Loading ...
Sorry, an error occurred while loading the content.

performance

Expand Messages
  • liaoyanchun <antie@21cn.com>
    Anyone know the performance of pnx1500 vs equator s MAP-CAP or BSP?
    Message 1 of 6 , Jan 1, 2003
    View Source
    • 0 Attachment
      Anyone know the performance of pnx1500 vs equator's MAP-CAP or BSP?
    • chuck.peplinski@philips.com
      TriMedia costs less than Equator. The 1500 is more powerful than the 1300 (clock close to 300Mhz, 24 new instructions), but in terms of raw MHz, the Equator
      Message 2 of 6 , Jan 1, 2003
      View Source
      • 0 Attachment
        TriMedia costs less than Equator. The 1500 is more powerful than the 1300 (clock close to 300Mhz, 24 new instructions), but in terms of raw MHz, the Equator parts probably run faster. The cache strategies are very different. This makes the optimization
        tasks for the two very different. I can't comment on which is better. I don't know much about the Equator's I/O capabilities.

        Chuck





        "liaoyanchun
        <antie@...> To: trimedia@yahoogroups.com
        " <antie cc: (bcc: Chuck Peplinski/SVL/SC/PHILIPS)
        Subject: [trimedia] performance
        01/01/2003 07:54
        AM Classification:
        Please respond
        to trimedia






        Anyone know the performance of pnx1500 vs equator's MAP-CAP or BSP?



        Sent via the TriMedia mailing list
        trimedia@yahoogroups.com - archive at http://groups.yahoo.com/group/trimedia

        Your use of Yahoo! Groups is subject to http://docs.yahoo.com/info/terms/
      • 建军 刘
        Hi: As far as I know that equator s developing tools are very expensive and you should buy the 3rd party software to build your own applications!And Equator is
        Message 3 of 6 , Jan 2, 2003
        View Source
        • 0 Attachment
          Hi:
          As far as I know that equator's developing tools are
          very expensive and you should buy the 3rd party
          software to build your own applications!And Equator is
          not as big as Philips!That is my own option!

          Happy New Year!
          crazyflower77cn


          _________________________________________________________
          Do You Yahoo!?
          "洋节土节都要过,大礼小礼不放过"
          http://cn.promo.yahoo.com/holiday/
        • Chris Bore <chris@bores.com>
          ... Comparing these two is not easy directly without setting up the target application on both and measuring. Then there are issues of cost, and of the
          Message 4 of 6 , Jan 3, 2003
          View Source
          • 0 Attachment
            --- In trimedia@yahoogroups.com, "liaoyanchun <antie@2...>"
            <antie@2...> wrote:
            > Anyone know the performance of pnx1500 vs equator's MAP-CAP or BSP?

            Comparing these two is not easy directly without setting up the
            target application on both and measuring. Then there are issues of
            cost, and of the relative level of development tools. In a gross
            oversimplification, I used to say that the Equator was faster than
            pnx1300 but the pnx1300 was cheaper and had better development tools.
            This is no longer a valid comment (it was always a silly comment
            since as I said, comparison without application benchmarks is not
            sensible).

            pnx1500 is more than just a faster pnx1300: the new Custom Operations
            each offer speed improvements of at least 2x over the pnx1300 in the
            target applications, and there are 20 of these new instructions. The
            pnx1500 peripherals and coprocessors are MUCH more powerful than
            those on the pnx1300. pnx1500 also addresses some needs in video
            enhancement, in driving diverse displays (including LCD), and in
            networking, that are not addressed by Equator.

            Let me outline some characteristics of pnx1500 and BSP-15, and
            suggest some differences. I am not even going to attempt to compare
            software tools. This is inevitably a lightweight assessment, and I
            offer it under that understanding:

            As for the core, the main difference comes from Equator's 64-bit and
            128-bit register sizes as compared with TriMedia's 32-bit. pnx1500
            has more functional units than did pnx1300, and has 20 new
            instructions that greatly speed up many applications - these make
            pnx1500 much faster in practice than pnx1300 and probably faster than
            BSP-15 at target applications.

