laptop was fan-noisy and downthrottling at work stuff and things like Google Earth, which wasn't happening with my old one.

So, what would be the recommended GPU stress test, comparably hard as Heaven but with a steady GPU load (no changing of scenes or such)?

 

How does Prime95 fry a laptop? If you want to undervolt, you need to test stability and prime95 is a great way of doing so. If your laptop is getting too hot during prime then you need to work on cooling.

 

Prime95 only maxes my CPU at like 62C. Actually way less heat than say realbench or a CPU/GPU combo...heck BF1 heats my laptop more... Having your laptop go nuclear is a combination of CPU and GPU stress...especially since the heatsink is 1 unit for both...

 

Right, my point is that Prime95 is a great stability test (especially 28.10) and is safe to run. I'm not sure where Pete got the hesitation to run it.

 

Well definitely don't run FurMark unless you want to shorten the life of your laptop...

 

I wanted to ask, can you suggest another popular stress test which runs stably like Furmark but with less GPU load, comparable to Heaven or such (something normal for a GPU in games)?
Because Heaven keeps changing scenes, producing disturbances in temperature graphs.

 

It shouldn't change temps that much between scenes... GPU usage should be pegged at 99% throughout the benchmark...

You could also run Unreal Engine 4 - Realistic Scenes stress test...that's pretty cool that I've got... it's a virtual room that you can walk through with realistic rendering... stresses the GPU hardddd

 

The GPU may be utilised 100%, but different types of utilisation result in different heat loads, depending for instance on how much time is spent waiting for data from memory instead of computing. I'd like to see flat lines as much as possible when I'm not changing anything (this may not be possible with discrete control, for instance if fan speed only switches among a few different rpm values not continuously).
Furmark produces a rather static image, so the temperatures are probably rather static too, but haven't run enough time to verify.

 

What is your goal? If you're gaming then temps will vary with time. Not sure what you're going for.

 

I'd like repeatable load conditions where temperatures can be better compared among computers. Also if for trying out different mods. Prime95 is fine. With Heaven, everything is drifting a bit.

 

I mean repeatability is nice, but you're talking about comparing systems that already have a ton of variability. Unless room temps, surface mounting of the laptop, airflow etc are all the same the numbers are going to be off. I think a few loops of heaven with something to log temps in the back is going to be about as good as it gets. You can run Furmark, but I don't recommend it. It'll sure heat up your gpu though, lol...

 

My temperatures between AIDA64 and HWInfo64 is slightly different: is anyone able to identify what "Ambient" temperature is on HWInfo? It corresponds to DIMM temperatures on AIDA64, but HWInfo has its own info for DIMM. Which is accurate, and is 105C too high?

http://imgur.com/TiQLnnD

Edit: So after some research, it seems to be the VRM. I've already undervolted, so the only way to cool it now is through thermal pads?

 

They probably just sample at different times and rates so temps and fq etc. are never the same.

I think the first generic ambient sensor is the vrm (should be the hottest). Can't remember what the others but I think m.2 drive, 2.5" drive, maybe the network card, maybe the battery.

GoNz0 has a 9550 picture with the vrm temp sensor circled. Now you know what it looks like. Then you could just open up the laptop and identify the different sensors and fire some cool air at them to see what HWiNFO64 says

105*C is high. Electronics fail at increasingly rapid rates with heat. Basic physics. Some parts are rated up to 95*C or 100*C but those temps still will shorten life. Neighboring parts that are not rated for those temps may fry quickly. Also if that temp comes from a local temp sensor, actual component temps may be significantly higher...

I really try to keep things below 80*C as much as possible. . .

 

You need to live on the wild side Pressing, come on man, do a 24 hour furmark run

 

hehehe

 

Meth,

As reference, the vrm of the similar 9550 seems to use the AOz5019 mosfets for the CPU. See on page 6 forward see how performance falls off a cliff at high load and high temperatures in an ideal test environment. Dell runs only about 1v to the CPU but pushes these mosfets hard with little air circulation, making the vrm very sensitive to temps...

http://www.aosmd.com/res/data_sheets/AOZ5019QI.pdf

Nobody has a great cooling solution yet. Some people try thermal pads. Brian Anderson (?) had some success with a $5 usb fan blowing at his keyboard towardsd the f7/f8 keys...

This thread will help you but I don't know if anyone has gotten 9560 i7 thermals under control...

http://forum.notebookreview.com/thr...rature-observations-undervolt-repaste.785963/

 

I have an i5, with undervolt of 140mV CPU/100mV GPU, but VRM temps were 105C with paste/padding on VRAM area.

After 4mm (1mm x 4) pads on VRM area, temps are high 70s as opposed to the 100s range; highest as of now is 79C for VRM after pads. 3mm pads equalled same height as the chokes, so added one more pad on top of that. Thinking about padding the other VRM area (according to this image, http://i.imgur.com/ED8MmpJ.jpg), but not sure what sensor that would correlate to. Anyone have an idea of what the "unknown" area that was padded was? I saw another photo that had the same area padded, but that person did not know what it was either.

Edit: Highest is actually low 80s (~82), but briefly.

 

Bottom right in the photo where it say IDK? That's the on-board chipset. It'll show up on HWINFO as PCH.

 

So padding that area might eliminate the fans turning on while charging (PCH 45C triggers 2500 RPM fans). Will it have any negative effects over time by padding that area? My PCH peaks around 79C. Seems the fans only have 2 settings: 2500 RPM or 5000 RPM. 5k RPM hits around 60C PCH. My CPU is 40C idle while on battery power, and 30C while plugged in b/c of the fans.

Edit: I placed pads on that area and diode temps decreased from 48C to 38C while charging with light browsing. The left side of my keyboard (near the Shift key) has also cooled down a bit. Bottom aluminum chassis feels more or less the same temperature. However, fans still kick in shortly after charging even with PCH diode at under 40C. Not a big problem, but fan settings may be linked to something else; maybe the 5th "Ambient" sensor on my HWInfo controls the fans. Thanks again.

Edit2: I've noticed that padding the VRM area increases my idle CPU temps from mid-30s to high 40s. Might be because of reduced airflow? The trade-off of padding the VRM area vs. thermal throttling is worth it, but for everyday use (browsing/Youtube videos) CPU would hit low 50s. Anyone have the same problem?

Edit3: I removed the pads from my PCH diode, and CPU idles/light use is back to mid 30s. Not sure if tradeoff is worth it: a 10C decrease in CPU temps during light use vs a 25C difference on PCH diode (55C w/ pads vs 80~85C without pads) under load. PCH diode idle temps were high 30s without pads vs low 30s with pads.