Announcement

Collapse
No announcement yet.

Benchmarking 11 Linux Distributions Across Dozens Of Workloads On The Intel Core i9 10980XE

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

  • splorgo
    replied
    I also would like to have Arch in there
    The out-of-the-box in Arch lands you in the install environment. To even reach the chroot you would have to have made choices regarding your file system which determines quite a lot.

    There is no default install of Arch, even if you just follow the wiki and find yourself just booted, you'd still have to install quite a lot of packages of your personal choosing even to be able to run most of the tests.

    ‚Äč

    Leave a comment:


  • Danielsan
    replied
    It is not a surprise that the faster distros are the lesser bloated...

    Leave a comment:


  • Grinch
    replied
    Originally posted by edwaleni View Post
    So its very possible (I haven't looked under the hood) that Intel did do some optimizations.
    No, sorry, but this is just not possible under any other circumstances other than the benchmark being broken. x265 is an extremely optimized program, it uses hand-written assembly (which the compiler does not affect in any way) for practically all hot paths, and even if you disable assembly optimizations the differences won't be anywhere near this large unless you are using something like -O3 on one test and something like '-O0, -Og, -O1' on the others. So either the configuration is botched or the result gathering is botched, this is not a correct result, same goes for the VP9 encoding benchmarks, and I also find the DAV1D results suspect.

    It's a shame because these tests are very interesting, but when the results are as here, clearly bogus, it makes them all suspect.

    Leave a comment:


  • halo9en
    replied
    Originally posted by lumks View Post
    I also would like to have Arch in there. I mean common, it's a one time setup you have to do. And no, Manjaro is not Arch, also not in performance of some things.
    Or at least Arcolinux if setup time is an issue, at least it's current Arch unlike Manjaro.

    Leave a comment:


  • edwaleni
    replied
    Originally posted by Grinch View Post
    The x265 benchmarks are clearly botched, there's no way that Clear Linux is ~40% faster than the rest, most of the hot code in x265 is hand optimized assembly. Are the other builds not using any optimization settings at all and thus defaulting to -O0 ? And of course the same goes for VP9 encoding test, and possibly the DAV1D test as well.

    Please fix these, they reflect poorly on Phoronix benchmarking.
    It was about a year ago I made on comment on Phoronix that Clear had a problem with some of the encode/decode benchmarks. They were "middle of the road" results and this was surprising compared to its other performance optimizations. Arjan from Intel responded that he agreed. So its very possible (I haven't looked under the hood) that Intel did do some optimizations. I can't verify it, but I have seen responsiveness from the Clear team on github and other places.

    Leave a comment:


  • Grinch
    replied
    The x265 benchmarks are clearly botched, there's no way that Clear Linux is ~40% faster than the rest, most of the hot code in x265 is hand optimized assembly. Are the other builds not using any optimization settings at all and thus defaulting to -O0 ? And of course the same goes for VP9 encoding test, and possibly the DAV1D test as well.

    Please fix these, they reflect poorly on Phoronix benchmarking.

    Leave a comment:


  • birdie
    replied
    Originally posted by Charlie68 View Post

    I'm not so sure that the mitigation settings are so similar, there are also other security tools like selinux vs apparmor are not similar. The default security settings vary from distribution to distribution and certainly have an impact on performance. Personally I have tried many distributions and I have never noticed great differences, especially for a desktop user, for this reason I don't give much importance to the benchmarks, I wouldn't like that even the distributions began to think more about performance than about security, they would end up like Intel.
    SeLinux/AppArmore barely have any impact on most benchmarks.

    Leave a comment:


  • Charlie68
    replied
    Originally posted by birdie View Post
    I was talking about GCC compilation flags, you countered with ... CPU HW vulnerabilities mitigations which are enabled on all Linux distros by default and all have roughly the same performance impact - actually recent kernels are less affected because some of these mitigations have been optimized unlike RHEL where they use the initial "slow" versions of them.
    I'm not so sure that the mitigation settings are so similar, there are also other security tools like selinux vs apparmor are not similar. The default security settings vary from distribution to distribution and certainly have an impact on performance. Personally I have tried many distributions and I have never noticed great differences, especially for a desktop user, for this reason I don't give much importance to the benchmarks, I wouldn't like that even the distributions began to think more about performance than about security, they would end up like Intel.

    Leave a comment:


  • 9Strike
    replied
    Nice to see Debian on the 2nd-ish place (Clear Linux doesn't really count in my eyes and CentOS Stream and CentOS 8 are basically the same), Buster was a really good release. Only the boot time is a mess compared the other Distros.

    Leave a comment:


  • vegabook
    replied
    Originally posted by birdie View Post
    Too bad these results won't result in anything on Fedora's side for "the sake of security of our users". Yet, most Fedora users are on the desktop where these security measures are an absolute overkill.
    The problem is Fedora is the testbed for an enterprise product. And that will only accelerate under IBM. And canonical's pivot toward enterprise means this trend is likely to pickup even more. We'll soon all be running massively bloated "desktop" editions of code that's meant for hyperscalers.

    Leave a comment:

Working...
X