Announcement

Collapse
No announcement yet.

ARM Cortex-A9 PandaBoard ES Benchmarks

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

  • #31
    I should add, it's very likely TI subsidizes that, so the real price would be more than 189$. They are known to have done so in the past with their dev boards.

    Comment


    • #32
      Originally posted by curaga View Post
      Thanks Michael, it's great to have some good ARM numbers.



      Not in perf/$. For 189$ you can get an E-450.
      I agree. It's not competitive not from a price/performance nor from a price/features point of view. The Raspberry Pi (US$25 to $35) looks much more interesting especially if they can get it to run XBMC.

      Comment


      • #33
        diff --git a/arch/arm/mach-omap2/id.c b/arch/arm/mach-omap2/id.c
        --- a/arch/arm/mach-omap2/id.c
        +++ b/arch/arm/mach-omap2/id.c
        @@ -231,7 +231,7 @@ static void __init omap4_check_features(void)
        case 1:
        default:
        /* Standard device */
        - omap_features |= OMAP4_HAS_MPU_1_2GHZ;
        + omap_features |= OMAP4_HAS_MPU_1_5GHZ;
        break;
        }
        }

        The above is a patch that will run the 4460 at 1.5ghz, replacing the 1.2ghz state. Pretty easy to see what I did is just fudge a board id check.. I put a GPU memory heatsink on the thing just for safety... I tried to drop the mv down to 1350 instead of 1375... but did it quick, and probably missed quite a bit of loose ends.. and thus that was fail..

        Thermal throttling is pretty annoying when compiling a kernel.. ./kern/drivers/staging/thermal_framework/governor/omap_die_governor.c has the goods for the whens and whats of that.. The actual temp sensor (sysfs) readout is quite different from what the thermal throttling setup repots as values (in kmsg when certain temp reached).. so... thats a bit odd.. temp1_input seems to be more realistic.

        Currently, this setup is painful.. Lots of glitchy things happening with X when using the latest pvr crap.. unity is... unity.. but even worse when such tomfoolery is going on.. For anyone who is setting their PB-ES up now, save yourself the time and pain -- and move the rootfs off of the sd card and onto nfs, usb-stick, usbhdd.. whatever.. just not the SD.. I don't care if its class 20000x mega-super.. it sucks.. morse code has more iops..

        Comment


        • #34
          Hardfloat?

          Michael,

          Could you please also run a comparison of an ARM softfloat vs. hardfloat userlands? The standard softfloat solution has a huge performance overhead for EVERY single floating point instruction, which unnecessarily disadvantages the processor in comparison to x86. Also, various FPU options may be of interest (neon, vfp).

          Thanks!

          Comment


          • #35
            erm.. I'm probably doing something wrong/off.. but for comparison's sake..

            7-Zip 9.20 Copyright (c) 1999-2010 Igor Pavlov 2010-11-18
            p7zip Version 9.20 (locale=C.UTF-8,Utf16=on,HugeFiles=on,2 CPUs)

            RAM size: 641 MB, # CPU hardware threads: 2
            RAM usage: 425 MB, # Benchmark threads: 2

            Dict Compressing | Decompressing
            Speed Usage R/U Rating | Speed Usage R/U Rating
            KB/s % MIPS MIPS | KB/s % MIPS MIPS

            22: 1097 139 767 1067 | 25756 199 1166 2325
            23: 1105 154 730 1126 | 25525 200 1170 2337
            24: 1095 146 808 1177 | 25146 199 1171 2333

            and.. i forgot to turn off a swap partition I had enabled on the sd card.. so I ran it again..

            7-Zip 9.20 Copyright (c) 1999-2010 Igor Pavlov 2010-11-18
            p7zip Version 9.20 (locale=C.UTF-8,Utf16=on,HugeFiles=on,2 CPUs)

            RAM size: 641 MB, # CPU hardware threads: 2
            RAM usage: 425 MB, # Benchmark threads: 2

            Dict Compressing | Decompressing
            Speed Usage R/U Rating | Speed Usage R/U Rating
            KB/s % MIPS MIPS | KB/s % MIPS MIPS

            22: 1112 153 708 1081 | 25799 200 1165 2329
            23: 1106 154 732 1127 | 25574 200 1172 2341
            24: 1102 156 757 1185 | 24896 197 1170 2310
            Last edited by Blades; 12-30-2011, 01:28 PM.

            Comment


            • #36
              Hey Michael,

              Love the ARM articles! Would love for you to go more in-depth on the ARM architecture. Some ideas would include:

              - Soft Float vs VFP.
              - NEON vs VFP
              - All of the above vs x86 (Atom)
              - Power consumption of mainboard/cpu only (arm vs atom).

