Announcement

Collapse
No announcement yet.

How Well Is AMD CnQ Working For You On Linux?

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

  • How Well Is AMD CnQ Working For You On Linux?

    Phoronix: How Well Is AMD CnQ Working For You On Linux?

    One of the latest possible Linux power-related regressions I've heard about is that AMD Cool 'n' Quiet may no longer be functioning too well on Linux-based systems...

    http://www.phoronix.com/vr.php?view=MTIxNjE

  • #2
    The only device with an AMD processor that I have right now is a Brazos-based netbook, and it doesn't seem to have any power management issues. If anything, it exceeds the specifications. I didn't check the power management when CnQ was disabled, though.

    That said, isn't CnQ/EIST losing is relevance lately, in light of ACPI processor (C) states? If nothing is using resources, it just goes to complete sleep by itself. No need to change the frequency, so there is no performance loss.

    Comment


    • #3
      asus f1a75-M pro

      I have a f1a75-M pro in combination with AMD's A8-3870K and after enabling the powernow or any other power saving function in bios the system will locks up. No idea if this is due to the hardware or linux.
      Mard0

      Comment


      • #4
        Originally posted by mard0 View Post
        I have a f1a75-M pro in combination with AMD's A8-3870K and after enabling the powernow or any other power saving function in bios the system will locks up. No idea if this is due to the hardware or linux.
        Mard0
        If PowerNow leads to lockups, this is almost certainly a hardware or BIOS issue. On K10 and later CPU generations, p-state transitioning is done in hardware, so there's little chance for any driver to screw up.

        PowerNow really did suck on K8 generation CPUs, but it's great on later CPUs.

        Comment


        • #5
          I got the impression C'n'Q is and always has been a BIOS related thing, not OS, so i don't see why linux wouldn't use it. I guess depending on how you load the CPU, other OSes compared to windows might not use it as effectively but in the end, its how much power your system draws when idle that matters most.

          Comment


          • #6
            For me it's perfect.

            Originally posted by schmidtbag View Post
            I got the impression C'n'Q is and always has been a BIOS related thing, not OS, so i don't see why linux wouldn't use it. I guess depending on how you load the CPU, other OSes compared to windows might not use it as effectively but in the end, its how much power your system draws when idle that matters most.
            For me C'n'Q works perfectly. It keeps my FX8120 CPU at almost room themperature most of time even at low cooler speed. Really remarkable efficiency I admit. Yet that CPU is able to offer quite impressive performance when it's actually needed. The only thing I wish: AMD GPUs should work the very same ways with Linux opensource drivers. Unfortunately it's not a case so far

            Comment


            • #7
              Originally posted by 0xBADCODE View Post
              For me C'n'Q works perfectly. It keeps my FX8120 CPU at almost room themperature most of time even at low cooler speed. Really remarkable efficiency I admit. Yet that CPU is able to offer quite impressive performance when it's actually needed. The only thing I wish: AMD GPUs should work the very same ways with Linux opensource drivers. Unfortunately it's not a case so far
              I really do suspect that AMD will find a way to get proper powersaving out for their GPUs. Now that they are putting them onto the same die with the CPU, having them run on full blast all the time (and thus negating nearly all powersaving done by the CPU) is a big problem.

              Since these APUs were one of the driving forces behind releasing documentation and OSS drivers (the way I undestand it, at least), having them not work properly is not a good way to proceed. So I'm guessing that proper powersaving (already written) will be released once the technical review is done.

              Comment


              • #8
                CnQ, FX-4170, GA-990FXA-UD3

                I run an FX-4170 under Linux (Ubuntu mostly) on a Gigabyte 990FXA-UD3 and CnQ works perfectly.

                This motherboard is dodgey though, it wont boot USB sticks that are created with Ubuntu's startup disk creator. Very annoying. It does work however with MultiSystem, and also sticks made with PenDriveLinux under Windows.

                Comment


                • #9
                  Originally posted by mard0 View Post
                  I have a f1a75-M pro in combination with AMD's A8-3870K and after enabling the powernow or any other power saving function in bios the system will locks up. No idea if this is due to the hardware or linux.
                  Mard0
                  I own a f1a75-i that fails to adjust vcore with all the powernow features enabled.
                  The same cpu/socket/chipset/software on an asrock A75M-ITX and a gigabyte GA-A75M-UD2H motherboard perform perfectly as expected.
                  This leads me to believe that asus has not properly implemented their acpi tables in the bios. Or maybe their digi-vrm nonsense and special proprietary "features" are causing issues.
                  Last edited by Soul_keeper; 10-28-2012, 01:20 AM.

