Slashdot Mirror


ATI PCI-Express Devices Revealed

JohnQ writes "According to Xbitlabs and AnandTech, the specifications for ATI's newest graphics cards have been revealed. Interesting to note is that all of these next generation video cards will run exclusively on the PEG (PCI-Express x16) interface. This does not bode well for those of us who just paid top dollar for the last generation of AGP cards. Read more about the roadmaps on Anandtech and Xbitlabs"

14 of 344 comments (clear)

  1. For those of you like me... by zegebbers · · Score: 5, Informative

    who know nothing about this so called PCI Express x16, check out these useful sites... True, they're not in english, but as if it's any harder to read than xbitlabs and anandtech ;-) .

    1. Re:For those of you like me... by wowbagger · · Score: 5, Informative
      In a nutshell:

      PCI, AGP, and ISA are all parallel systems - you have a wire for each data bit.

      PCI express uses a VERY high-speed serial bus to carry the data. How high speed? One serial channel will carry more data than a standard 64 bit 66MHz PCI bus.

      The advantages to a serial system are:
      1. No timing skew. With a high speed parallel bus, you have to insure that all the data paths are the same electrical length, or else some bits get to the card before other bits, and you have to reduce the clock rate to prevent errors. With a serial system you have far fewer lines to worry about.
      2. Simpler board design. When you are dealing with a 64 bit data bus, 64 bit address bus, plus control lines, the board design gets a bit complex. With a serial system, you have less than ten lines per channel - a much simpler board layout.
      3. Serial systems usually use a MUCH lower signaling voltage, resulting in MUCH lower EMI and on-board noise.
      4. PCI express allows you to gang serial channels for more bandwidth. Video card saturating one channel? Use two.
      5. Unlike AGP, PCI express is a bus - so each device can busmaster to system memory or to other cards as needed. This helps you when your video card wants to store textures in system RAM rather than on-card. Imagine how much fun John Carmack could have with a video card that can support 1G of textures.
      6. It is far easier to design disconnects for a serial bus, thus allowing for PCI hotswap. In fact, the PC card group is working on a new standard for PC cards (nee PCMCIA) that brings 2 PCI-X channels to card - Cardbus on steroids.
      7. It is possible to route a PCI-express channel out of the computer case to an external chassis. While this is of limited use to the usual computer user, for guys like me it is a boon to be able to have an external chassis that looks just like it is on the main system bus, because it IS.


      PCI express *could* allow you to have a computer that has bays that accept anything - hard disk, video card, extra CPU, NIC, whatever, and plug them it without restarting (unless you are running Windows (cheap shot, as I beleive MS is working on fixing that)). It will allow your video card to REALLY have fast access to system RAM, and especially in 64 bit systems, that could be a LOT of system RAM.

      Good stuff - I can hardly wait 'til it becomes commonly accepted.
    2. Re:For those of you like me... by imsabbel · · Score: 3, Informative

      Actually, the single bandwith of one link isnt that big. See the story, they use 16Links for Graphics.

      But hypertransport is the same in that reagard. Its MUCH easier to run x links a y bits than one link with x*y bits.
      For Example, you have your own clock line for every link, so there is no need for temporal coherence between the links. At Ghz-speeds in copper on a pcb, a few millimeters lenght difference would be enough to kill your signal. But of course you cant put the lines to close to each other because of crosstalk, ect.
      With multiple seriel busses, you just give every link a big enough fifo at each end and no problems...

      --
      HI O WISE PRINCE. WHT TOOK U SO DAM LONG?
  2. Um, no by thetzar · · Score: 5, Informative

    JohnQ, are you some kind of idiot? If you READ the article, you'd see that ATI is releasing dual chipsets of identical performance, one each for PCIe and AGP.

  3. Re:buyers of last-gen AGP cards? by ckotchey · · Score: 5, Informative

    Note that there are two differnet standards - PCI-X is different than "PCI Express".

  4. Re:buyers of last-gen AGP cards? by DrMindWarp · · Score: 5, Informative
    PCI-X != PCI-Express

    Check out PCI-SIG.

  5. Call me crazy... by rqqrtnb · · Score: 4, Informative

    Call me crazy, but it seems to me that the changes ATI is making with R4xx are much less drastic than what Nvidia is doing with NV4x. Nvidia is claiming 3X perfomance increase over NV3x, and up to 8X performance increase in Pixel Shader operations. Yeah, it's all theoretical at this point, but it's something to think about. Of course, if you compare R3xx to NV3x, it appears that ATI just had a better design than Nvidia, for the most part, so they didn't need to change as much.

