Announcement

Collapse
No announcement yet.

AMD Linux Graphics Driver To Switch To More Aggressive Power Heuristics By Default

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

  • yump
    replied
    That said...

    Originally posted by aufkrawall View Post
    By that logic, you can also ban Zen chiplet models (i.e. those which aren't monolithic APUs) and AM5 desktop (there are exceptions, but few) as a whole, as they burn energy like crazy in idle doing nothing. Should sum up to many tons of CO2 and other energy production waste on a global scale.
    But instead of a ban for existing products, I'd rather see policies that force desktop computer makers to match mobile devices idle power draw in the future. AMD just gives a crap otherwise.

    While I'm at it: Maybe they should be forced to avoid this utter nonsense also with GPUs. As RDNA3 dGPUs are also just terrible garbage in that regard. >40W TGP when playing a 480p video or gif is dumb even for AMD standards. And this is on Windows with "properly" working "power-saving" features...
    There are tradeoffs, actually. Energy that you carry around with you in a Li-ion battery costs many times as much as energy that comes from a wall plug, and if wall-plug computers were as efficient as battery computers, somebody is doing very bad engineering. What you are proposing is for big government to force someone to do bad engineering.

    If you Regulators insist on regulating something, force the transition to single-voltage PSU architecture (dual voltage, really, including the standby rail). Intel has already done most of the work with ATX12VO. Lenovo/Dell/HP switched to single-voltage years ago on their own initiative, because it's cheaper for them and makes it easier to hit regulatory efficiency targets. The DIY market got stuck on multi-rail ATX because of a combination of everybody not being able to move at once and backwards compatibility ludditism.

    Leave a comment:


  • yump
    replied
    Originally posted by cakeisamadeupdrug View Post

    A vaninshingly small number of home desktop users are using anything other than one chiplet, your criticism is restricted to 12-core CPUs and up, but even so the fact that Intel can start drawing hundreds of watts the second you open a web browser makes the idle use not matter so much.
    It's actually not restricted to 3-chiplet (12 and 16 core). Even the 2-chiplet (6 and 8 core) ones have very high idle power. And I'm given to understand that using XMP memory (which ~everybody does) makes it worse.

    Leave a comment:


  • Spacefish
    replied
    Think i will configure it to stay at the old profile on my system.
    Don´t think the additional power draw justifies the minimal performance gains.

    I was always happy with the AMD Gpus especially how little power they draw compared to nvidia during idle times. performance of my 7800XT is more than sufficient for all games that i play.

    I am quite happy with the performance even in VR.

    The games tend to feel "smoother" on Linux anyway, as there are far less micro stutters due do shader compilations (thanks steam for pre-caching these) or other things running in the background. It´s night and day for some titles "feeled" performance wise.

    Leave a comment:


  • cakeisamadeupdrug
    replied
    Originally posted by aufkrawall View Post
    Entirely not true for idle or low load power draw. You can have a 13900K with DDR5 7000 XMP at ~15W package idle. Good luck trying that with Zen 3, 4 or 5 chiplet desktop CPU.
    A vaninshingly small number of home desktop users are using anything other than one chiplet, your criticism is restricted to 12-core CPUs and up, but even so the fact that Intel can start drawing hundreds of watts the second you open a web browser makes the idle use not matter so much.

    Leave a comment:


  • LinuxNoob
    replied
    Originally posted by agd5f View Post

    FWIW, this does not affect idle power draw. It only affects how quickly the clocks ramp up/down when there is load.

    It does on my Machine. I can see the power ramp up when I switch from "Boot Up Default" to "3D Full Screen" using LACT. With no other change power usage jumps immediately to 40 watts from an average of 11 watts This is on Fedora 40 and with a 7900XTX.

    Leave a comment:


  • Jeff
    replied
    Yes, lets talk about low power issues again. AMD changed the GPU minimum power back in kernel 6.7 and now we can't set low power limits. I used to be able to set my 6750 to 160-180W (or even as low as 100W), and now I can't set less than 205+ watts or something.

    For those who don't know, you can easily set power limit to like 20% lower than stock and only lose like 2% FPS (this is based on my tests in many games as an average). Your card will run cool and fans will run slow without any noise.

    I still can't really accept that this was done, and this change wasn't reverted back.
    Stop increasing your GPU power by 50% to get like 5FPS, it doesn't make sense! lol

    And I know some will say you can still limit the GPU clock it does the same.... no it doesn't, some GPU work load can still use a lot of power even at low clock, and setting the clock even lower will then impact other workload. Ideally you need both: limit GPU clock to set back the factory clocks (to get rid of the stupid OC card overclocked settings and restore back AMD stock clock speeds), and limit power to your desired acceptable heat and noise level.

    Leave a comment:


  • mathieu.cossette
    replied
    Nice post 👍.... Gaming on Linux is becoming better than on Windows.

    Leave a comment:


  • aufkrawall
    replied
    Originally posted by cakeisamadeupdrug View Post
    I don't know how you can say this given everything Intel has been doing since Zen's launch. At its absolute worst Ryzen has always been more energy efficient than anything comparable by Intel.
    Entirely not true for idle or low load power draw. You can have a 13900K with DDR5 7000 XMP at ~15W package idle. Good luck trying that with Zen 3, 4 or 5 chiplet desktop CPU.

    Leave a comment:


  • cakeisamadeupdrug
    replied
    Originally posted by aufkrawall View Post
    By that logic, you can also ban Zen chiplet models (i.e. those which aren't monolithic APUs) and AM5 desktop (there are exceptions, but few) as a whole, as they burn energy like crazy in idle doing nothing. Should sum up to many tons of CO2 and other energy production waste on a global scale.
    But instead of a ban for existing products, I'd rather see policies that force desktop computer makers to match mobile devices idle power draw in the future. AMD just gives a crap otherwise.

    While I'm at it: Maybe they should be forced to avoid this utter nonsense also with GPUs. As RDNA3 dGPUs are also just terrible garbage in that regard. >40W TGP when playing a 480p video or gif is dumb even for AMD standards. And this is on Windows with "properly" working "power-saving" features...
    I don't know how you can say this given everything Intel has been doing since Zen's launch. At its absolute worst Ryzen has always been more energy efficient than anything comparable by Intel.

    There's been a lot of hyperbole thrown around but I think the actual idea that AMD should be trying to come up with an elegant and intelligent solution instead of blasting power at boot time does seem to be a reasonable one.

    Leave a comment:


  • aufkrawall
    replied
    Originally posted by agd5f View Post
    FWIW, this does not affect idle power draw. It only affects how quickly the clocks ramp up/down when there is load.
    Yes, I'm also ok with the change (afair I actually asked to make it default), as it's the lesser evil and power saving shouldn't negatively affect user experience.
    But it's undeniably a crutch and there really should be some kind of heuristics in maybe the firmware that autonomously-ish interprets workloads and clocks accordingly. Sounds like a fantastic use case for an NPU.

    Leave a comment:

Working...
X