Announcement

Collapse
No announcement yet.

Four-Way ARM Linux Distribution Comparison

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

  • Four-Way ARM Linux Distribution Comparison

    Phoronix: Four-Way ARM Linux Distribution Comparison

    The latest ARM Linux benchmarks to share at Phoronix is a comparison of Ubuntu 12.10, Linaro 12.10, Fedora 17, and Arch Linux when running from the dual-core Cortex-A9 OMAP4460-based PandaBoard ES development board.

    http://www.phoronix.com/vr.php?view=18113

  • #2
    I'm a little surprised at these results. I didn't expect arch to fall that far behind, although i have got the impression their ARM support is a little lacking. One of the things i found so weird about arch is its relatively poor support for anything that isn't x86, because you'd think a distro that sometimes compiles things on-the-fly would be easily cross platform.

    Comment


    • #3
      Originally posted by schmidtbag View Post
      I'm a little surprised at these results. I didn't expect arch to fall that far behind, although i have got the impression their ARM support is a little lacking. One of the things i found so weird about arch is its relatively poor support for anything that isn't x86, because you'd think a distro that sometimes compiles things on-the-fly would be easily cross platform.
      Well, Arch is no Gentoo.

      Comment


      • #4
        Disappointing that the recently released openSUSE 12.2 for ARM wasn't included in the mix but I imagine Michael was way to deep into these tests when it came out to include them.

        Comment


        • #5
          Originally posted by deanjo View Post
          Disappointing that the recently released openSUSE 12.2 for ARM wasn't included in the mix but I imagine Michael was way to deep into these tests when it came out to include them.
          Exactly what I was about to say.

          It would be also interesting to see results for some mid or low end amd64 system for comparison as to how much slower arm is.

          Comment


          • #6
            Unity

            Why are all distributions being tested from the console while Linaro is being teted with Unity2D?
            This seams to be a handicap for Linaro (which otherwise does quite well, together with Ubuntu)

            Comment


            • #7
              Debian?

              Oy, where's Debian? It's probably one of the oldest ARM distributions out there, how could you forget it?

              --Coder

              Comment


              • #8
                Further benchmarking

                Thanks for your work, I read you every day.

                But I would like to know this benchmarks:

                Multiple pandaboards as server -

                How well they scale?

                How much are needed to equal an i7 more or less?

                How much would cost this multi Panda board server vs a single i7? cost/performance

                And how many energy this multi PAnda spend vs a single i7 energy efficiency

                Also how well i7 scales vs ARM, then 4 i7s vs 4 multi Panda Borads alike systems, with the intermediate steps.

                I know it would be quite a work but nobody is actually doing this benchmarks, and surely it would be a great benchmark to read if you are thinking in making a server or a small suoercomputer.

                I recently watched a video where a university makes its own supercomputer and they waited to the new i7 to make it instead of using cheaper ARM boards, perhaps with future ARM64 boards the choose would have been different, but as a tech reader I cannot find now benchmarks between multi intel i7 / AMDs and ARM for servers and supercomputers with actual costs and energy budget for a similar power.

                Thanks in advance for your future work in this field if you accept the suggestion.

                Comment


                • #9
                  Originally posted by coder111 View Post
                  Oy, where's Debian? It's probably one of the oldest ARM distributions out there, how could you forget it?

                  --Coder
                  Perhaps because it may have been faster than Obonto?

                  Comment


                  • #10
                    Originally posted by Cyber Killer View Post
                    Exactly what I was about to say.

                    It would be also interesting to see results for some mid or low end amd64 system for comparison as to how much slower arm is.
                    H.264 Video Encoding
                    1.4 GHz ARM9 Calxeda Quad Core - 11.82 fps
                    3.2 GHz Core i5 3470 - 107.36 fps
                    9 times slower
                    4 times slower at same frequency

                    7-Zip Compression
                    1.4 GHz ARM9 - 2176 MIPS
                    3.2 GHz Core i5 - 13778 MIPS
                    6.3 times slower
                    2.8 times slower at same frequency

                    Source:
                    http://www.phoronix.com/scan.php?pag...000_atom&num=3
                    http://www.phoronix.com/scan.php?pag...ei5_3470&num=5

                    Comment


                    • #11
                      @JS987: Thanks

                      Comment


                      • #12
                        Originally posted by mitcoes View Post
                        Multiple pandaboards as server -
                        How well they scale?
                        Michael already did this:
                        http://www.phoronix.com/scan.php?pag..._cluster&num=1

                        How much are needed to equal an i7 more or less?
                        How much would cost this multi Panda board server vs a single i7? cost/performance
                        And how many energy this multi PAnda spend vs a single i7 energy efficiency
                        Also how well i7 scales vs ARM, then 4 i7s vs 4 multi Panda Borads alike systems, with the intermediate steps.
                        http://www.phoronix.com/scan.php?pag...cluster&num=16
                        When comparing to an i7, ARM clusters seem to be much more expensive, a bit less power efficient, and not much faster, if at all. I think you're underestimating the power of Sandy Bridge and Ivy Bridge i7s. Even AMD systems are likely more power, time, and cost efficient. Keep in mind that whenever you do cluster computing, for every computer you add you're wasting electricity on powering a whole separate system, you drastically lower latency (by communication, error checking, translation between gathering the data and interpreting it for use, timing everything properly, the relatively lengthy wires, etc), you spend more time setting it up, and the more systems you have there's a higher probability of failure. In other words, if you're looking to massively parallel something without going beyond 64 cores, you're wasting time, energy, effort, and money on a cluster. You can get up to 64 threads in 1 system with a SPARC T3 setup. Also, there are server motherboards that allow 4 AMD CPUs at a time. I'm not sure if they support socket G34 (I think that's the socket name) but if they do you can have up to 48 cores using 4 of AMD's 12-core Opterons. Those Opterons are overall a much much better value than the Intel competitor.
                        Last edited by schmidtbag; 11-09-2012, 10:34 AM.

                        Comment


                        • #13
                          There are some 16-core opterons, so you can get 64 x86 cores in a box too.

                          Comment


                          • #14
                            Originally posted by curaga View Post
                            There are some 16-core opterons, so you can get 64 x86 cores in a box too.
                            I don't think those are available yet, and they're actually 8-module systems, so 16 threads. The 12-core Opterons are actually 12 full cores, so they're probably a little more reliable.

                            Comment


                            • #15
                              Thread hijack: Bulldozer module edition

                              One BD module = two int cores, one shared fp core. So only for floats you can say it's not two cores - for integer math it is two cores.

                              Comment

                              Working...
                              X