Announcement

Collapse
No announcement yet.

AMD P-State v4 Linux Testing On A Ryzen 5 5500U

Collapse
X
 
  • Filter
  • Time
  • Show
Clear All
new posts
    jonix
    Junior Member

  • jonix
    replied
    It seems this new pstate driver actually provide some power saving in ondemand and schedutil. Certainly an improvement for laptops and handhelds such as Steam Deck.

    Leave a comment:

  • shmerl
    Senior Member

  • shmerl
    replied
    Originally posted by wooque View Post
    Basically useless
    I don't think so. Lowering performance just a bit while increasing power savings could be a good trade off especially for laptops or handheld devices like Steam Deck.

    Leave a comment:

  • dibal
    Phoronix Member

  • dibal
    replied
    Latency (especially with pulsed Loads) would be much more interesting then pure crunching performance with the new driver.

    Leave a comment:

  • agd5f
    X.Org ATI Driver Developer

  • agd5f
    replied
    I don't think the benchmarks are particularly useful for what this driver is intended to do. The the new driver can't magically make the CPU faster. You get what you get when the CPU is running at a particular clock speed. The idea is to give a wider range of clock values to the OS to allow it to have finer grained control of the performance. The old pstate driver tends to perform better on performance-oriented tests because it only has three clock levels to pick from. No matter how much performance you need for a good user experience, you will automatically snap to the nearest clock level. If there were a bios that only defined two clock levels it would probably perform even better than the current sbios with three clock levels. The new CPPC interface gives you a continuum of clock levels. The idea is set the performance level to something that would accomplish the task in a reasonable time, not necessarily the fastest time. If you always want to accomplish the task in the fastest time, always select the highest performance level.

    I think these results would be more telling if you were to look at battery life or total power draw or average clocks over time while browsing the web or using a text editor or watching video using the different drivers. If you are doing any of those tasks, you don't need the absolute best performance, you just need the applications to feel smooth. If you can do that, and gain an extra hour or more of battery life or a decent power reduction, most people would prefer that.

    As a concrete example, consider the following, say you have a CPU with three defined clock levels using the old pstate interface, 1Ghz, 2Ghz, and 3 Ghz. Say you have a application that needs 2.1Ghz to hit it's target performance. With the old pstate interface, you'll constantly be at the 3Ghz clock, with CPPC, you can be at the target clock you want, 2.1Ghz. The absolute performance won't be as good, but you will still get the required performance to hit your deadline, and you'll save a lot of power in the process.
    agd5f
    X.Org ATI Driver Developer
    Last edited by agd5f; 24 November 2021, 02:48 PM. Reason: add example

    Leave a comment:

  • aufkrawall
    Senior Member

  • aufkrawall
    replied
    If you're not CPU bound, you can also just limit maximum clock with performance governor; efficiency will increase. For watching videos or browsing, you can also save power with the conservative governor. System will be more laggy, but still far better than powersave. Another very suboptimal governor isn't really progress.

    Anyway, the main issue likely is schedutil not to be good enough.

    Leave a comment:

  • CaptainLugnuts
    Junior Member

  • CaptainLugnuts
    replied
    For the compilation tests it would be better to have the results in time and Joules of energy used so you can determine which is more efficient.

    Leave a comment:

  • dragonn
    Senior Member

  • dragonn
    replied
    Originally posted by wooque View Post

    -0.4W is not something you will notice
    Well I can say in practice the average on a non benchmark load is more, often at least 1W less, sometimes ~2W

    Also, anyway, let say you laptop draws about ~7W and the battery last up too 7 hours, so you get ~21 minutes more with -0.4W, that is noticable.

    Leave a comment:

  • wooque
    Junior Member

  • wooque
    replied
    Originally posted by tildearrow View Post

    ...unless you want to save power.
    -0.4W is not something you will notice

    Leave a comment:

  • tildearrow
    Senior Member

  • tildearrow
    replied
    Originally posted by wooque View Post
    Basically useless
    ...unless you want to save power.

    Leave a comment:

  • mlau
    Senior Member

  • mlau
    replied
    Originally posted by wooque View Post
    Basically useless
    Not quite, I noticed that with amd-pstate the max turbo is reached more often and longer.

    Leave a comment:

Working...
X