Announcement

Collapse
No announcement yet.

AMD FX-8150 Bulldozer On Ubuntu Linux

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

  • AMD FX-8150 Bulldozer On Ubuntu Linux

    Phoronix: AMD FX-8150 Bulldozer On Ubuntu Linux

    Two weeks ago AMD introduced the Bulldozer FX-Series CPUs to much excitement, although many were letdown by the initial results, and it was months after showing the first Linux benchmarks of an AMD Dual-Interlagos pre-production system. In the days that followed I delivered some initial AMD FX-4100 Linux benchmarks when securing remote access to a low-end Bulldozer system running Ubuntu 11.04 (and there were also some Linux benchmarks from independent Phoronix readers), but then last week a Bulldozer kit arrived from AMD. The centerpiece of this kit is an eight-core AMD FX-8150 CPU, which is now being used to conduct a plethora of AMD Bulldozer benchmarks on Linux.

    http://www.phoronix.com/vr.php?view=16572

  • #2
    Looks like I'll have to run some comparisons with the 1090T.

    PS. You really have to get those system configuration tables to show up right on your articles. Right now it doesn't even show the system being benched.

    Comment


    • #3
      Lol

      Is it just me, or does the CPU name state: "AMD FX(tm)-8150 Eight-Care Processor"

      While the part that says: "cpu cores: 4"

      Even the OS sees it a quad core? A quad core with increased integer performance.

      Comment


      • #4
        Surprising

        I'd say these are surprising results for Bulldozer on linux. Almost all the Windows reviewers found it to barely be competitive with the i5 2400 or 2500 depending on the test. Even with the expected 10% improvement from Windows 8, it wouldn't win at all on a Windows platform.

        But I guess Linux's better threading is winning here. I'm surprised by the encoding tests where it showed to be competitive.

        It seems these chips may actually be great server parts.

        Comment


        • #5
          Originally posted by zeealpal View Post
          Is it just me, or does the CPU name state: "AMD FX(tm)-8150 Eight-Care Processor"

          While the part that says: "cpu cores: 4"

          Even the OS sees it a quad core? A quad core with increased integer performance.
          Something's mixed up here, or we're missing some kind of context, because it also says "siblings: 8" and "core id: 7".

          Comment


          • #6
            Originally posted by zeealpal View Post
            Is it just me, or does the CPU name state: "AMD FX(tm)-8150 Eight-Care Processor"

            While the part that says: "cpu cores: 4"

            Even the OS sees it a quad core? A quad core with increased integer performance.
            If it had 8 cores, then Sun Microsystems's SPARC T1 would have 32 cores. Only an ignorant person would think that way though.

            Originally posted by hechacker1 View Post
            I'd say these are surprising results for Bulldozer on linux. Almost all the Windows reviewers found it to barely be competitive with the i5 2400 or 2500 depending on the test. Even with the expected 10% improvement from Windows 8, it wouldn't win at all on a Windows platform.

            But I guess Linux's better threading is winning here. I'm surprised by the encoding tests where it showed to be competitive.

            It seems these chips may actually be great server parts.
            The comparison was biased. The Intel Sandy Bridge processors tested do not support SMT:

            http://ark.intel.com/products/52210

            You could say that you compared things in the same price range, but those Gulftown processors are far more expensive than Sandy Bridge processors with SMT support.

            Originally posted by Ex-Cyber View Post
            Something's mixed up here, or we're missing some kind of context, because it also says "siblings: 8" and "core id: 7".
            It does the same thing with Intel's SMT (i.e. HyperThreading). Those figures refer to virtual cores.
            Last edited by Shining Arcanine; 10-24-2011, 02:51 AM.

            Comment


            • #7
              Great and i was almost disappointed, i just hate intel and am glad that when i buy a new processor i wont have to make a really bad choice just because i don't want to support a company i hate.

              anyway i think it would be interesting to see how well llvm-pipe works on the bulldozer, compared maybe to intel processors.

              Comment


              • #8
                I realized that the benchmarks were biased a little late, so I tried editing my post and I made a mistake in it.

                Anyway, "You could say that you compared things in the same price range, but those Gulftown processors are far more expensive than Sandy Bridge processors with SMT support." should have been:

                "You could say that Michael compared things in the same price range, but those Gulftown processors are far more expensive than Sandy Bridge processors with SMT support."

                Additionally, the Core i7-2600 costs $20 more than Bulldozer:

                http://www.newegg.com/Product/Produc...82E16819103960
                http://www.newegg.com/Product/Produc...82E16819115071

                In comparison, the Core i5 2500k that was tested costs $50 less.

                http://www.newegg.com/Product/Produc...82E16819115072

                Furthermore, the Core i7 2600k costs only $35 more:

                http://www.newegg.com/Product/Produc...82E16819115070

                Had Bulldozer been around in 2008, it would have been very competitive, but right now, there is much better hardware that you can buy.

