Announcement

Collapse
No announcement yet.

AMD Releases R600/700 Programming Guide

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

  • #21
    It's probably about break-even in terms of direct costs; there are some extremely capable developers working on the open source drivers who are not on AMD's payroll, but the planning and execution of the IP review process eats up time from our most senior technical people, the ones who would otherwise be working on the GPUs and drivers we'll be selling 2-4 years from now.

    The same goes for intangibles (risk, market benefits, "halo" effect, indirect benefit for other markets like embedded etc..) only the numbers are even harder to quantify.

    What I think it boils down to is that if we were still a pure GPU company it would be hard to justify doing this, but as a GPU/CPU/platform company it does make sense. The hard part is that you still end up comparing apples and oranges when trying to determine the overall cost/benefit.
    Test signature

    Comment


    • #22
      So far only AMD CPU's have nested page tables, which improves the speed of para-virtualization a lot.

      So I am thinking, which these specs, would it be possible to implement virtual page tables for AMD GPU's, so you can swap out an entire page table?

      Comment


      • #23
        Oh my gosh so I could finally get my 4870X2 working 3d in Fedora 11???

        Comment


        • #24
          Originally posted by ap90033 View Post
          Oh my gosh so I could finally get my 4870X2 working 3d in Fedora 11???
          I doubt it. I'd already consider it lucky if 3D got from experimental to basic (as in, could maybe even handle *gasp* glxgears or Compiz) by Fedora 12. Chill out, you'll likely have upgraded your distro version at least once or even more times if you expect to get it without installing it yourself from git. (just to clarify: I did not just say that current git would be usable for any real use case, still locks up on eg glxgears)

          Comment


          • #25
            Originally posted by bridgman View Post
            The hard part is that you still end up comparing apples and oranges when trying to determine the overall cost/benefit.
            Not to mention, do you have the benefits yet? I'm currently running nVidia with the proprietary drivers, and I see absolutely no reason to buy an AMD card in order to run Catalyst. Not unless you're really trashing nVidia in Windows game performance, but there you're usually competitively priced anyway (margins are another matter). I want to choose open source, but I'm fairly pragmatist.

            That means, what can I get out of the box on mainstream distros using open source? I don't expect them to be default but compiling git trees is pretty much out. I don't expect it to win FPS shootouts but I want a 50% unoptimized acceleration not a 1% software rendering solution. You're not at the point where I want to buy it yet, but you're getting closer.

            I'll probably be ready for another purchase once the Radeon R800 generation arrives. The question is if your open source drivers will be ready to reap the benefits, or at least give me credible faith they will soon. Yes, I know you can't talk about unreleased products and certainly not product lines but as a general advice, try to be on top of new releases. They're often the ones that cause purchase decisions, which is when you have to be there.

            Comment


            • #26
              Originally posted by Kjella View Post
              I'll probably be ready for another purchase once the Radeon R800 generation arrives.
              I think there's general suspicions (R800 isn't out yet so no one really knows at this point and people who might otherwise know couldn't really even tell they know since they'd have signed NDA's anyway) R800 would be relatively similar R6xx/R7xx when it comes to the 3D engine so the work on them could then be ported over when developers get to know the differences. (assuming the expectations are true, the faster R6xx/R7xx 3D gets in a working shape, the more likely it is that R800 will start getting useful support ported to it soon after release) Sadly no one can really promise anything, neither in how soon the support will be written nor in that the new chipset family would have similar 3D engine. We'll see.

              Comment


              • #27
                Yeah, a lot of the code to support new GPU generations piggybacks on code for earlier products, so it's important to get the previous generations solid first. We put docco for the 3xx-5xx generation out first so the devs could be working with that while we were writing docs and sample code for 6xx/7xx; the critical docs and sample code came out about 6 months ago.

                Similarly, we need 6xx/7xx 3D support to be solid in order to provide a foundation for 8xx work. I can't comment on the degree of change between 7xx and 8xx, of course, but we have already started collecting info for the docs and sample code.

                Originally posted by Kjella View Post
                That means, what can I get out of the box on mainstream distros using open source? I don't expect them to be default but compiling git trees is pretty much out. I don't expect it to win FPS shootouts but I want a 50% unoptimized acceleration not a 1% software rendering solution. You're not at the point where I want to buy it yet, but you're getting closer.

                I'll probably be ready for another purchase once the Radeon R800 generation arrives. The question is if your open source drivers will be ready to reap the benefits, or at least give me credible faith they will soon. Yes, I know you can't talk about unreleased products and certainly not product lines but as a general advice, try to be on top of new releases. They're often the ones that cause purchase decisions, which is when you have to be there.
                What you get out of box right now with open source drivers on the latest chips is solid EXA and Xv acceleration plus software 3D, which actually meets the needs of a surprising number of users. The 3D driver is written and had started integration testing in April, but we had to take a little side-trip to adapt to the radeon-rewrite code base which changed the lower level bits of the stack quite a bit (albeit in a good way).

                Porting 6xx/7xx 3D to the -rewrite code base is finished and we have resumed integration testing. We are down to a single buffer management issue that causes soft hangs after a few frames; once that is fixed I think things should seem to progress pretty quickly (as they normally do when all the hard work was done months ago ).

                Top priority for the next round of distro releases is getting the kernel driver code in the box, since updating userspace drivers via package manager is usually a no-brainer.

                There is a disconnect between the release cycles of typical distros and the market inflection points that drive new hardware releases, in the sense that distro releases tend to come out just *before* new hardware (which sucks for users who prefer open source drivers), but I think that will evolve into something that works a bit better for everyone over time. Ubuntu release timing is pretty good in the sense that you can usually *just* squeeze new hardware support in, but even there the timing is extremely tight and you end up working outside the distro packager's comfort zone.
                Last edited by bridgman; 11 July 2009, 12:24 AM.
                Test signature

                Comment


                • #28
                  Originally posted by bridgman View Post
                  Ubuntu release timing is pretty good in the sense that you can usually *just* squeeze new hardware support in, but even there the timing is extremely tight and you end up working outside the distro packager's comfort zone.
                  That and I think they also seemed quite aware of how the drivers were progressing for the 9.04 (not very surprising considering all the noise about radeon EXA/Xv in Linux land) and they decided to put it into their 2.6.28 kernels. A cynical viewer could see a pattern here where distros find kernel releases to come so seldom they end up having to backport the interesting changes themselves thus putting themselves in kernel maintainers' boots. In Ubuntu the stuff got in two kernel versions early and in Fedora one early; in upstream Linux the functionality only got to 2.6.30. I'm not aware how it ended up being in other distros. I'd assume Ubuntu and Fedora would backport (as in, new stuff for older kernels) the 3D stuff into their systems too if it gets in a reasonably working state before freezes and the kernel release dates are inconvenient.

                  Comment


                  • #29
                    So what you are saying is sell my 4870x2 since its crap in linux?

                    Comment


                    • #30
                      No, just wait. You can't play games (not like: pop cd in, install, play) on Linux anyway. Keep Windows for another six months or so.

                      Comment

                      Working...
                      X