                  Comment


                  • #10
                    Originally posted by schmidtbag View Post
                    I got the impression C'n'Q is and always has been a BIOS related thing, not OS, so i don't see why linux wouldn't use it.
                    It's really not, at least not entirely. What the BIOS needs to do for PowerNow/CnQ to work is fill the p-state table in the CPU's model-specific registers, and that's it. This table contain values for operating frequency, voltage, etc. for all states. The OS requires a driver that reads the table and instructs the CPU to switch to a certain state, depending on load, according to some governor. (BTW, if the BIOS fails to set up the p-state table correctly, you can do that from userspace before loading PowerNow drivers.) The transition between states is handled by the CPU itself, so the driver is quite simple.

                    Earlier on K8, the driver also needed to accurately do and time all the nitty-gritty low-level details of transitioning between states, and this was quite error-prone, depending on hardware.
                    Last edited by brent; 10-28-2012, 11:15 AM.

                    Comment


                    • #11
                      I just upgraded my kernel from 3.6 to 3.7-rc3 and wondered why CnQ stopped working at all, so I looked into the boot messages and found:
                      powernow-k8: this CPU is not supported anymore, using acpi-cpufreq instead.
                      As I didn't had acpi-cpufreq compiled into the kernel the issue was found.

                      Not sure if I would call that a bad thing through. CPU is:
                      model name : AMD Athlon(tm) II X3 455 Processor

                      Comment


                      • #12
                        Originally posted by pingufunkybeat View Post
                        I really do suspect that AMD will find a way to get proper powersaving out for their GPUs.
                        But currently they broke already existing code even further . In recent kernels (3.5-3.7 kernels or so), dynpm mode is completely broken for my GPU and /sys/kernel/debug/dri/0/radeon_pm_info could output numbers not matching actual state of things in some cases. Now driver unwilling to reclock GPU to 157 MHz (lowest clock, ok for idle desktop) and only uses 600MHz after boot (far hotter and power hungry mode). Earlier kernels were better in this regard.

                        Amazing. I fail to understand: hey, dear AMD guys, why can't you just get some 2-3 GPUs of each 5xxx, 6xxx, 7xxx family and APUs and actually boot some Linux on that hardware to see how it performs? AMD should surely have enough resources for very little test lab where it's possible to see all major quirks in code. But it looks like if it's not a case.
                        Last edited by 0xBADCODE; 10-28-2012, 11:59 PM.

                        Comment


                        • #13
                          Originally posted by 0xBADCODE View Post
                          But currently they broke already existing code even further . In recent kernels (3.5-3.7 kernels or so), dynpm mode is completely broken for my GPU and /sys/kernel/debug/dri/0/radeon_pm_info could output numbers not matching actual state of things in some cases. Now driver unwilling to reclock GPU to 157 MHz (lowest clock, ok for idle desktop) and only uses 600MHz after boot (far hotter and power hungry mode). Earlier kernels were better in this regard.

                          Amazing. I fail to understand: hey, dear AMD guys, why can't you just get some 2-3 GPUs of each 5xxx, 6xxx, 7xxx family and APUs and actually boot some Linux on that hardware to see how it performs? AMD should surely have enough resources for very little test lab where it's possible to see all major quirks in code. But it looks like if it's not a case.
                          It looks like 4xxx are no better, after setting the profile to mid on HD 4850, I'm still at 625MHz core and 993MHz memory i.e. "the defaults".

                          Good thing It's a desktop card and I have dual-slot cooling...

                          Comment


                          • #14
                            This "article" started out ok, but as soon as I got to the first TEST, the lack of intelligence behind it became instantaneously obvious.

                            When you put the CPU under LOAD, it ramps up to MAXIMUM POWER CONSUMPTION. CnQ has ZERO effect when the CPU is loaded. You need to test power consumption when IDLE and under varying (controlled) levels of load LESS than fully loaded, and compare the power consumption at the same work loads.

                            From personal observation, when unloaded, my AMD CPUs all sit cooly at their minimum frequencies. As others have noted (and mine being no exception), power consumption is drastically better than manufacturer's specs.

                            Comment


                            • #15
                              My personal experience with CnQ is that its doing its job excellent if you do normal work.

                              However, if you have full 100% load over large period of time, its better to turn it off and set governor to performance to bump the frequency. This yields in ~10% performance gain over large time period, possibly due to less logic and less ons/offs.

                              Comment

                              Working...
                              X