Announcement

Collapse
No announcement yet.

Intel's PCIe Cooling Driver Ready For Linux 6.13 To Reduce Bandwidth When Running Hot

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts

  • Intel's PCIe Cooling Driver Ready For Linux 6.13 To Reduce Bandwidth When Running Hot

    Phoronix: Intel's PCIe Cooling Driver Ready For Linux 6.13 To Reduce Bandwidth When Running Hot

    For the past year Intel software engineers have been developing a PCIe cooling driver to reduce the PCIe link speed to cope with thermal issues. In the future with PCI Express 6.0 this driver may be further adapted to also reduce the PCIe link width when encountering thermal problems. This cooling driver is now ready for merging with the upcoming Linux 6.13 kernel...

    Phoronix, Linux Hardware Reviews, Linux hardware benchmarks, Linux server benchmarks, Linux benchmarking, Desktop Linux, Linux performance, Open Source graphics, Linux How To, Ubuntu benchmarks, Ubuntu hardware, Phoronix Test Suite

  • #2
    What PCIe generation was it where cooling started to become an issue? On my Ryzen 3600 system, which I think is PCIe gen 4, I'm hoping I don't have this problem.

    I do however have the X570 chipset fan so that's something I have to worry about a bit.

    Comment


    • #3
      Originally posted by hamishmb View Post
      What PCIe generation was it where cooling started to become an issue? On my Ryzen 3600 system, which I think is PCIe gen 4, I'm hoping I don't have this problem.

      I do however have the X570 chipset fan so that's something I have to worry about a bit.
      As mentioed in the article, it's not a fan cooling or similar. It reduces the sensed PCIE temperature by throttling the PCIE bandwidth.

      Comment


      • #4
        Originally posted by Modu View Post

        As mentioed in the article, it's not a fan cooling or similar. It reduces the sensed PCIE temperature by throttling the PCIE bandwidth.
        Oh yeah I new that, I was just asking if anyone knew which PCIe generation was the one that started having overheating problems.

        Comment


        • #5
          How about not designing buses that are so easy to overheat?

          Comment


          • #6
            From what I understand, this is just a laying the groundwork for eventual PCIe (thermal) throttling. It gives device drivers a way to control the link bandwidth for any reasons really, not just thermal. Nothing's going to happen until individual drivers are taught about this - there's no generic way to read a PCIe's device temperature.

            Originally posted by hamishmb View Post

            Oh yeah I new that, I was just asking if anyone knew which PCIe generation was the one that started having overheating problems.
            Originally posted by anarki2 View Post
            How about not designing buses that are so easy to overheat?
            It's not "the bus" (that would be the CPU, chipset) that has thermal issues, but rather things like NVME drive controllers which are given the opportunity to put out more heat the higher the speed. Reducing the bus link speed is one way of throttling the device.

            Why don't NVME drives just throttle themselves?, you may ask. They *usually* do, but there are cases where the internal thermal limits are poorly chosen (see Gen.5 drives plain crashing under stress), so having another way to do it can be desirable. For instance, the NVME driver could poll a drive's S.M.A.R.T. log to query the temperature and lower the link speed based on some user-configurable limits.
            Last edited by alexenv; 27 October 2024, 09:24 AM.

            Comment


            • #7
              Originally posted by hamishmb View Post
              What PCIe generation was it where cooling started to become an issue? On my Ryzen 3600 system, which I think is PCIe gen 4, I'm hoping I don't have this problem.
              For NVMe drives, each jump to a new PCIe version is challenging now. Early PCIe 4.0 NVMe drives ran fairly hot and we had to deal with thermal throttling under sustained load. We've obviously seen the ridiculous coolers for first gen PCIe 5.0 drives. The controllers get more refined over time and we get cooler running drives. Being space / cooling constrained like laptop form factors makes this more challenging.

              I have one PCIe 5.0 x4 NVMe slot on my desktop that I was going to put a 4TB gen 5 Crucial drive in, but I think I'll actually put a matching 4TB PCIe 4.0 x4 WD SN850X in there instead to have 12TB RAID 0 pool.

              Comment


              • #8
                Originally posted by anarki2 View Post
                How about not designing buses that are so easy to overheat?
                They're playing fastest and loose trying to squeeze all they can out of the hardware.

                Once upon a time, I overclocked/volted my own shit; now I have to worry about the manufacturers doing it.

                It all comes down to... the desperation of marketing and MBA scumbags.

                Comment


                • #9
                  'Cooling' term honestly makes it look like the word 'throttling' is forbidden at Intel. Seriously, suppose I'm an engineer from the future and I want to find some way to throttle my hot PCIe device... How should I grep kernel code / search the web for it?

                  Comment


                  • #10
                    Originally posted by Type44Q View Post

                    They're playing fastest and loose trying to squeeze all they can out of the hardware.

                    Once upon a time, I overclocked/volted my own shit; now I have to worry about the manufacturers doing it.

                    It all comes down to... the desperation of marketing and MBA scumbags.
                    That's because customers by and large want everything to be as fast as possible.

                    Comment

                    Working...
                    X