Announcement

Collapse
No announcement yet.

A Detailed Look At The ATI Linux Power Management

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

  • Ard Righ
    replied
    I was wondering, is there anyway to volunteer to help get some enhancements for 5xxx mobile GPUs moving forward?

    I have an Acer laptop (manu # LX PMB02 171), which has an ATI HD5650 GPU, and the battery on my laptop seems to chew through quite fast, and I get the feeling this is probably due a lot to my GPU running at full speed the entire time.
    The cpu monitoring apps show CPU speed drops back to 1.2Ghz, which is what I'd expect with an i5.

    So as per the comment above about lack of access to a laptop with these cards in them, is it possible to volunteer to test?

    Leave a comment:


  • piete
    replied
    Hi! I wrote to another thread (http://www.phoronix.com/forums/showt...t=23447&page=6) about my experiences of the power consumption improvements, and only afterwards I remembered this article. I know that Thinkpad is frugal with power, but the consumption measured here is a lot less than on my HP NC8430, so I'd like to see where this test computer arrives with fglrx on Lenny or Ubuntu Hardy plus on Windows, just to compare. At the moment I'm really satisfied with the power management's temperature control, but not that satisfied with the power consumption that is practically double of that of Windows.

    So +1 for comparison request with fglrx (and Windows, if possible)

    Leave a comment:


  • agd5f
    replied
    A couple quick notes regarding the current pm state:

    - Voltage drop is already implemented for r1xx-r5xx chips in the new pm code
    - Some additional asic features (like dynamic sclk and dynamic voltage where the asic scales the clocks/voltage rather than the driver doing it) are also already implemented in the driver for r1xx-r5xx asics. Whether or not these get enabled depends on the flags in the power state entry.
    - Most r6xx/r7xx desktop cards have power saving modes

    Leave a comment:


  • frej
    replied
    Better reporting/gathering of data

    Make the graphs more... ehm interesting? We don't know if it's some extra wireless interrupt that decreases the FPS or the power settings.. so

    Please add standard deviation of several runs....then it's likely that low FPS differences can be argues as equivalent. (Ie ignored).

    Also, sometimes percent is a better scale than than FPS

    Leave a comment:


  • Kalessian
    replied
    I'd say this is a good start, but I think that the real benefit will come with lowered voltages.

    On a related note, is there any way at all to lower the GPU voltage through software in ubuntu? I have to use fglrx because I need ATI powerplay to keep my laptop from getting too hot under normal use. This forces me to use the old LTS ubuntu since my mobility x700 is no longer supported. I would love to upgrade to ubuntu 10.04.

    Leave a comment:


  • sylware
    replied
    It may be a good idea to see if the frequency governor system of linux could fit that power management system.

    Leave a comment:


  • Louise
    replied
    Would it be possible to turn off the GPU completely, if you have a server you only access by SSH?

    Or are on one that risk-averse?

    Leave a comment:


  • bridgman
    replied
    I was under the impression that UPS mode was more like a pending shutdown than a low power mode; or maybe this is something new that I hadn't heard about

    Are you saying that "UPS mode" uses the same ACPI messages/whatever as an AC-to-battery transition in a laptop ? Is that separate from the "you might wanna be shutting down before I takes your power away" mode ?

    Leave a comment:


  • nanonyme
    replied
    Originally posted by bridgman View Post
    Desktop systems tend not to have battery mode - but IIRC all of the test results in this article were made when running on battery.
    UPS counts as battery too there then, I think. Might also be possible to trick the system into thinking it's running on battery with some hackery. (virtual battery whose power level you could change should be interesting for power management benchmarking since then you could see if different power management levels are triggered on different battery levels)

    Leave a comment:


  • bridgman
    replied
    Most power management implementations turn on more aggressive power saving functions when running in battery mode.

    Desktop systems tend not to have battery mode - but IIRC all of the test results in this article were made when running on battery.

    Leave a comment:

Working...
X