Announcement

Collapse
No announcement yet.

Intel Sandy Bridge VA-API Performance

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

  • #31
    Originally posted by crazycheese View Post
    You will not, granted you possess the clip.
    So if A/B==gfx cards of same owner, owning the clip and only publishing the results - its not a problem.
    But if A/B==different people with different machines they will all be required to license the clip, its logical. Yes, thats what Michael told me.
    The problem with using custom clips that cannot be used by everyone is that publishing results and comparison to other setups become meaningless to everyone except the actual tester.

    And if we create such clip - from big buck bunny lower bitrate h264 as source or multiple big buck bunny screens rendered from one lower bitrate formats into new file especially for phoronix test? It won't be commercial encode, the source is also free - should be no problem. Of course it will require some location to be buffered on the internet.
    What do you think?
    When it comes to video decoding performance you generally want a realistic "worst case" clip. Low bit rate clips do not stress the system enough to provide any meaningful data. When I'm shopping to purchase a truck for example, I don't see if it can handle a 25 pound bag of spuds if I possibly will be using it to haul around a camper.

    Comment


    • #32
      Originally posted by deanjo View Post
      The problem with using custom clips that cannot be used by everyone is that publishing results and comparison to other setups become meaningless to everyone except the actual tester.
      Yes, yes - I didn't reply to Michael because of this too, its clear.


      Originally posted by deanjo View Post
      When it comes to video decoding performance you generally want a realistic "worst case" clip. Low bit rate clips do not stress the system enough to provide any meaningful data. When I'm shopping to purchase a truck for example, I don't see if it can handle a 25 pound bag of spuds if I possibly will be using it to haul around a camper.
      Say you take 4Mbit BBB sample and put it together 4x on same frame while resizing it to 1/4 of fullhd on each side. You have already 16Mbit/s detail stream. You can take cut and combine different moments from BB - the static as well as dynamic ones. With less/more decent editor you should be able to pick the most intensive segments. You could make a loop and play it several times within the test. There may be good advantages in such manually respinned test track - full control, no restriction to modification, no license problems for use in test. The only disadvantage I see is the resulting size(again depending on how long you want to make it) that should be stored somewhere.

      Comment


      • #33
        Originally posted by crazycheese View Post
        Say you take 4Mbit BBB sample and put it together 4x on same frame while resizing it to 1/4 of fullhd on each side. You have already 16Mbit/s detail stream.
        Not quite, you have 4 x 4 Mbit streams. There are various issues with this as the various decoders have limitations as to how many HD streams they are capable of playing using the HW decoder. IIRC Nvidia officially only supports one, intel supports 2 HD streams and ATI supports one HD stream and one SD stream.

        Comment


        • #34
          I should also note that for nvidia at least, even though it only officially supports one stream it seems that it is purely based on if the decoder has enough resources free to accommodate the additional load up to it's max capabilities.

          Comment


          • #35
            Originally posted by deanjo View Post
            Not quite, you have 4 x 4 Mbit streams. There are various issues with this as the various decoders have limitations as to how many HD streams they are capable of playing using the HW decoder. IIRC Nvidia officially only supports one, intel supports 2 HD streams and ATI supports one HD stream and one SD stream.
            Yes, I'm also aware of this fact. However my proposal was completely different, unless I missunderstand you.

            My proposal was to combine four videostreams of the free licensed movie into one videostream using demux, decode, quality resize, picture-in-picture filters with final 16Mbit 2 pass compression.

            The other possibility is to generate huge amount of raw noise in fullhd frame and encode it into any preferable format, setting encoder two pass average bitrate @ desireable.

            Comment


            • #36
              a 40 mbit peak file would be much better, average maybe 30 mbit for video.

              Comment


              • #37
                Originally posted by crazycheese View Post
                Yes, I'm also aware of this fact. However my proposal was completely different, unless I missunderstand you.

                My proposal was to combine four videostreams of the free licensed movie into one videostream using demux, decode, quality resize, picture-in-picture filters with final 16Mbit 2 pass compression.
                The final bitstream since you are combining into one stream could be at any bit rate since you are reencoding. It's going to be far easier to encode from raw png and just get a good quality high bitrate video with peaks of video data @ 40 mbit. Like I say the best open movie that would be a tough decode would be elephant dreams where there isn't a lot of redundant data because of the massive amounts of detailed movement going on in the background.

