Search:

Type: Posts; User: fenrus; Keyword(s):

Page 1 of 2 1 2

Search: Search took 0.13 seconds.

  1. also one general note: for benchmarks that run...

    also one general note: for benchmarks that run fully flat out, don't expect many changes; that's the easiest case where pretty much every governor will pick the highest performance point and stay...
  2. don't select governors.,...

    The native P state driver does NOT use cpufreq governors.... no point selecting or setting them (in fact, nothing good can come out that so please just don't do that)
  3. Replies
    20
    Views
    3,641

    this is not about physical destroyed. The...

    this is not about physical destroyed. The hardware has protection for that of course.
    When those protections trigger, they trigger hard, and performance suffers a LOT.

    This is about avoiding the...
  4. I'd love to get others to help out. Having said...

    I'd love to get others to help out. Having said that, I don't quite know how that is going to work yet... I have a load of documentation to write on how to use, and then how to contribute, before...
  5. Now I think you're confused. The average of "half...

    Now I think you're confused. The average of "half as fast" and "twice as fast" is not "neutral" in any math that I can think of (before my morning coffee at least)
  6. fwiw this linearity is useful, including for the...

    fwiw this linearity is useful, including for the graphing software (R with ggplot2) to calculate and draw the std deviation bars
  7. I've been thinking about this, how to fix it, and...

    I've been thinking about this, how to fix it, and I haven't figured it out yet. In some way, "-50%" indicates "half as bad", and "+100%" is "twice as high". It also allows me to do math on the data...
  8. since you claim it is alpha... I assume you tried...

    since you claim it is alpha... I assume you tried it?

    you did not read the website though since this is not an Intel project.
  9. the colored dots are the average results for the...

    the colored dots are the average results for the release in question. there are also tiny black dots which are the results of each individual PTS test (so sometimes you see a "smear" of them, meaning...
  10. I don't think this one was recorded unfortunately.

    I don't think this one was recorded unfortunately.
  11. it makes "the same" be "zero". A lot of people...

    it makes "the same" be "zero".
    A lot of people really do not like graphs without the zero axis there; and if the axis would have to always span 0 - 100 then it becomes useless to see small changes.
  12. yeah I'll try to clarify the page tonight,...

    yeah I'll try to clarify the page tonight, clearly it's not well done right/clear enough right now; thanks for the feedback.

    let me give a specific (but made up) example.
    Say some test gives...
  13. (I presented about this at Plumbers last year,...

    (I presented about this at Plumbers last year, several key RH guys were there and we talked about it afterwards as well)

    Delta RPMs don't really work well. In that for small changes, they're 100x...
  14. hmm. Gentoo wouldn't be the first distro to come...

    hmm. Gentoo wouldn't be the first distro to come to my mind while developing high efficient software update technology ;)
    (doing binary delta's of only those files that changed, solving the...
  15. for each test, the "best of the rest" score is...

    for each test, the "best of the rest" score is used as an index (e.g 100%). the score for Fenrus Linux is then indexed against this score and a delta percentage comes out.
    (the scores are always...
  16. there is absolutely room for real performance...

    there is absolutely room for real performance improvements using compiler flags (see url I posted way earlier in the thread).
    prefetch/prefetchw is not part of that however.
  17. oh compiling with a new enough CPU is a huge gain...

    oh compiling with a new enough CPU is a huge gain at times (just see the graphs of my distro that I posted earlier in this thread).
    prefetchw or prefetch are not part of that however.
  18. eh how? if you think that the compiler can...

    eh how?
    if you think that the compiler can insert a better prefetchw than the hardware prefetchers.. please speak up with an example...


    (disclaimer: I work for Intel on Linux, and also have my...
  19. prefetch

    The Prefetch instruction (and in this case "prefetchw") is almost always a loss. Hardware nowadays has pretty aggressive prefetchers that work on the actual access pattern, and those are very...
  20. some benchmark graphs

    Fenrus Linux went to the RC on March 17th; the graphs at http://linux.fenrus.org/performance/ show that there's modest increases in the cray and graphics-magick tests, and a regression in...
Results 1 to 20 of 21
Page 1 of 2 1 2