              I know some of the benchmarks may be a bit tricky, especially considering Ubuntu 11.10 lacks good arm support; however, maybe with the latest GCC and kernel, some of these tests may become viable. Knowing what the performance limitations of the ARM core vs Atom and the power consumption of the ARM dev board vs. an Atom (ie N510) would be very interesting!

              Comment


              • #37
                There is something seriously wrong with these arm benchmarks.

                The phoronix got 1154 DMIPS out of their pandaboard with their dhrystone benchmark (DMIPS=dhrystone score/1575) . ARM got 2075 DMIPS out of their 830mhz single core. These benchmarks are also discussed in realworld tech forums, and someone there got 2.5 times phoronix result on the 7zip benchmark on equivalent 1.2Ghz dual core cortex A9. (exynos based). So maybe the operating system didn't get all the drivers right and runs it at fraction of real clockspeed.

                Comment


                • #38
                  Originally posted by josmala View Post
                  The phoronix got 1154 DMIPS out of their pandaboard with their dhrystone benchmark (DMIPS=dhrystone score/1575) . ARM got 2075 DMIPS out of their 830mhz single core. These benchmarks are also discussed in realworld tech forums, and someone there got 2.5 times phoronix result on the 7zip benchmark on equivalent 1.2Ghz dual core cortex A9. (exynos based). So maybe the operating system didn't get all the drivers right and runs it at fraction of real clockspeed.
                  Well, fwiw.. the kernel included in the 11.11 linaro 'distro' had my pandaboard-es running at a smoking 920mhz.. I don't know if this is the case with the latest bundled kernels - as that inspired me to get the source and attain the proper clock speed in any way I could... so, its possible something like that is going on - that and running everything off of the sd card is awful (from a usability standpoint) - I wouldn't be surprised to impact benchmarks in ways you wouldn't think of..

                  Comment


                  • #39
                    Originally posted by gururise View Post
                    - NEON vs VFP
                    NEON isn't a replacement for VFP as it doesn't support IEEE754. In particular no double precisions and just one rouding mode. ARMv8 ISA will have proper IEEE754 SIMD instructions.

                    Comment


                    • #40
                      BTW, here are my benchmark results from gentoo (hardfp) running on origenboard (dual-core ARM Cortex-A9 @1.2GHz): http://openbenchmarking.org/result/1...AR-1112277AR91

                      Comment


                      • #41
                        Wow, nice! Much more in-line with what I was expecting. And a Pandaboard with hardfp should get in the same ballpark as that, right? Suddenly the power question is in much sharper relief.

                        Comment


                        • #42
                          Originally posted by ssvb View Post
                          BTW, here are my benchmark results from gentoo (hardfp) running on origenboard (dual-core ARM Cortex-A9 @1.2GHz): http://openbenchmarking.org/result/1...AR-1112277AR91
                          Thanks for doing that! These results make me think the Pandaboard was misconfigured.

                          Comment


                          • #43
                            How much is from hardfp and how much from a tweaked arm core (exynos)? I recall samsung's pr saying the exynos outruns omap4.

                            Comment


                            • #44
                              Originally posted by Wyatt View Post
                              Wow, nice! Much more in-line with what I was expecting. And a Pandaboard with hardfp should get in the same ballpark as that, right? Suddenly the power question is in much sharper relief.
                              Yes, Pandaboard ES and Origen board should have similar performance.

                              Originally posted by curaga View Post
                              How much is from hardfp and how much from a tweaked arm core (exynos)?
                              Using hardfp API should help on some floating point heavy tests which do lots of function calls. Running on real 1.2GHz clock speed (vs. allegedly 920MHz) should help everywhere. Also if gcc in ubuntu is configured to target thumb2 code generation by default, it also could affect performance depending on the quality of the compiler.

                              I recall samsung's pr saying the exynos outruns omap4.
                              Exynos4210 may have quite a bit faster memory controller, especially if compared to older OMAP4430 which had some serious problems with memory performance. But OMAP4460 was supposed to resolve the issue.

                              Originally posted by ldesnogu View Post
                              Thanks for doing that! These results make me think the Pandaboard was misconfigured.
                              The question is how could it happen after all the supervision that ubuntu got from linaro? Especially considering linaro focus on the kernel and toolchain areas.

                              Also a major problem for ARM when running tests like this is the missing sane support for runtime cpu features detection (-march=native and -mtune=native options support in gcc, reliable neon detection and use in all the neon optimized libraries). ARM Ltd. has been aware of the problem for years, but did nothing to address it

                              Comment


                              • #45
                                Originally posted by ssvb View Post
                                hardfp API
                                This was a typo and should be "hardfp ABI".

                                Comment

                                Working...
                                X