Announcement

Collapse
No announcement yet.

Quad-Core ODROID-X Tested Against PlayStation 3

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

  • Quad-Core ODROID-X Tested Against PlayStation 3

    Phoronix: Quad-Core ODROID-X Tested Against PlayStation 3

    Here are some more benchmarks of the ODROID-X, a $129 ARMv7 development board that packs four Cortex-A9 cores along with Mali-400 graphics to provide a fairly impressive punch. There's even some comparative numbers to a Sony PlayStation 3 running Linux...

    http://www.phoronix.com/vr.php?view=MTE2NzI

  • #2
    Wasn't the cell processor in the PS3 supposed to be really fast or something?

    How on earth an ARM chip is faster.

    Comment


    • #3
      Likely the cell vector units are not being used, so it is only a comparison to a single-core PowerPC @ 3.2Ghz? Uns

      Comment


      • #4
        Originally posted by 89c51 View Post
        Wasn't the cell processor in the PS3 supposed to be really fast or something?

        How on earth an ARM chip is faster.
        The Cell CPU is only superb if you can more or less utilize all its 8 "co-processor" units (SPEs). Linux can't, but it is no surprise (not Linux's fault) since these are hardly general purpose CPUs in themselves, for example they do not have linear access to system memory. So when you run Linux on a PS3, you are basically running it on a single, outdated 3.2Ghz PowerPC core (the PPE in the Cell), and contrasting that to a modern 4-core 1.4Ghz CPU is less than fair.

        I've had the joy of programming on the Cell. The problem is that using its 8 SPE's doesn't just mean writing parallel code. Since they do not have access to main memory, keeping their tiny 256KB local storage constantly filled with overlapped DMA transfers, and making efficient use of their dual-instruction-path CPU engine which is only possible with manual ASM coding is anything but fun. In consequence, the Cell CPU is mainly useless for almost anything except for multimedia processing, which is why it makes sense to put it in a console meant for games. Otherwise, its PowerPC core is slow and coupled with the PS3's limited 256MB of RAM it is absolutely no wonder that it is outperformed by any current CPU when running an OS on it. However, the Cell could still outperform the ODROID-X in multimedia if it was running something optimized for the Cell, not just something compiled for the PowerPC.

        Michael sometimes makes news out of nothing extraordinary, this is one of those cases. Or I could say the news is not that the ODROID-X runs Linux faster than the PS3, but that it is a feature-rich and cheap development board.

        Comment


        • #5
          Actually if anyone paid attention, it says the PS3 is a dual core. I wonder if the affinity was set like that deliberately so it would be a better comparison to the dual core ARM chip?


          Either way, one thing I found interesting is anything media related, PS3 shot up pretty quick. I'm not sure if it's GPU has drivers yet but it seems to me that a PS3 would still be better than most other systems to work as an HTPC.

          Comment


          • #6
            vS. quad core Krait

            i am waiting for the PS3 vS. quad core Krait benchmark.

            Comment


            • #7
              Originally posted by schmidtbag View Post
              Actually if anyone paid attention, it says the PS3 is a dual core.
              LOL, no it isn't. It is a dual-threaded processor, or better said, a processor with a dual-issue pipeline. That is, it has two instruction pipelines, but instructions are executed in every alternate cycle. So if you insist on viewing it as a dual-core CPU, then it is only 1.6Ghz instead of 3.2Ghz. Except worse then a dual core, since many things are not replicated, for example both pipelines share caches (even the L1 cache). People are often mislead into thinking that it is a dual-core CPU, because the OS sees two logical CPUs, the same way it would show two CPUs for an Intel hyperthreaded core. But they are only fake/logical cores.

              Originally posted by schmidtbag View Post
              Either way, one thing I found interesting is anything media related, PS3 shot up pretty quick. I'm not sure if it's GPU has drivers yet but it seems to me that a PS3 would still be better than most other systems to work as an HTPC.
              As I said, the Cell would win with multimedia, but only if you can get its 8 SPUs to work. In practice though it wouldn't be a good HTPC, because there is no publicly available multimedia player/decoder that would use them. All Linux players would just use the single PPC core.

              Comment


              • #8
                Here's one more interesting fact for the interested. The SPUs also have a "dual-issue" pipeline, but not just any, a very peculiar one at that. It can theoretically execute two instructions at the same time, BUT each instruction pipeline can only execute specific instructions - otherwise it stalls. To be precise, one can execute only data (load/store..), the other only arithmetic instructions. So to get to the theoretical maximum 100 Gflops as advertised in marketing materials, you'd need to perfectly alternate between data and arithmetic instructions. Good luck with that! The compiler sucks at it, by manual ASM-coding you can get better results but still far from anything perfect, as very often it is just not possible. And of course it has no out-of-order execution, so if it's not the compiler nor you doing it, you get half of the advertised performance.

                Given realistic programs, the Cell's speed is easily matched by dekstop CPUs from even 2-4 years ago, and they are waaaay-much easier to develop for. Though don't get me wrong, the Cell does shine at least from one aspect, which is power/performance-ratio. The Cell offers a relatively competitive performance for much less power consumption (ok, I don't have exact numbers for consunption right now), so the Cell is very attractive, for example, in supercomputers, where you jam thousands of CPUs into a single machine. And no surprise here, the Cell is used in multiple supercomputers. But believe me, avoid it for the desktop.