                Comment


                • #9
                  @Shining Arcanine: You mentioned a compiler benchmark, compiling LibreOffice.
                  I second that! That's really what I'm missing, and was one of the things I would have known, when I had to chose the base for my new PC.
                  It got a i7 2600K...
                  And now I don't know if libreoffice is the best package to test compiler performance, as libreoffice does a lot more then just throwing files into gcc. There is a lot self-baked file processing and java involved, so no benchmark for pure gcc-performance.
                  That lead to those numbers:
                  libreoffice:
                  Athlon II X3 435@ 2,9 GHz (stock): 1h 33min
                  i7 2600K: 46min
                  glibc:
                  Athlon: 21min 15sec
                  2600k: 7min 48sec

                  So sb is 3x faster at glibc but only 2x faster on libreoffice.
                  When comparing other packages (e.g. kdelibs, qt, glib, gtkmm, ...) numbers are always between 2x and 3x faster, with a great mayoritiy going towards 3x faster.
                  The guys over at anandtech have chosen chromium as Visual Studio-benchmark, look at e.g. the sandy bridge review. Probably that also is a good package, to bench gcc?

                  Comment


                  • #10
                    Michael, did you compile these tests with -march=bdver1 for Nulldozer, and whatever option is for SB?

                    Comment


                    • #11
                      ekopath

                      With all that Ekopath buzz when it was open sourced why not try and see how that works ?
                      Also it would be nice to see how long it takes to compile using Bulldozer ( timed Apache compile ? )

                      Comment


                      • #12
                        Originally posted by hechacker1 View Post
                        I'd say these are surprising results for Bulldozer on linux. Almost all the Windows reviewers found it to barely be competitive with the i5 2400 or 2500 depending on the test. Even with the expected 10% improvement from Windows 8, it wouldn't win at all on a Windows platform.

                        But I guess Linux's better threading is winning here. I'm surprised by the encoding tests where it showed to be competitive.

                        It seems these chips may actually be great server parts.
                        Nah, this was pretty comparable to the windows results I saw. It slightly beats out the 2500k on most highly threaded tests, but loses by a fair margin to the i7 2600k (same chip but with hyperthreading enabled, and more expensive of course). It's the lightly threaded tests where it is only at 2400 levels.

                        A lot of it just has to do with the fact that most windows tests aren't that highly multi-threaded. Michael runs a bunch of parallel tests, which make the chip look better compared to the windows reviews.

                        I do think this could be a decent server chip, but we'll have to wait and see.

                        Comment


                        • #13
                          Originally posted by Drago View Post
                          Michael, did you compile these tests with -march=bdver1 for Nulldozer, and whatever option is for SB?
                          or just everything with -march=native

                          Comment


                          • #14
                            Among the other tests that are still being carried out at Phoronix with the FX-8150, which will be published in the coming days, include: [...]
                            Looking forward to that, and I applaud the effort.
                            Edit: seems like the openbenchmarking.org results are already out, correct?
                            Edit 2: what where the memory settings on each config?

                            Originally posted by Shining Arcanine View Post
                            The comparison was biased.
                            I think "incomplete" is a more adequate statement. The tests were run on what hardware was available.
                            Also I challenge your price comparisons. First, one might as well include the more affordable FX-8120 (3,1 GHz stock instead of 3,6 GHz, so a slight overclock and bam...). Second, one has to consider the platform as a whole. Third, for current 890/990-chipset mobo owners, I doubt the price difference of "only" $35 holds much weight.

                            Originally posted by smitty3268 View Post
                            It slightly beats out the 2500k on most highly threaded tests
                            So I guess we have a different definition of "slightly".

                            Nice review in any case.
                            Last edited by PsynoKhi0; 10-24-2011, 07:41 AM.

                            Comment


                            • #15
                              Oh noe! Water c00l1ng :/ I hate watercooling en masse. :/

                              Its great stuff all and all, if you aim for overclocking straight, forgetting purchase costs (but then, real overclockers would use custom llang ddc-based watercooling with 3x120 or 6x6x120 anyway!), or even head for refrigerators/liquid nitrogen.
                              But 1x120 "closed system" is just laughable in terms of "advantage", packaging it in series is laughable 2x.

                              Because this basically means
                              1) (for AMD) you cannot solve your thermal problems (1x120 is nothing for overclocking)
                              2) (for buyer) you are a ricer with no understanding

                              Btw, awesome review, Michael! Much thanks!
                              Kudos to AMD for sending sample.

                              But a small test regarding actual power consumption of systems (provided they use same gfx card) in
                              1) idle
                              2) full load CPU
                              3) full load GPU
                              states would awesome to add up.

                              Thanks again!
                              Last edited by crazycheese; 10-24-2011, 07:39 AM.

                              Comment

                              Working...
                              X