Announcement

Collapse
No announcement yet.

AMD Releases FX-Series Bulldozer Desktop CPUs

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

  • #21
    Originally posted by deanjo View Post
    This is true to an extent. However other introduced processor architectures (i3/i5/i7, K7, etc) have shown great leaps in performance even without rebuilding of binaries. When the binaries were redone that leap in performance just got greater.
    i3/i5/i7 are all the same architecture with different performance grades.
    Intel has also been known in the past for building-to-benchmarks. AMD has a history of ignoring the benchmarks and just building the best chip they can.

    K7 is a different kind of case, from a different era. It wasn't just an architectural change, it was also just plain MUCH MUCH FASTER. You'll recall that it had a DDR FSB. The benchmarks at that time certainly would have been made for K6/P2, but even with that against it, it was ***SO MUCH FASTER*** that it didn't matter.

    Comment


    • #22
      What people really should be asking is if it will improve the performance of your particular application that you spend most of your cpu cycles on.

      Comment


      • #23
        Originally posted by crispy View Post
        What people really should be asking is if it will improve the performance of your particular application that you spend most of your cpu cycles on.
        Well again, you can't tell right now due to the different architecture. We already know that the current process schedulers will kill the performance of these chips. In the *very least*, the kernel must be updated to support this.

        Comment


        • #24
          @deanjo

          that means you will switch to ivi bridge next year or what

          Comment


          • #25
            Take these tests by X-bit labs, Anandtech, etc with a big grain of salt.

            Most of these tools used for testing (Sisandra, various games, etc) are compiled using Intels C/C++ compiler which generates fast and optimized codepath's for Intels own processors but very bad and inefficient codepaths for AMD processors. Very unfair to AMD and Bulldozer.

            Ofcouse these tests will show Intel as a big leader over AMD as the Intel code runs optimized and AMD does not.

            Do the tests using AMD's own Open64 C/C++ compiler and you will get a different result.

            Comment


            • #26
              Originally posted by droidhacker View Post
              i3/i5/i7 are all the same architecture with different performance grades.
              Intel has also been known in the past for building-to-benchmarks. AMD has a history of ignoring the benchmarks and just building the best chip they can.

              K7 is a different kind of case, from a different era. It wasn't just an architectural change, it was also just plain MUCH MUCH FASTER. You'll recall that it had a DDR FSB. The benchmarks at that time certainly would have been made for K6/P2, but even with that against it, it was ***SO MUCH FASTER*** that it didn't matter.
              With the craptastic performance the cache seems to have i highly doubt optimizations are gonna get you that much. For BD to even be a decent arch it needs to improve by at least 20%, which is not going to happen. I mean an 8 core 32nm "Llano core(Is Stars the codename?)" CPU would have performed significantly better while probably also consuming less power. The only explanation i can think of is that there is some bottleneck that didn't show up in the simulations and caused AMD to miss their performance target by a large degree. Otherwise i really can't understand why they didn't simply cancel the project. BTW a while back the guy that came up with MCM idea had commented that the design didn't really come out that great.

              Comment


              • #27
                Originally posted by droidhacker View Post
                You don't necessarily need to optimize *all* your binaries. Probably the kernel by itself will make a big difference.
                That's not going to save BD unless the kernel does something very stupid when it encounters an unknown AMD cpu.
                BD is slower then an equally clocked Phenom II when it comes to single-threaded applications as well as most multithreaded applications.
                That is by design, as BD was designed to have lower IPC then previous generations but have higher clock rate to compensate.
                Unfortunately the reality is that the transistor techs and materials used today doesn't allow BD to overclock that much further then Phenom II.
                [Edit:]Sweclockers.com reached 4.7-4.8 GHz with watercooling.Google translated to english/Swedish original text

                I bet a 32nm Phenom II X6 would reach just as high clock rates but would outperform BD apart from some specialized application.
                If you're a full-time user of those applications, then BD will be better then a Phenom II but shouldn't you be looking in the server segments of the CPU market instead?
                Compile-time or realtime optimizations of the workload sounds good on paper, we heard it when Intel talked about P4, but unless you have a very
                static workload it takes a lot of work to optimize all the workload.

                Sorry, I didn't mean to sound so harsh and negative but I didn't expect my Phenom II X6 to be just as good or better then BD.
                Last edited by a7v-user; 12 October 2011, 11:43 AM. Reason: Added overclocking result

                Comment


                • #28
                  Most of you are missing the point of this processor. Look at the die size compared to performance (just ignore per core/process performance for the moment). Bulldozer actually beats current Sandy Bridges by a good margin here, and this is a very important metric in the server market. The only downside is that the Interlagos chips must be pulling more power than their Xeon competitors.

                  The Bulldozer looks pretty lackluster as a desktop part, but it will probably get AMD some big wins in the server market which, along with the no-brainer Llano wins for the mobile segment, will keep them humming along in their usual role as Intel's red-headed stepchild.

                  Overall, PC enthusiasts represent a pretty small market. If they didn't, AMD would have gained a lot more market share than they did in the Athlon64/P4 days. It sucks for us because it will keep retail prices for the good stuff from Intel higher, and their low-end chips artificially crippled.
                  Last edited by psycho_driver; 12 October 2011, 11:36 AM.

                  Comment


                  • #29
                    Originally posted by mirv View Post
                    Bulldozer was designed with heavily threaded/multi-process environments (aka server systems) in mind. Many desktop applications are headed that way too, but they're not there yet.
                    We've been hearing this since the Pentium 4 got Hyperthreading in 2002. That's a decade of "everything will be written with parallelism in mind". Most applications do not need to be massively parallel, nor deal with the complexities that come with it. Fast single threaded performance will remain one of the most important aspects for years to come on the desktop.

                    Comment


                    • #30
                      Originally posted by skies View Post
                      Take these tests by X-bit labs, Anandtech, etc with a big grain of salt.

                      Most of these tools used for testing (Sisandra, various games, etc) are compiled using Intels C/C++ compiler which generates fast and optimized codepath's for Intels own processors but very bad and inefficient codepaths for AMD processors. Very unfair to AMD and Bulldozer.

                      Ofcouse these tests will show Intel as a big leader over AMD as the Intel code runs optimized and AMD does not.

                      Do the tests using AMD's own Open64 C/C++ compiler and you will get a different result.
                      But this don't really explains why BD single core performance is much worse than Stars single core performance.

                      Comment

                      Working...
                      X