[MUD-Dev] Jeff's Rant: A World Full of Wheel-Makers

[Brian Hook]

At 06:23 PM 5/22/01 -0400, Dave Rickey wrote:

> Similar issues arise with software poly occlusion, by the time
> you've decided you don't need a poly, you would have been better off
> to draw It.

While this is sometimes true, a good portaling scheme that does scene
level portal frustum clipping is nearly always a win if your data is
arranged in a hierarchical fashion that allows quick boundary/frustum
checks.

> However, the Kyro II chipset may render that moot, the tile-mode
> rendering essentially provides hardware occlusion at no performance
> hit.

Kyro II has some interesting capabilities, however it's a dollar short
and a day late.  Most modern rendering algorithms have huge amounts of
_required_ overdraw, specifically multipass algorithms that accumulate
the results in destination color or destination alpha channels.  The
current work I'm doing has passes upwards of twenty, most of which are
required and can't be trivially optimized away by deferred rendering
or occlusion testing (and in the cases that I can, the hierarchical
Z-buffers that Radeon and GF3 offer give me almost as big a win just
by doing rough front/back sorting).

> by geometry counts.  Basicly, the Kyro II is the first chipset that
> isn't designed strictly for Quake-type games, corridor shooters, and
> will be a major boon for high-geometry, high-overdraw situations
> like most MMOG's.

No chipset that I'm aware of has ever been optimized for Quake-type
games (and keep in mind that the original Quake didn't support
hardware acceleration until the GLQuake patch was released...most of
Quake's graphical advances were a result of Carmack and Abrash's very
serious software optimizations).

Since Voodoo1 pretty much all mainstream graphic chipsets have simply
tried to draw as many textured triangleas possible.  Fundamentally,
almost every immediate mode graphics accelerator operates the same way
-- take a triangle, draw it with some basic per-pixel operations like
Z-test, alpha test and blending.  This is the age old (back to SGI
workstaiton days) immediate mode paradigm.

Now, a lot of DRIVERS have been optimized for Quake because its
benchmarks are so prevalent when comparing different graphics chips
(and CPUs), but that's simply a matter of prioritizing what code paths
you want to aggressively optimize.

Brian Hook

_______________________________________________
MUD-Dev mailing list
MUD-Dev at kanga.nu
https://www.kanga.nu/lists/listinfo/mud-dev