Alienware 13 R3 Thermal Testing - Should you cool your PCH with mods?

Discussion in '2015+ Alienware 13 / 15 / 17' started by shadowyani, May 18, 2017.

  1. shadowyani

    shadowyani Notebook Deity

    Reputations:
    210
    Messages:
    789
    Likes Received:
    132
    Trophy Points:
    56
    You should read, but TL;DR, yes, the PCH is a part that requires cooling. The conclusion is located on Page 4.

    So when playing games I would notice the performance of my entire system suffer despite the CPU and GPU being in very safe temperature ranges. This would manifest as sudden drops in CPU clock speeds for up to 10 seconds at a time. I would get results like the following:

    perfDrop1.png
    perfDrop2.png

    So warranty came out to swap my motherboard and the technician ruined the computer. It failed 3 days later. But during those 3 days I was able to play games and get the same results. After the computer failed, a replacement came to me. I also got the same results. That makes for a total of 3 different motherboards all with the same results. I decided to investigate as to why this was happening and why Dell considered the computer to be working despite this behavior.

    It became clear that the PCH temperatures were alarming; anything in your system that has performance that is sensitive to heat typically has a temperature sensor that can be monitored. Yes even SSDs can throttle. The PCH is a very important part of the system and has a huge effect on the system’s performance; it replaces what we used know as Northbridge and Southbridge. I speculated that this chip might not be connected to any cooling at all. I took the computer apart and located it, and sure enough there are no heatsinks or fans attached to the thing. Ok fine, so why does Dell think it’s okay to have it passively cooled? Probably because everyone else leaves it passively cooled, its probably within spec…except their competitor’s computers are larger than the svelte for a gaming laptop Alienware 13 r3. I needed to be more rigorous and proper with my testing.

    In order to properly test thermals you need to eliminate as much variable behavior as possible. This means for the Alienware 13 r3 you need to hop into the BIOS and enable CPU and Fan Performance modes. In Windows you need to prefer the Nvidia GPU, set it to prefer maximum performance, and then set Window’s power plan to high performance. The result is a stable test platform where clock speeds for the CPU and GPU should never fall below base clocks during idle and the Nvidia GPU stays on when no intensive graphics are running. Furthermore, the fans stay at maximum speed and never variate.

    To illustrate this here’s the baseline that results once all those options are ticked. That means that the screenshots from here on out are the best results possible for a stock machine since the fans are never allowed to rest. To keep my testing accurate, all parts are stock except my choice of memory and storage and no re-paste has been attempted:

    baseLineWithPerfModesOn.jpg

    I also shut down unnecessary background applications to get real idling. Yup. Clock speeds never drop below base and my fans are roaring. After each test, I give the system a chance to return to this base state.

    So my first test was to do what the Dell remote technician did for me over the phone. He simply ran a CPU stress test. Here’s my version of that same test but using OCCT for the CPU load:

    OCCT.png

    With these results the technician was ready to tell me there was nothing wrong with the system. If all your computer has is a CPU then yeah sure that’s probably true. Notice though that an average temp of 89C and a spike of 93C would be a huge cause for concern for many people but look at the actual CPU clocks. The average was a multiplier of 32x which indicated the computer was well into Turbo Boost territory with short minimums down to 28x which is the base. The computer simply could care less that its 90C or it wouldn’t be on Turbo. You might get your lap burned but your computer doesn’t care about your feelings

    Now let’s look at another test in vacuum. The GPU only. I used Unigen Heaven for this. Dell uses Valley, but the tech never used it he declared to computer to be working with the CPU test only. Here are GPU-only results:

    Heaven.jpg

    There are a few notes about this result. First Unigen Heaven is a highly variable test, in order for me to keep the GPU load at 100% I had to pause the camera movement. I also max out all the detail settings so that the GPU load stays at 100% regardless of whether or not the CPU is being taxed. Tessellation and its associated sliders are maxed out in the 3D window’s interface; they aren’t available in the menu.

    The GPU under load is less hot than the CPU is under load. However the PCH is doing something very different here than it does in the CPU test; it climbs all the way up to 83C whereas in the CPU test it only went to 77C. It isn’t possible to directly target the PCH with the tools I have.

