[rescue] Vesa Local Bus video cards

Francisco Javier Mesa-Martinez lefa at ucsc.edu
Tue Apr 1 03:24:23 CST 2003

> Marketing. PCI and especially AGP are "inventions" of Intel. When 32/33
> PCI became to narrow for graphics, Intel invented AGP to be the market
> leader in chipsets. Other chip set producers had no AGP and the PeeCee
> crowd wanted that new, fast graphics thing, so they had to go with Intel
> chip sets. Other chip set producers got AGP later, but they had to pay
> $$$ to Intel for licences.... So AGP is a child of marketing for 100%, a
> typical PeeCee thing. Some ceap kludge to make more $$$ instead of
> proper engineering. "Real" Machines like Alphas or HP9k had 64/66 MHz
> PCI for graphics quite early.

Not 100% accurate, AGP shares a lot with PCI, but it is not 100% PCI, it
is more "burst" oriented than PCI, since AGP was designed for very low
count server/client bus designs (i.e. 1 device per bus). Basically, the AGP
is a fast PCI that can transfer on the 2 edges of a single clock (very crude

The reason is that 64/66 PCI slots are EXPENSIVE for a motherboard
manufacturer, a fast 32bit slot is way cheaper... and for all intents and
purposes you can actually get better performance from an normal AGP port
than a fancy 64bit PCI bus. The same applies for the other end, i.e. the
actual gfx board design, it is easier if your interface is narrower... you
are even more constrained when it comes to routing on an add on PCI/AGP

Of course real GFX computers use switched means to interconnect GFX
subsystems and the rest of the system (ala SGI). I think the next
generation of perfieral interconnect as proposed by intel is going to be
point to point switches (PCI express?)


More information about the rescue mailing list