On Fri, Feb 11, 2005 at 03:50:32PM -0500, Jack Howarth wrote: > Dave, > Well now I am thoroughly confused. I had been told that the > DRI support of PCI-E cards was done using the agpgart code in > manner similar to how pcigart support was. pcigart doesn't use agpgart. The two are completely separate. > If this isn't the case doesn't it imply that any DRI support > for PCI-E cards will be software based? Only if the driver doesn't recognise/support the PCI-E cards. > That hardware DRI doesn't really exist for PCI-E cards and thus they will be > a major step backwards from the existing DRI/agpgart rendering with AGP > based cards. hardware DRI doesn't necessarily need AGP. There have been several PCI cards that work just fine with DRI for example. You seem somewhat confused, I'll try and explain the state of things. First, take for example, a old PCI Radeon. Nothing to do with AGP, so for handling on-card textures, the driver falls back to using pcigart. This is kinda sorta like an agpgart, but not quite as fast. It's also card & driver specific (ie, the code lives inside the radeon driver). Then, someone invented AGP Radeons. Same deal, only now, we get to poke bits in the chipset northbridge to allow the graphic card to address system RAM directly. (This is AGPGART). Next up.. PCI-E cards. This is sort of a hybrid of both AGP/PCI GART. The actual GART remapping is done on the graphic card, so the chipset doesn't need to be poked (hence, no agpgart driver needed, and everything is self-contained again in the radeon driver [well, the gfx hardware itself does most of the work in this case]). Hope this clears things up a little ? Dave