    [continued below]
     
    Last edited: May 28, 2017
  2. shadowyani

    shadowyani Notebook Deity

    Reputations:
    210
    Messages:
    789
    Likes Received:
    132
    Trophy Points:
    56
    For the next test I do the CPU and GPU combined using OCCT + Heaven. It is important to set Heaven to what I’ve specified preceding this because the introduction of a CPU load can limit the graphics load if the CPU isn’t giving the GPU enough to work with. You’ll find lots of simultaneous tests like this out on the internet where the GPU load is as low as a few percent when care isn’t taken to ensure the GPU always has work; this is variable and the results are unreliable. So here are my results:

    OCCT + Heaven.jpg

    These results are telling. For the majority of the test system performance is relatively stable with the CPU and GPU clocks variating slightly but always above base clocks. Its very clear that the chips are aiming for 90C as their performance target and throttling isn’t ever happening. However towards the end of the test, not even 10 minutes in, the PCH temperatures have run away. The CPU drops a full 300 Mhz below its base, still not thermal throttling however, at least not from the CPU’s perspective and HWINFO’s throttle indicator confirms this. The GPU however has a massive valley in its graph. Its temperatures are also nowhere near dangerous nor are any thermal throttle flags raised by HWINFO. The way the PCH works isn’t as transparent to me as the CPU and GPU but I believe it is unable to service the CPU and GPU when its own temperatures run rampant which causes the performance drops. I had already seen enough to know that the PCH, without a heatsink or fan, was going to continue to run away and I didn’t want to risk damage.

    For the next test I do real world. Let’s play some Ark! Here are the results:

    ArkOnly.jpg

    I ran this test for over 30 minutes because I was frankly surprised to not see any throttling. PCH temperatures are still going rampant though. CPU load averages 31% and the GPU averages over 91%. It would seem to me that despite the high PCH temperatures, it is able to continue its work in this situation because the combined load during Ark is still less than that of OCCT + Heaven. Obviously it brought a smile to my face to see my game stop throttling.

    But that wasn’t good enough when I play Ark I don’t just have a 31% CPU load. I have browsers in the background to access the Ark wiki, I have Discord to chat with my tribemates, and I have Crashplan running because I backup the private server we play on. Here are my results with those things running:

    ArkAndBackgroundTasks.jpg

    So the CPU load is only slightly higher with all those things running, wow the 7700HQ is a beast! But the PCH has gone up a few degrees. This isn’t slight; every degree increase is in heat is not a linear increase in work. But still the combined load isn’t enough to choke the system.

    Here’s what I think all this means. I think it means that in the best case scenario the Alienware 13 r3 can in fact handle insane loads without any user intervention provided that you have your fans on full blast at all times and you set your system up with all the performance options I’ve listed at the beginning of this article. It only starts to choke when you have a true 100% load on BOTH the CPU and the GPU when it is configured this way, and even so, its only when the PCH temperature begins to run away. Were the thing cooled, you could probably run indefinitely at 100% loads.

    But that’s not the real-world. In the real-world you let the entire system manage itself; fans are variable and so are clock rates. Sure enough when using the machine as Average Joe might use it, I see this:

    RealWorldAverageJoe.jpg

    And I’ve seen this on my first machine, its replacement motherboard, and the full system replacement I have now. Is it any wonder I’m surprised that you guys are getting such variable results?

    When you’re designing a computer and you go down the list of specs it’s easy to target for a situation where all the multipliers and fans are locked. Once you get it into the hands of the users who real-world use cases require variation in clock speed and fans you get a whole different ballgame. This is why we as users need to be more precise in our testing; when someone quotes a massive drop in temperatures because of a re-paste job more often than not the before and after tests are as variable as the choice of paste itself. We can easily mislead each other.

    [continued below]
     
    Digital1337 and ThatOldGuy like this.
  3. shadowyani

    shadowyani Notebook Deity

    Reputations:
    210
    Messages:
    789
    Likes Received:
    132
    Trophy Points:
    56
    So where do I go from here? Well I’m going to attempt to actively cool my PCH similar to what fellow user LM64792 has done as shown here:

    Mod.jpg

    It’s an elegant mod that uses a graphite sheet to tap into the system’s active cooling. When I’ve done my version of his mod I’ll post those results. Will it work? Maybe. Maybe not. Is it worth my time? Definitely!