            1) Equator is 4-slot VLIW, TriMedia is 5-slot VLIW. BUT, TriMedia has
            many more available functional units (27 in pnx1300, 31 in pnx1500)
            than issue slots, and some of TriMedia's performance comes from the
            compiler being able to choose suitable units for a wide range of
            applications and parts of applications and so make better use of the
            five issue slots that are potentially usable.

            2) Both Equator and TriMedia have some SIMD units that can carry out
            two or four operations in each issue slot. Equator can also do eight
            1-byte operations, due to its 64-bit register size. Equator's are
            called Media Instrinsics, TriMedia's are called Custom Operations.
            Understanding of SIMD instructions, and how the target application
            maps onto these, is essential in assessing either of these processors.

            3) Equator has 64-bit (and 128-bit) registers.

            4) Equator's advantage in core speed will be eroded by pnx1500's
            extended instruction set, new larger icache, and additional
            functional units.

            Often, the true differences appear when considering the peripherals.
            The pnx1500 peripherals are also very much more powerful than those
            on the pnx1300. In some areas where Equator had an advantage over
            pnx1300, that advantage will be eroded and overwhelmed by pnx1500's
            new peripherals:

            1) Equator's Display Refresh Controller (DRC) performs a similar role
            to the pnx1300's Video Output: scaling is handled separately by the
            BSP-15 Video Filter, whereas pnx1300 does this in the Video Output,
            but that is a trivial difference. The video blending and overlay are
            handled by pnx1300 Video Output, and by BSP-15 DRC, respectively,
            with relatively minor differences. pnx1500 replaces Video Output with
            a much more powerful Video Composition Processor (VCP) that can
            output in HD, VGA and LCD modes, and that can handle 10-bit video: I
            think this exceeds the BSP-15 capabilities significantly.

            2) Equator's DRC has 2D drawing capability which pnx1300 lacked (2D
            drawing being in software). pnx1500 now has a 2D drawing coprocessor
            engine that matches or exceeds BSP-15's performance in this area.

            3) pnx1500 VCP now handles Video Quality Enhancement, (for instance
            histogram modification and skin tone correction) in hardware. This
            goes beyond anything available on the BSP-15.

            4) The pnx1500 Memory Based Scaler (MBS) goes far beyond the old
            Image CoProcessor (ICP). The MBS now performs de-interlacing and anti-
            flicker filtering in hardware, as well as scaling. These capabilities
            are not available on BSP-15.

            5) pnx1500 has a new and very versatile General Purpose I/O interface
            that has much more capability than the BSP-15 general purpose port.

            6) pnx1500 has a new 10/100 LAN interface. There is no such interface
            on BSP-15.

            Overall I am convinced that the pnx1500 is a very great improvement
            on pnx1300, and very competitive with Equator except where many 64-
            bit or 128-bit operations are required, which would be unusual for
            the target applications of pnx1500. The main message, though, is that
            just making such a comparison is not at all easy or straightforward.
            The only reasonable way to proceed is to define the target
            application closely, and make a proper assessment of how the required
            tasks and operations map onto the underlying architecture - taking
            full account of the System-On-Chip peripherals and the nature and
            match of the SIMD units as well as raw core performance.

            I am working on a new set of classes on the pnx1500, so my own
            knowledge and understanding should grow over the next few weeks. I
            need to make clear that what I say is my personal opinion (which may
            not be balanced), and is based on my own knowledge and understanding
            (which may not even be correct, let alone complete). :-) I would be
            glad for any corrections though.

