News, tips, advice, support for Windows, Office, PCs & more. Tech help. No bull. We're community supported by donations from our Plus Members, and proud of it
Home icon Home icon Home icon Email icon RSS icon
  • More laptop thermals– liquid metal and undervolting, full throttle

    Posted on Ascaris Comment on the AskWoody Lounge

    Home Forums AskWoody support PC hardware More laptop thermals– liquid metal and undervolting, full throttle

    Viewing 1 reply thread
    • Author
      Posts
      • #2287411 Reply
        Ascaris
        AskWoody_MVP

        Previously, in my posts about using laptop thermals on my Dell G3 gaming laptop, I’ve used Folding@Home as the load, set to “Full” on folding power. That is supposed to use 5 cores (out of 6), but in reality, it ends up maxing four cores and leaving the other two near idle.

        With laptops, the performance potential of the CPU usually exceeds the heat removal performance of the cooling system, so there will usually be throttling under heavy sustained loads. The CPU temp will be whatever you set, or the hardware vendor has set, as the maximum; it will simply throttle performance up and down to stay below the max temp.

        The folding performance I had before with standard voltage and using high-end (but not metal) thermal paste, Gelid GC-Extreme, had my CPU maxed at 3100 MHz (3.1 Ghz). I don’t remember what the temps were, specifically.

        The CPU on this laptop has a max of 4100, but it only does that if a single core is loaded heavily. I thought sustained speeds at 3100 was decent for 2/3 of the cores being heavily loaded, given that this is a laptop with a skimpy heat sink… but if it could be improved, it should be improved!

        I’d undervolted at some point during the process, but I had not realized that Dell had taken my extra cooling away when I installed the latest firmware revision, which addresses a security issue that does not affect me called Plundervolt ™.  It’s one of those security issues that has its own logo and everything, so I gave it a trademark.

        I wrote a post not long ago about how I got my undervolting back, but before I did that (and before I realized my undervolting was gone), I’d noticed the higher temps and started looking at the possibility of using the liquid metal thermal material. I wrote a post about that too.

        With the liquid metal, I had about 3400ish on the same “five” core folding setup (maybe a touch more). It was again only really four cores, but it was set to “auto” as in the previous test, and the display said cpu:5, indicating 5 cores. A gain of 300+ MHz just for changing the thermal material is quite nice, I’d say.

        Then I realized my undervolt was missing, and after I got it back, I found that my boost would go all the way to 3900 on four cores. I got 500 more MHz out of it from undervolting (during which I also removed the long CPU power limit, so it can stay at 45 watts all day as long as the cooling’s there, instead of about 8 seconds).

        Impressive, but still only loading four cores.

        I wanted to try all of the cores.  I did, and here’s what I got:Screenshot_20200809_095254

        I should have closed the sensor browser pane first, but the important info is still there. The test started right where things began to ramp up on the left.

        It held 3500 MHz for the whole time of the test (however long it took to scroll all that way, a minute or two), all six cores at 100% utilization, with temps on all cores at 70C or less, and a package temperature (in the system tray lower right) of 67C.  It’s still holding that level of performance now, a half hour after I took that screenshot!

        The

        It takes the fans a while to ramp up, but it seems well tuned with regard to the thermal inertia of the heat sink… even with the slow ramp up, temps didn’t get too high.

        Intel’s thermald service that I previously praised was inactive during this test. It had made a big difference in temps with no frame rate loss when gaming, but the benchmark performance (which Folding@home is quite similar to, with its artificially high sustained CPU load) was poor.

        Right now, it’s bouncing off of the 45w limit, not the thermal limit.  Usually it comes preconfigured to go to 45w (the short limit) for about 8 seconds, then to go down to a lower watt limit, the long limit (I don’t know what the wattage of that is), but the undervolting program allowed me to remove the short limit. As you can see, I still have plenty of thermal headroom, as this CPU has a max temp of 100C. I could boost the CPU more (not sure if it would work or not, though), but since this is a gaming PC, I want to have some cooling left for the GPU too.The heat sink/fan setup has two fans, one by the CPU and one by the GPU, and each fan with its own finned heat sink, but the CPU and GPU are tied together by the two large, flat heat pipes. The GPU is under no load in my test, but you can see it’s still up to 51C in the screenshot (in the systray). That’s nearly all CPU heat… the idling GPU would be in the low 30s otherwise.

        Ambient temp is 78F/26C during the test. The laptop is flat on the desk, other than my laptop lifts (rubber feet) stuck to the rearward portion of the bottom. It’s been 48 minutes now, still holding the same temps and CPU speed.

        It took some effort to get to this point (which Dell could have done too, though it would have raised the price of my laptop), but I’d bet I get more CPU performance from this than I would from my overclocked Sandy Bridge desktop (4500 MHz, 4 core, no power limit).  I haven’t even tried to see if I can give it a severe overclock… 4500 is what I could manage with no more than a 10% boost in voltage, which is what an Intel rep said was the maximum safe voltage. It was at a conference somewhere, and they tried to pin him down to exact voltages, but all he would say is “ten percent.” So I looked at what my stock voltages were, added 10%, and considered that my target. I got 4500 MHz, and there it has been for several years, with no instability noted… so it seems to be good.

        I doubt I’d ever reach sustained 100% CPU on all cores in gaming, though.

         

         

        Group "L" (Fedora 32 Linux w/ KDE Plasma).

        Attachments:
        1 user thanked author for this post.
      • #2287448 Reply
        Ascaris
        AskWoody_MVP

        Well, I managed to resist the temptation to try to shoot for the sky for a couple of hours. I set the power limits above 45, and as I had suspected, the system firmware (presumably) vetoed that idea. After entering the new parameters (I told it to go to 52w), I looked at the wattage meter (using the intel-undervolt program in Linux, the command is intel-undervolt measure ), and it was still bouncing around at about 44.9 watts, clearly doing its best to stay clear of the wattage limit. I tried other numbers, even down to 46, and it refused to go higher than 45.

        I’m not too unhappy about that, as I did want to keep some thermal headroom available for the GPU. It’s fun to experiment though!  It’s possible that the VRM MOSFETs can’t handle the extra current, even for a short time. I hope they can handle the extra heat of being at the higher power level for over the default short power limit. They do have heat pads connecting them to the heat sink, so they’re actively cooled.  The heat sink they’re attached to is not being pushed to the maximum heat level that they would usually see with a non-undervolted, thermal throttled CPU (CPU temp in the mid to high 90s C), so they should hypothetically be cooler than they were before, since they can’t possibly be cooled below the temp of the CPU that shares their heat sink.

        It’s not as though I expect to be pushing this hard a lot, though. I’m using the FAH client as a stress program, but for actual folding, I skip the CPU and let the GPU have the full capacity of the heat sink. The performance contribution of the CPU to the folding effort is far lower than its heat contribution to the heat sink!

        Group "L" (Fedora 32 Linux w/ KDE Plasma).

        • This reply was modified 1 month, 1 week ago by Ascaris.
    Viewing 1 reply thread

    Please follow the -Lounge Rules- no personal attacks, no swearing, and politics/religion are relegated to the Rants forum.

    Reply To: More laptop thermals– liquid metal and undervolting, full throttle

    You can use BBCodes to format your content.
    Your account can't use Advanced BBCodes, they will be stripped before saving.