Some random ranting follows: *g* > maybe using main memory as framebuffer memory like i810 does for example, > is the key here.
Yeah, but that is only an option, if you have a hell of a bus going to and from that memory. IIRC SGI does/did it that way. However a few hours ago, I talked to a friend of mine, who'd been laughing his ass off today, that his new workstation at the university featuring an 1.2GHz processor, but a decent graphics subsystem was considerably faster than what most colleges got out of 2.x GHz boxes with SHM-Graphics onboard. It's a pretty simple calculation: They are typically using 1280x1024x32, for that's what the native resolution of the LCD screens is. Now if you do that at 75Hz (I know, silly for normal LCDs, but nevertheless common), you get: 393216000 Bytes/second that you need from the main RAM. aka 375 MB/sec. That means eating up something between 14 and 40 percent of total memory bus capacity just for refreshing the display. Sometimes I really wonder what bullshit people are willing to buy, because marketing tells them to. Pretty similar for simple IDE-"Raid"-Controllers - c't tested those a few months ago, and many performed worse than just doing it in Software ... Or IDE itself. They are putting Quad-controllers in, give silly cabling length restrictions, just to be able to do something SCSI does easily (many devices, high speed) and continue selling cheap IDE disks for the mass market and high-priced "professional" SCSI Disks ... Enough ranting ... CU, Andy -- = Andreas Beck | Email : <[EMAIL PROTECTED]> =