            You can find some details on pnx1500 at:

            http://www.semiconductors.philips.com/news/content/file_895.html

            Chris
            ------------------------------------
            Chris Bore
            TriMedia Yahoo Group moderator
            BORES Signal Processing
            http://www.bores.com
            chris@...
          • Tim Roberts
            ... I, for one, would like to extend a word of thanks to Chris Bores for his scholarly and objective comparison of the Equator and the PNX1500. This one is
            Message 5 of 6 , Jan 6, 2003
            View Source
            • 0 Attachment
              On Fri, 03 Jan 2003 15:27:33 -0000, "Chris Bore" <chris@...> wrote:
              >
              >--- In trimedia@yahoogroups.com, "liaoyanchun <antie@2...>"
              ><antie@2...> wrote:
              >> Anyone know the performance of pnx1500 vs equator's MAP-CAP or BSP?
              >
              >Comparing these two is not easy directly without setting up the
              >target application on both and measuring. Then there are issues of
              >cost, and of the relative level of development tools. In a gross
              >oversimplification, I used to say that the Equator was faster than
              >pnx1300 but the pnx1300 was cheaper and had better development tools.
              >This is no longer a valid comment (it was always a silly comment
              >since as I said, comparison without application benchmarks is not
              >sensible).

              I, for one, would like to extend a word of thanks to Chris Bores for his
              scholarly and objective comparison of the Equator and the PNX1500. This one is
              getting filed for the future!

              --
              - Tim Roberts, timr@...
              Providenza & Boekelheide, Inc.
            • Torsten Fink
              my 2 cents: Chris, good job! I would like to add that Equator greatly benifits from its programable DMA engine (64-channel DataStreamer DMA controller). It
              Message 6 of 6 , Jan 28, 2003
              View Source
              • 0 Attachment
                my 2 cents:

                Chris, good job!

                I would like to add that Equator greatly benifits from its programable DMA
                engine (64-channel DataStreamer DMA controller). It pratically eliminates
                D-Cache misses for MPEG like SW components. On the other hand, it also
                complicates the porting of SW. Equator also has an MMU, which is not
                available on the 1500.

                The 1500 on the other hand comes with proprietary video blocks which
                leverage decades of basic video research. The MBS supports three different
                deinterlacing modes and the measurement of all sorts of video data: black
                bar, black level, temporal noise, UV bandwidth, histogram etc. The video
                output block supports a variety of video quality improvement options, like
                skin-tone correction, histogram modification, sharpness enhancement, gamma
                correction etc. Apart from this, one should also look at the availability of
                software. The 1500 should have the clear advantage here, because pretty much
                all existing SW components can be just recompiled (except for those using
                older pnx1300 peripherals).

                Torsten

                -----Original Message-----
                From: Chris Bore <chris@...> [mailto:chris@...]
                Sent: Friday, January 03, 2003 7:28 AM
                To: trimedia@yahoogroups.com
                Subject: [trimedia] Re: performance


                --- In trimedia@yahoogroups.com, "liaoyanchun <antie@2...>"
                <antie@2...> wrote:
                > Anyone know the performance of pnx1500 vs equator's MAP-CAP or BSP?

                Comparing these two is not easy directly without setting up the
                target application on both and measuring. Then there are issues of
                cost, and of the relative level of development tools. In a gross
                oversimplification, I used to say that the Equator was faster than
                pnx1300 but the pnx1300 was cheaper and had better development tools.
                This is no longer a valid comment (it was always a silly comment
                since as I said, comparison without application benchmarks is not
                sensible).

                pnx1500 is more than just a faster pnx1300: the new Custom Operations
                each offer speed improvements of at least 2x over the pnx1300 in the
                target applications, and there are 20 of these new instructions. The
                pnx1500 peripherals and coprocessors are MUCH more powerful than
                those on the pnx1300. pnx1500 also addresses some needs in video
                enhancement, in driving diverse displays (including LCD), and in
                networking, that are not addressed by Equator.

                Let me outline some characteristics of pnx1500 and BSP-15, and
                suggest some differences. I am not even going to attempt to compare
                software tools. This is inevitably a lightweight assessment, and I
                offer it under that understanding:

                As for the core, the main difference comes from Equator's 64-bit and
                128-bit register sizes as compared with TriMedia's 32-bit. pnx1500
                has more functional units than did pnx1300, and has 20 new
                instructions that greatly speed up many applications - these make
                pnx1500 much faster in practice than pnx1300 and probably faster than
                BSP-15 at target applications.

