Announcement

Collapse
No announcement yet.

Red Hat Recommends Disabling The Intel Linux Graphics Driver Over Hardware Flaw

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

  • polarathene
    replied
    Originally posted by intelfx View Post

    Hm, which exactly CPU and kernel version was this?

    My i7-8550U (Kaby Lake) is definitely broken with 5.4.x, see here: https://gitlab.freedesktop.org/drm/intel/issues/614
    i3-10110U (Comet Lake) UHD620 graphics. Running the current stable Manjaro KDE ISO(18.1.5 with kernel: 5.4.6-2-MANJARO) via USB, powertop shows in the "Idle stats" section 2nd column under "GPU":

    Code:
    Powered On 21.8%
    RC6                78.2%
    RC6p                0.0%
    RC6pp              0.0%
    While typing that those two values changed to 11.7% / 88.3%, and then 18.6% / 81.4%, and once more to 36 / 64 and shortly after back to the 20 / 80 range, and after moving the mouse around on elements a bit to 95 / 4, I tried moving the mouse on/off a UI element to spam it's hover state on/off and that gave a 100 / 0. So it seems to be working?
    Last edited by polarathene; 01-18-2020, 03:14 AM.

    Leave a comment:


  • gukin
    replied
    Originally posted by starshipeleven View Post
    Does Intel Graphics even run Crysis? I'm pretty sure they still can't.
    Au contraire, I was fiddling around with my laptop (sporting an Intel 8250U processor with 620 graphics) and with DXVK 1.51 Crysis is very playable and without the awful "flashing grass" issue that has dogged Intel graphics on Crysis before. Of course this is at 1366x768 and high settings. Still the frame rates were, as I recall, in the 40-50 FPS range.

    The flashing grass issue is still present when not using DXVK though and it really makes the gaming experience unpleasant.

    Mageia 7, wine 5.0 RC5, Mesa 19.2.3, llvm 8.0. kernel 5.4.10.

    I don't have anything useful to add but if you're stuck in an elevator with your Intel laptop and your GOG collection, you can play Crysis.

    Leave a comment:


  • intelfx
    replied
    Originally posted by polarathene View Post
    All I remember is was seeing rc6 in the idle stats tab I think for powertop and it had the bulk of the % assigned to it.
    Hm, which exactly CPU and kernel version was this?

    My i7-8550U (Kaby Lake) is definitely broken with 5.4.x, see here: https://gitlab.freedesktop.org/drm/intel/issues/614
    Last edited by intelfx; 01-17-2020, 12:22 PM.

    Leave a comment:


  • polarathene
    replied
    Originally posted by intelfx View Post
    Hm, which exactly case are we talking about?
    Sorry there was a similar article that I mixed your response up with, which had the benches for performance regression but lacked impact on battery which a user claimed went up a fair bit at idle prior to the patch.

    Originally posted by intelfx View Post
    I wasn't talking about 4.19...
    I know, I mentioned it because it's the LTS prior to 5.4, so most install media ships with it by default. On Manjaro which I use I'll have to check but I think kernels that aren't LTS get EOLed so using kernels inbetween would require getting source and compiling my own I guess as my Comet Lake doesn't seem like it boots with 4.19 kernel for whatever reason(could just be their stupid hardware detection getting in the way and assigning a bad driver or something since the IDs for Comet Lake would be missing in the kernel?).


    Originally posted by intelfx View Post
    Anyhow, to check rc6 residency you can either look into powertop or run turbostat with something like this:

    Code:
    sudo turbostat -S -s GFX%C0,GFX%rc6,Busy%,Pkg%pc2,Pkg%pc3,Pkg%pc6,Pkg%pc7,Pkg%pc8,Pkg%pc9,Pk%pc10
    Then look at GFX%rc6 and (as an integral measure of how good is your power saving) Pkg%pc8 to pc10.
    I'll give that a go soon, thanks. All I remember is was seeing rc6 in the idle stats tab I think for powertop and it had the bulk of the % assigned to it.

    Leave a comment:


  • intelfx
    replied
    Originally posted by polarathene View Post
    Michael tested a Whiskey Lake for Gen9 to see if there was any power regression as a result despite the minimal performance regression for Gen9 graphics, and found no change in power draw, unless he didn't test it right(eg tested on at load rather than idle..)
    Hm, which exactly case are we talking about? I was talking about the mitigation described in the linked article, i. e. blacklisting the i915 kernel driver. Applying the proper mitigation for gen8/9 doesn't seem to carry a significant penalty indeed.

    Originally posted by polarathene View Post
    As for the RC6, how to I verify that? My Comet Lake seems to report RC6 in PowerTop, but perhaps I'm looking at the wrong place, don't know too much about such yet. Not sure that I can test on older kernels either as 4.19 didn't seem to boot from install media(live usb) with such
    I wasn't talking about 4.19... Anyhow, to check rc6 residency you can either look into powertop or run turbostat with something like this:

    Code:
    sudo turbostat -S -s GFX%C0,GFX%rc6,Busy%,Pkg%pc2,Pkg%pc3,Pkg%pc6,Pkg%pc7,Pkg%pc8,Pkg%pc9,Pk%pc10
    Then look at GFX%rc6 and (as an integral measure of how good is your power saving) Pkg%pc8 to pc10.

    Leave a comment:


  • polarathene
    replied
    Originally posted by intelfx View Post

    You're looking at a 1.5x-2x reduction in battery life on any recent mobile Intel platform. They really require very tight integration/coordination between all available parts of the CPU and the chipset to have any power savings at all.

    That said, if you're on Gen9, then kernels 5.3.8 and newer already come with disabled rc6 And nobody is doing anything about it.
    Michael tested a Whiskey Lake for Gen9 to see if there was any power regression as a result despite the minimal performance regression for Gen9 graphics, and found no change in power draw, unless he didn't test it right(eg tested on at load rather than idle..)

    As for the RC6, how to I verify that? My Comet Lake seems to report RC6 in PowerTop, but perhaps I'm looking at the wrong place, don't know too much about such yet. Not sure that I can test on older kernels either as 4.19 didn't seem to boot from install media(live usb) with such

    Leave a comment:


  • r08z
    replied
    Wow Intel sucks big monkey balls...

    Leave a comment:


  • libv
    replied
    Now if only DRM wasn't such a mashed together mess, and there was clear separation between display and the various engines (of which 3d is only one). But that's just how fd.o/drm people do.

    Leave a comment:


  • kpedersen
    replied
    If these are for the older drivers then in most cases you would already be running in software mode.

    For example Blender requires GL 3.x and above. The older GMA drivers only implement OpenGL 2.0. So you would need LIBGL_ALWAYS_SOFTWARE=1 anyway.

    Interestingly I seem to have a more responsive experience when using the software renderer than these older GPUs anyway :/

    Possibly the only issue is the limited display resolutions of vesa.

    Leave a comment:


  • intelfx
    replied
    Originally posted by polarathene View Post

    Not a problem until the mitigation via kernel update enforces it. I'm still curious for other confirmations about increase in watts drawn for Gen9+, I'll take the local vulnerability over losing a notable hit on battery life..
    You're looking at a 1.5x-2x reduction in battery life on any recent mobile Intel platform. They really require very tight integration/coordination between all available parts of the CPU and the chipset to have any power savings at all.

    That said, if you're on Gen9, then kernels 5.3.8 and newer already come with disabled rc6 And nobody is doing anything about it.

    Leave a comment:

Working...
X