                Comment


                • #9
                  Originally posted by ultimA View Post
                  LOL, no it isn't. It is a dual-threaded processor, or better said, a processor with a dual-issue pipeline. That is, it has two instruction pipelines, but instructions are executed in every alternate cycle. So if you insist on viewing it as a dual-core CPU, then it is only 1.6Ghz instead of 3.2Ghz. Except worse then a dual core, since many things are not replicated, for example both pipelines share caches (even the L1 cache). People are often mislead into thinking that it is a dual-core CPU, because the OS sees two logical CPUs, the same way it would show two CPUs for an Intel hyperthreaded core. But they are only fake/logical cores.


                  As I said, the Cell would win with multimedia, but only if you can get its 8 SPUs to work. In practice though it wouldn't be a good HTPC, because there is no publicly available multimedia player/decoder that would use them. All Linux players would just use the single PPC core.
                  I didn't say that I insist it's a dual core, I'm stating that's what the benchmark posted it as - way to be an ass about it. After some research, I'm not finding anything stating that PS3's processor is dual-threaded from several different sources. While I don't fully understand the idea of SPEs, they seem to be actual physical cores, and there's 8 of them. I have always been aware that one of them was usually disabled, with another one being used for the OS. Look at pictures of the design and you can distinctively see them, however, since I don't understand it perfectly, you could compare it to the Bulldozer architecture where you can see 8 distinct separate cores but there's technically only 4.

                  Either way, I have yet to find 2 of anything (by default) in the cell processor, which led me to believe that whoever performed these benchmarks intentionally disabled all other SPEs (assuming they're recognized as CPU cores).

                  Comment


                  • #10
                    The information in the thread matches what I know of the PS3. It is basically a 3.2Ghz PPC that performs on-par with the old G5 iMac from 2006. The video card was supposedly a NV47 variant. It has 8 SPEs, of which I believe 7 are accessible from linux. Nothing automatically uses these SPEs though, as it would require specifically crafted binaries. It was an incredibly powerful chip 'if' you were able take full advantage of the hardware, which was said to be difficult in most circumstances.

                    Cell was a neat technology when it was released. Unfortunately, it looks as if IBM didn't/couldn't follow it up with a second version that fixed the problems that plagued the original.

                    I'm not surprised by the PS3 benchmark results, but understand that they are not truly representative of the Cell's full breadth of capabilities.

                    F

                    Comment


                    • #11
                      Originally posted by schmidtbag View Post
                      didn't say that I insist it's a dual core, I'm stating that's what the benchmark posted it as - way to be an ass about it.
                      I'm sorry, I didn't mean it that way. What I meant is, "if you'd prefer to look at it as a dual-core CPU", but I'm not implying that you actually prefer it.

                      Originally posted by schmidtbag View Post
                      After some research, I'm not finding anything stating that PS3's processor is dual-threaded from several different sources.
                      I wrote those things from my memory, since I haven't worked with the Cell in a long time, but a search on Google turned up this (link). "The PPU is a dual-issue, in-order processor with dual-thread support."

                      Originally posted by schmidtbag View Post
                      While I don't fully understand the idea of SPEs, they seem to be actual physical cores, and there's 8 of them.
                      I think you misunderstood me. Yes, there are 8 SPEs and they are physical. What I said about not being physical cores was regarding the PPE, not for the SPEs. There is 1 PPE, which is a derivateve of a PowerPC core, and there are 8 SPEs, which are meant for stream processing. Linux applications by default, when not optimized explicitly for the Cell, only make use of the PPE. The SPEs can also be used under Linux, but not by Linux, and they need special programming/support. What I'm pretty sure of is that none of the phoronix tests used the SPEs. The SPEs weren't disabled explicitly, they just weren't used, at least I guess, because I am not familiar with the test suit. It's like running 16bit applications on a 64bit CPU and then complaining that the CPU has bad performance. The comparison simply wasn't fair/done correctly, because if my guess is right, the tests don't even try to utilize the full potentional of the Cell. Maybe someone could confirm that?
                      Last edited by ultimA; 08-24-2012, 01:35 PM.

                      Comment


                      • #12
                        Originally posted by russofris View Post
                        Cell was a neat technology when it was released. Unfortunately, it looks as if IBM didn't/couldn't follow it up with a second version that fixed the problems that plagued the original.
                        There was actually a Cell 2.0, which fixed the terrible "double-precision floating-point performance bug" the original Cell had. I remember it quite vividly, since that release of the Cell 2.0 was what rendered my work with the Cell up to that point useless

                        Comment


                        • #13
                          Comparing the ARM system to the PS3 is absolute nonsense. Not only is the PS3 running a linux distro from THREE YEARS ago versus the ARM system on the latest kernel/gcc, it has to do everything through a hypervisor. A huge downside to that is that the GPU is inaccessible. So you have no GPU and a hypervisor that helps throw some performance out the window. If you could ACTUALLY install linux on a PS3 (on bare metal), its numbers would be much higher. Would the ARM system still win? No idea. But this setup is far from scientific; it was probably just done for fun.

                          Comment


                          • #14
                            As for why the PS3 wins in some tests, it most likely has to do with the fact that it has a huge advantage in memory (RAM) bandwidth. Its RAM runs at 3.2 GHz!

                            Has anyone managed to fully hack a PS3 to the point where the hypervisor could be bypassed entirely?

                            Comment


                            • #15
                              Clarification

                              As the submitter of these results I would like to clarify a few things regarding the PlayStation3 and these tests.

                              The Playstation3 has a general purpose PowerPC core capable of running three threads. In addition, the system has six other threads called SPEs which ARE accessible in GNU/Linux through the CELL SDK and given that you write specific code for them. If the software is properly implemented 160 GFLOPS of FP performance are possible.

                              The tests that were run in PTS employed the PPE core only (one PowerPC core capable of two threads).

                              Comment

                              Working...
                              X