    So should you demand that your gaming laptop be able to handle a simultaneous CPU and GPU stress test? Yes! Will Dell demand that? Maybe not! Will Dell honor the warranty if a simultaneous stress test kills your computer? Maybe not! But you should still demand more, eh?

    Don’t worry, this is fun for me.

    [end article]
     
  4. ThatOldGuy

    ThatOldGuy Notebook Deity

    Reputations:
    614
    Messages:
    1,656
    Likes Received:
    1,125
    Trophy Points:
    181
    Its hard to tell from the picture. Is the fan using a regular 2 pin JST? If it is then the micro fan mod should work with a Y splitter
    upload_2017-5-18_16-23-34.png

    https://www.amazon.com/Crazepony-In...UTF8&qid=1495138943&sr=1-28&keywords=2pin+jst

    Would you say there is about 5mm of space for a micro fan?
     
    Digital1337 likes this.
  5. shadowyani

    shadowyani Notebook Deity

    Reputations:
    210
    Messages:
    789
    Likes Received:
    132
    Trophy Points:
    56
    My makeshift graphite sheet heat-pipe should be arriving soon so when that arrives I'll have my computer opened. While I don't plan on mounting a micro fan at the moment I am more than happy to help you take the measurements and look carefully at the fan connector while I'm at it. Do you know the dimensions of the fan and how you plan on mounting it? Where does the airflow go? Probably a bunch of other things that need very careful consideration that I haven't even thought about yet; I'm currently thinking about the graphite sheet mod so that's kinda consumed my brainpower at the moment.
     
  6. ThatOldGuy

    ThatOldGuy Notebook Deity

    Reputations:
    614
    Messages:
    1,656
    Likes Received:
    1,125
    Trophy Points:
    181
    I would remove the square box part of fan with dremel. Mount passive round center directly to PCH with double sided thermal tape (same stuff used to mount raspberry pi heat-sinks). Wouldn't modify the case till I saw results of this, the in case air flow may be enough.

    Just theory crafting here though. Mine is at the "It ain't broke, so don't open it" phase
    Only things I would need to know is if the current fan connector is 2 pin JST connection and If there is at least 5mm clearance over PCH
     
  7. shadowyani

    shadowyani Notebook Deity

    Reputations:
    210
    Messages:
    789
    Likes Received:
    132
    Trophy Points:
    56
    Oh I see. So you'll need 5 mm of vertical space just for the fan? The thermal pad is going to make it a little thicker won't it? I'm visualizing your end result its gonna look like a traditional CPU fan on a desktop that sits directly on the chip right with pad instead of heat sink? I will also take a clear look at the fan connectors and provide that information as well.

    Yes I agree opening the laptop if its working ok really won't help but make the fit and finish worse especially since modern laptop designs are over-reliant on plastic tabs that get worn down with each snap and unsnap cycle. So to save you the trouble I will remember to help you take measurements since mine will necessarily have to be open very soon.
     
    ThatOldGuy likes this.
  8. ThatOldGuy

    ThatOldGuy Notebook Deity

    Reputations:
    614
    Messages:
    1,656
    Likes Received:
    1,125
    Trophy Points:
    181
    Yes, 4mm fan 1mm pad (ideally the 2.5mm fan, just can't find it for sale). Just like a CPU desktop fan, just tiny.

    These things are crazy to open up now. Just did a re-paste on a Vaio Z Flip I picked up the other day (50% off retail). was like 30 screws and the snaps were tiny but all around, also had to take the rubber feet off to get to screws. Brilliant little laptop though.
     
  9. Digital1337

    Digital1337 Notebook Evangelist

    Reputations:
    146
    Messages:
    604
    Likes Received:
    93
    Trophy Points:
    41
  10. rinneh

    rinneh Notebook Virtuoso

    Reputations:
    88
    Messages:
    2,663
    Likes Received:
    738
    Trophy Points:
    131
    That mod will pretty much not do anything...

    @iunlock did a better solution. Use heat resistant 3M elictrical tape and tape of the green substrate part of the PCH. Then use a fairly thin heatsink but more wide than the chip die itself and use a glue gun to glue down the heatsink onto the elictrical tape. Thus you can remove the heatsink after if needed but until then you have a solid connection between your heatsink and PCH.
     
    Pete Light, iunlock and Vasudev like this.
Loading...

Share This Page