                1) Equator is 4-slot VLIW, TriMedia is 5-slot VLIW. BUT, TriMedia has
                many more available functional units (27 in pnx1300, 31 in pnx1500)
                than issue slots, and some of TriMedia's performance comes from the
                compiler being able to choose suitable units for a wide range of
                applications and parts of applications and so make better use of the
                five issue slots that are potentially usable.

                2) Both Equator and TriMedia have some SIMD units that can carry out
                two or four operations in each issue slot. Equator can also do eight
                1-byte operations, due to its 64-bit register size. Equator's are
                called Media Instrinsics, TriMedia's are called Custom Operations.
                Understanding of SIMD instructions, and how the target application
                maps onto these, is essential in assessing either of these processors.

                3) Equator has 64-bit (and 128-bit) registers.

                4) Equator's advantage in core speed will be eroded by pnx1500's
                extended instruction set, new larger icache, and additional
                functional units.

                Often, the true differences appear when considering the peripherals.
                The pnx1500 peripherals are also very much more powerful than those
                on the pnx1300. In some areas where Equator had an advantage over
                pnx1300, that advantage will be eroded and overwhelmed by pnx1500's
                new peripherals:

                1) Equator's Display Refresh Controller (DRC) performs a similar role
                to the pnx1300's Video Output: scaling is handled separately by the
                BSP-15 Video Filter, whereas pnx1300 does this in the Video Output,
                but that is a trivial difference. The video blending and overlay are
                handled by pnx1300 Video Output, and by BSP-15 DRC, respectively,
                with relatively minor differences. pnx1500 replaces Video Output with
                a much more powerful Video Composition Processor (VCP) that can
                output in HD, VGA and LCD modes, and that can handle 10-bit video: I
                think this exceeds the BSP-15 capabilities significantly.

                2) Equator's DRC has 2D drawing capability which pnx1300 lacked (2D
                drawing being in software). pnx1500 now has a 2D drawing coprocessor
                engine that matches or exceeds BSP-15's performance in this area.

                3) pnx1500 VCP now handles Video Quality Enhancement, (for instance
                histogram modification and skin tone correction) in hardware. This
                goes beyond anything available on the BSP-15.

                4) The pnx1500 Memory Based Scaler (MBS) goes far beyond the old
                Image CoProcessor (ICP). The MBS now performs de-interlacing and anti-
                flicker filtering in hardware, as well as scaling. These capabilities
                are not available on BSP-15.

                5) pnx1500 has a new and very versatile General Purpose I/O interface
                that has much more capability than the BSP-15 general purpose port.

                6) pnx1500 has a new 10/100 LAN interface. There is no such interface
                on BSP-15.

                Overall I am convinced that the pnx1500 is a very great improvement
                on pnx1300, and very competitive with Equator except where many 64-
                bit or 128-bit operations are required, which would be unusual for
                the target applications of pnx1500. The main message, though, is that
                just making such a comparison is not at all easy or straightforward.
                The only reasonable way to proceed is to define the target
                application closely, and make a proper assessment of how the required
                tasks and operations map onto the underlying architecture - taking
                full account of the System-On-Chip peripherals and the nature and
                match of the SIMD units as well as raw core performance.

                I am working on a new set of classes on the pnx1500, so my own
                knowledge and understanding should grow over the next few weeks. I
                need to make clear that what I say is my personal opinion (which may
                not be balanced), and is based on my own knowledge and understanding
                (which may not even be correct, let alone complete). :-) I would be
                glad for any corrections though.

                You can find some details on pnx1500 at:

                http://www.semiconductors.philips.com/news/content/file_895.html

                Chris
                ------------------------------------
                Chris Bore
                TriMedia Yahoo Group moderator
                BORES Signal Processing
                http://www.bores.com
                chris@...



                Sent via the TriMedia mailing list
                trimedia@yahoogroups.com - archive at http://groups.yahoo.com/group/trimedia

                Your use of Yahoo! Groups is subject to http://docs.yahoo.com/info/terms/
              Your message has been successfully submitted and would be delivered to recipients shortly.