Announcement

Collapse
No announcement yet.

to bridgman: Radeon video card questions regarding DEC Alpha systems

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

  • to bridgman: Radeon video card questions regarding DEC Alpha systems

    Hi,

    Obviously your #1 constituency isn't users of weird architectures. None of the following questions are directly related to Alpha systems per se.

    Firstly, very few Alpha systems have AGP slots. The most common board with an AGP (the UP1500) has the 3.3V signalling notch, which limits the selection of cards to 9800 Pro and older, although its chipset supports AGP 4x.

    Alternatively, we could get PCI Radeons, such as the X1550 or 2400 Pro. Would an X1550 or 2400 Pro on a PCI bus be faster than a 9800 Pro on an AGP bus? Application dependent maybe?

    Since the 2400 Pro seems to have a 64-bit memory bus and the X1550 has 128-bit, I would assume the X1550 is faster. From a bit of research, the 2400 has 40 stream processors while the X1550 has 4. Which would actually be faster?

    Certainly we can expect the X1550/2400 to receive more driver support and in turn better features in the coming months.

    What about video acceleration? I've been following the XvMC thread, but I'm unsure which features are specific to R500/R600 cards and which to older generations. Are any of these cards really fast enough for significant video acceleration?

    Should we expect to see any R700 (or newer) PCI cards?

    Finally, we've been testing video card compatibility with this board. Our current theory is that cards with only the 3.3V notch work. That is, cards with both the 3.3V and 1.5V notches cause the system to hang on start up. We've found one exception thus far: the R200 series. Was the R200 series made using 3.3V signalling with 1.5V added on later? This would certainly explain this discrepancy. It may be the case that the 9800 Pro is just a fantasy and that actually the fastest AGP card is the 8500.

    I understand these questions are not normal, so I (and other Alpha users) appreciate your response a lot.
    Last edited by mattst88; 09 January 2009, 05:35 PM.

  • #2
    Originally posted by mattst88 View Post
    Firstly, very few Alpha systems have AGP slots. The most common board with an AGP (the UP1500) has the 3.3V signalling notch, which limits the selection of cards to 9800 Pro and older, although its chipset supports AGP 4x.

    Alternatively, we could get PCI Radeons, such as the X1550 or 2400 Pro. Would an X1550 or 2400 Pro on a PCI bus be faster than a 9800 Pro on an AGP bus? Application dependent maybe?
    OK, first note - take all my answers with a grain of salt for now; will ask around next week and see if I can at least get confirmation on them.

    My feeling is that the 9800 Pro is probably the faster chip anyways (see below), so that makes the PCI/AGP issue moot.

    Originally posted by mattst88 View Post
    Since the 2400 Pro seems to have a 64-bit memory bus and the X1550 has 128-bit, I would assume the X1550 is faster. From a bit of research, the 2400 has 40 stream processors while the X1550 has 4. Which would actually be faster?
    The rule of thumb I use for comparing older cards with unified shader cards is to count each vertex shader as 4 SPs, and each pixel shader as 5 SPs. That gives 4x4 + 5x8 = 56 SPs for a 9800 Pro, 4x2 + 5x4 = 28 SPs for an X1550, and 40 SPs for a 2400 Pro. Obviously clock rates, texturing power and memory type / bus width are a factor as well, but I think you would need to go at least to an X1650 or HD2600 to outrun a 9800 Pro.

    Originally posted by mattst88 View Post
    Certainly we can expect the X1550/2400 to receive more driver support and in turn better features in the coming months.
    For the proprietary drivers, that's probably a safe assumption (particularly 2400). For the open source drivers, I think the plan is to extend as many features as possible back through the entire radeon family. There are some hardware-dependent cutoffs, ie Gallium will probably only go back to R3xx, and some of the higher GL functions will be limited on pre-5xx due to the lack of dynamic flow control in the hardware, but that's about it.

    AGP support is, in some ways, easier on the older GPUs because we don't need to deal with the added complexity of bridge chips. The newer chips are easier to deal with in other ways, of course, simply by virtue of being newer.

    Originally posted by mattst88 View Post
    What about video acceleration? I've been following the XvMC thread, but I'm unsure which features are specific to R500/R600 cards and which to older generations. Are any of these cards really fast enough for significant video acceleration?
    The video overlay used to have a fairly powerful processing block on cards up to 4xx (ie 5xx and up did not have it). The block only accelerated Xv-type acceleration, so on newer chips we use Textured Video for Xv. The 9800 Pro is probably the only one of the three you mentioned with enough power to have a chance of doing much video decode acceleration on the shaders, unless we are able to open up the UVD block on the 2400 (which I'm saying "no" to right now until we have a chance to look at it more deeply). They can all do a good job of Xv-type acceleration, however.

    Originally posted by mattst88 View Post
    Should we expect to see any R700 (or newer) PCI cards?
    Don't know. I'm not sure it makes sense for us to support legacy buses with every generation of GPU, although it probably is worth making sure we do keep at least one good card available for each bus.

    Originally posted by mattst88 View Post
    Finally, we've been testing video card compatibility with this board. Our current theory is that cards with only the 3.3V notch work. That is, cards with both the 3.3V and 1.5V notches cause the system to hang on start up. We've found one exception thus far: the R200 series. Was the R200 series made using 3.3V signalling with 1.5V added on later? This would certainly explain this discrepancy. It may be the case that the 9800 Pro is just a fantasy and that actually the fastest AGP card is the 8500.
    I don't really know much about this at all, and some of the information may only exist in developer's heads. I'll see if I can find anything.
    Last edited by bridgman; 12 January 2009, 12:12 PM.
    Test signature

    Comment


    • #3
      ^ This is why ATI is great. Awesome responsivity, bridgman... this kind of customer-vendor dialogue without the encumbrance of the traditional support organization "filter" is just fantastic.

      Comment


      • #4
        Firstly, thank you so much for your reply. We really appreciate what you do.

        Originally posted by bridgman View Post
        will ask around next week and see if I can at least get confirmation on them.
        Thanks, that would be awesome.

        Originally posted by bridgman View Post
        I don't really know much about this at all, and some of the information may only exist in developer's heads. I'll see if I can find anything.
        Again, thanks.

        I'll try to grab a 9800 Pro to verify it works. If it doesn't, a lot of this is just academic, unfortunately.

        Comment

        Working...
        X