    Regardless of which chip you favor, it's shaping up to be an interesting battle come springtime! (Or more likely summer for those of use that don't get the very first cards direct from the manufacturers.) Can't wait! When these cards get released, I'll finally be able to afford a Radeon 9800XT. :)

  6. Re:Drivers by parksie · · Score: 5, Informative

    Linux 2.6 + XFree86 4.3

    Enable the "radeon" DRI driver in the kernel, use "radeon" in your XF86Config, and all is good. If you want to stick with 2.4, *disable* all DRI support in-kernel, and grab the DRI project R200 drivers.

  7. Re:PCI-E about features by LankyBoycie · · Score: 5, Informative

    Nope - AGP can go both ways too, this is not a new feature on PCI-Express. PCI-Express is all about replacing PCI and AGP with a common interface.

    Using the host processor "to make on the fly corrections to the image" would be madness as you would have to transfer the whole frame buffer off the GFX card to host mem and then back again. An incredible waste of bandwidth when you can do pretty much most things with pixel shaders anyway, without the round trip.
  8. Re:AGP doesn't make sense by Lumpy · · Score: 4, Informative

    are you on drugs?

    almost ALL Nvidia cards with VGA + DVI do dual head out of the box for $69.00 to $299.00 nothing expensive there... 3 head? easy, just buy a (gasp) PCI card to compliment it.

    matrox makes 4-8 head cards that are sub $500.00 which are in the same price ballpark as the go-fast latest shiny video card that also have great 3d.

    I suggest you learn about what you are complaining about before you publically complain about it... there are GOBS of goodies for super cheap multi-head.

    --
    Do not look at laser with remaining good eye.
  9. Re:I got screwed by lostchicken · · Score: 4, Informative

    You do realize that running a system open-cased actually reduces cooling performance, right? Coses work through airflow, pulling air in one side of the case, flowing through the case and pushing out out the other. Put your case back on. Your CPU (and ears) will thank you.

    --
    -twb
  10. Re:Cheaper prices by JPriest · · Score: 5, Informative
    "ATI's GPUs will be available in both PCIe and AGP flavors throughout 2004."

    I guess even the submitter did not RTFA.

    --
    Saying Java is nice because it works on all OS's is like saying that anal sex is nice because it works on all genders.
  11. Cards don't matter, but chipsets do by brucmack · · Score: 4, Informative

    Many people have pointed out that it really doesn't matter if one has just purchased an AGP card just because PCI Express versions are coming out this year... However, it may be influenced by the chipset support.

    Intel's roadmaps reveal that none of their next-gen chipsets will have AGP support.

    Similarly, SIS' roadmaps reveal that none of their chipsets will have AGP support either. That's for both Intel and AMD processors.

    However, VIA's roadmaps show support for AGP throughout 2004 for both Intel and AMD processors.

    So there's all the major players in the Intel game, and two for AMD. I would theorize that NVidia will go with whatever solution lets them pimp their high-end GPUs most effectively for their next NForce boards, but I don't remember seeing anything official about this. Anyone got a link?

  12. Re:Why? by The+Ego · · Score: 3, Informative

    As another poster stated, this in unlikely to happen.

    High-performance graphic cards will require a x16 slot, and most motherboards will only provide one x16, multiple x1 and maybe a couple of x4.

    Moreover the PCI-Express specs define power limits. All the Gfx vendors requested (and got) amazingly high power limits for graphic slots. Having two Gfx boards working at the limit would blow past the cooling abilities of most cases. While it will be possible for a PC manufacturer to provide multiple such slots, this will not happen in the value segment and may only be offered at a high cost premium (if at all).

    What I am hoping for is for "secondary" cards working from a 4x slot, with limited performance and limited consumption. I could use a (or two) secondary display(s) while using flight simulators (e.g., for auxiliary panels or peripheral vision).

    Note also that the PCI-SIG is close to making a decision on "second generation signaling rate". The debate is between 5Gb/s/lane (ie 2 times Gen I) or 6.25Gb/s/lane. A Gen-II 4x slot would provide enough bandwidth to feed a current high-end card.