Announcement

Collapse
No announcement yet.

ATI Linux Drivers Gain Support For Unreleased RS880

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

  • ATI Linux Drivers Gain Support For Unreleased RS880

    Phoronix: ATI Linux Drivers Gain Support For Unreleased RS880

    AMD's current flagship offering when it comes to integrated ATI graphics is the Radeon HD 3300 / 790GX. This IGP was introduced last fall as a minor refresh to the Radeon HD 3200 / 780G Chipset...

    http://www.phoronix.com/vr.php?view=NzE1OA

  • #2
    This is beyond great news!

    I do wonder though what will happen when AMD's Fusion (CPU and GPU on die) comes out.

    Will this require a completely new open source driver?

    Comment


    • #3
      The Fusion parts will almost definitely need driver *changes* but I don't think a whole new driver would be required. The important thing is making sure that we have a solid open source driver to start with.
      Last edited by bridgman; 03-22-2009, 09:29 AM.

      Comment


      • #4
        Adding new ids is something really impressive...

        Comment


        • #5
          What about if the CPU and GPU will be able to share cache?

          So it would quite cool if they shared cache on die when using OpenCL
          Last edited by Louise; 03-22-2009, 09:45 AM.

          Comment


          • #6
            I *definitely* can't tell you that. Sorry

            Comment


            • #7
              Originally posted by bridgman View Post
              I *definitely* can't tell you that. Sorry
              ... Of course that would require some sort of weird memory controller...

              It is all interesting

              Actually I don't quite get how the numbers of pins won't be a serious problem for Fusion. A CPU have ~1000 pins, and a GPU have maybe also ~1000.

              I read that CPU's would *never* get more can 16 cores, because the numbers of pins required to the ram banks would not be possible to fit on a die.

              Comment


              • #8
                Originally posted by Kano View Post
                Adding new ids is something really impressive...
                Despite your sarcasm... what's significant here is that the ID is making its way into the driver way before the product releases. So hopefully, the major distros will have support before end-users get their hands on it. This hasn't always happened in the past.

                Comment


                • #9
                  Fine, but please release a version and not only a git snapshot.

                  Comment


                  • #10
                    Originally posted by Louise View Post
                    ... Of course that would require some sort of weird memory controller...

                    It is all interesting

                    Actually I don't quite get how the numbers of pins won't be a serious problem for Fusion. A CPU have ~1000 pins, and a GPU have maybe also ~1000.

                    I read that CPU's would *never* get more can 16 cores, because the numbers of pins required to the ram banks would not be possible to fit on a die.
                    That's what gets interesting about Fusion parts -- nearly all of the pins on a GPU are either power/ground or for communicating with CPU and memory. Since the CPU and memory connections are now on chip, other than extra power/ground for the GPU and pins for the video outputs you don't really need a lot more pins when you add a GPU.

                    Comment


                    • #11
                      Originally posted by bridgman View Post
                      That's what gets interesting about Fusion parts -- nearly all of the pins on a GPU are either power/ground or for communicating with CPU and memory. Since the CPU and memory connections are now on chip, other than extra power/ground for the GPU and pins for the video outputs you don't really need a lot more pins when you add a GPU.
                      That's really interesting. Also in the sense of manufacturing cost. Pins should be quite expensive to wire in a CPU, which is why 64bit AMD CPU's only support 2^42bit memory and not 2^64bit -- for the time being

                      I bet Fusion chips will become quite cheap.

                      I predict that at some point X GB of RAM is "enough for anyone", and RAM will be on die too for Fusion products.

                      Comment


                      • #12
                        Originally posted by Louise View Post
                        I predict that at some point X GB of RAM is "enough for anyone", and RAM will be on die too for Fusion products.
                        I'm not sure that makes sense, just in terms of die size vs. yield. It would basically turn a small chip into a huge chip. A multi-chip package might work (cf. TI'S OMAP3 family and the Xbox 360 GPU), but I don't know how much RAM can be feasibly crammed into one package. This could have some interesting implications for netbooks...

                        Comment


                        • #13
                          Originally posted by Ex-Cyber View Post
                          I'm not sure that makes sense, just in terms of die size vs. yield. It would basically turn a small chip into a huge chip. A multi-chip package might work (cf. TI'S OMAP3 family and the Xbox 360 GPU), but I don't know how much RAM can be feasibly crammed into one package. This could have some interesting implications for netbooks...
                          Okay, that's a pretty good point.

                          Comment


                          • #14
                            RAM could become integrated just like sound cards, and ethernet/usb/firewire/etc ports

                            Comment


                            • #15
                              Just found this about Fusion:
                              http://www.xbitlabs.com/news/cpu/dis..._Products.html

                              The first accelerated processing units (APUs) - the term AMD uses to describe its CPU-GPU chips - will be code-named Llano and Ontario. The former - Lllano - will feature four cores, 4MB of cache, graphics processing engine and DDR3 memory controller; the latter - Ontario - will sport two cores, 1MB of cache and DDR3 memory controller. The Llano will be targeted at mainstream desktop market segment, whereas the latter will be aimed at ultra-portable notebooks.
                              How on earth can 4MB cache be split or shared between 4 CPU cores and 1 GPU ?

                              Or is "graphics processing engine" != GPU ?

                              Fusion is more interesting than I first thought

                              Last edited by Louise; 03-23-2009, 08:13 AM.

                              Comment

                              Working...
                              X