                Comment


                • #38
                  OC the 720p60 tall ships clip on the free x264 Blu-ray and AVCHD ISO is also a nice test



                  perhaps Michael should also house that in the Phoronix Test Suite tree along side other real life video clips i linked to OC

                  http://forum.doom9.org/showthread.php?t=154183
                  25th April 2010, 04:46
                  read the tread for more info and links/thoughts for higher bitrate etc.

                  Comment


                  • #39
                    Other thoughts

                    I've been an AMD/ATi fanboy for a long time. I really liked where they where heading in the opensource department too. That said ...

                    wow, this does look like a really nice and open solution for a mediacenter/htpc. Being able to decode 1080p in hardware with opensource drivers is a big plus for me.

                    I do have a question though, maybe slightly off-topic. AMD doesn't opensource this bit (yet) because the whole DRM debacle. Now AMD (and nVidia?) say they have specific asic's in their die's for this? UVD, purevideo. Does the intel video chip have this aswell? Or, like I always thought, is UVD, purevideo just some extra functions and does the driver/chip still do most of the heavy lifting via shaders?

                    Comment


                    • #40
                      Originally posted by oliver View Post
                      Does the intel video chip have this aswell? Or, like I always thought, is UVD, purevideo just some extra functions and does the driver/chip still do most of the heavy lifting via shaders?
                      I don't know about special unit to decode the video, but sandybridge itself has very strong hollywood sexual fantasy, ie drm (guess hence the name - sandy, ie filtering, bridge - to deliver content) that was heavily criticized. You can read about it if you google for sandybridge + drm, obviously.

                      Comment


                      • #41
                        Originally posted by oliver View Post
                        I've been an AMD/ATi fanboy for a long time. I really liked where they where heading in the opensource department too. That said ...

                        wow, this does look like a really nice and open solution for a mediacenter/htpc. Being able to decode 1080p in hardware with opensource drivers is a big plus for me.

                        I do have a question though, maybe slightly off-topic. AMD doesn't opensource this bit (yet) because the whole DRM debacle. Now AMD (and nVidia?) say they have specific asic's in their die's for this? UVD, purevideo. Does the intel video chip have this aswell? Or, like I always thought, is UVD, purevideo just some extra functions and does the driver/chip still do most of the heavy lifting via shaders?
                        the Nvidia pure-video and indeed the AMD UVD3 are fixed function ASIC as in they cant change much about the bug's inside them to remove the problem without ripping them out and totally replacing them all,and can only work software kludges run on whatever's available to them at the time.

                        however, by all account's the internal Intel Encode/Decoder is more a collection of basic fixed function and full collection of DSP's with a limited related re-programmability, much like the OMAP 4/5 ARM SOC design.

                        not as good as say simply sticking a bulk purchased state of the art 1.5GHz Achronix FPGA (apparently priced at ASIC prices to compete today) http://www.eetimes.com/electronics-n...artup-Achronix
                        on there somewhere and charging the end user a few quid more to cover the initial cost's, but far better long term than any fixed function ASIC today as you can just sell add-on functionality and fix found bugs with a simple BIOS/board FPGA update, but apparently that's to expensive and to far an innovative jump for some people in 2011/12, they do love their ASIC markets

                        Comment


                        • #42
                          So is VA-API + a Core i3-530 now good enough...
                          Such that one doesn't need to bother with a discrete GT_430* + VDAPU?


                          *I'm also considering a GTX 560 Ti...
                          I know it consumes way more power, but it's also a proper gaming card.
                          I'm a very casual gamer though, so I'm not sure if its worthwhile.
                          There's a single-slot wc block coming soon that can easily be added to my wc loop.
                          So the Ti needn't block any of my precious PCI/e slots.

                          Comment


                          • #43
                            Originally posted by Michael View Post
                            Where is any other high quality sample content that's widely available on HTTP/FTP mirrors and under a free license? I haven't seen such links.
                            Here is a bunch of 1080p files that I recorded myself you can test with: http://media.rawdod.com/?dir=/Music%20Videos

                            If you need me to write some special thing for the license let me know.

                            Comment


                            • #44
                              http://media.rawdod.com/files/Music%...harmcenter.mkv

                              http://media.rawdod.com/files/Music%...d_new_cont.mov

                              Comment


                              • #45
                                They will never use something which isn't widely available on mirrors because it has to be replicable by other PTS users, as always benchmarks accuracy doesn't matter.
                                ## VGA ##
                                AMD: X1950XTX, HD3870, HD5870
                                Intel: GMA45, HD3000 (Core i5 2500K)

                                Comment

                                Working...
                                X