Date: Thu, 17 Feb 94 09:26:27 CST
From: csaba AT vuse DOT vanderbilt DOT edu (Csaba A. Biegl)
To: djgpp-bounces AT sun DOT soe DOT clarkson DOT edu
Subject: Re:  Mach32 driver, Diamond Stealth
Cc: djgpp AT sun DOT soe DOT clarkson DOT edu

	From djgpp-bounces AT sun DOT soe DOT clarkson DOT edu Wed Feb 16 03:18:28 1994
	To: djgpp AT sun DOT soe DOT clarkson DOT edu
	Subject: Mach32 driver, Diamond Stealth


	I'm going to start using DJGPP for simulating some 3D-graphics stuff.
	I'm interested in getting high resolution/bit depth,
	and want to be able to do at least 24bit color at 640x480,
	preferably at 800x600.

	Will an ATI LX card (Mach32 chipset) or Diamond Stealth be able to swing
	this, and will any drivers exist?  (LIBGRX, or otherwise)

	Thanks,

	Matthew

GO32 1.11's built-in VESA driver works with the Stealth. The only ATI ULTRA I
have experience with is the original ULTRA from '92. At that time I was
not very impressed with the VESA support in the ATI BIOS-es. (Their
so-called VESA BIOS did not know about the higher resolution modes of
the card.) The GRX drivers which use these cards in accelerated modes can
only support 256 colors, for higher color depth you have to use 
a "dumb frame" approach. The current GRX release only supports 16 bit
color depth, but you can use Gregory's VESA library for 24 bit color.

If your 3D graphics goes beyond a simple wire-frame or flat polygon
approach (i.e. Goraud shading, ray tracing, etc..) then you don't need
an accelerated card. Your program will have to calculate and display
every pixel individually. In this case what you are looking for is a
card with the fastest bus interface. Many times accelerated cards are in
fact slower than simple "dumb frame" SVGA cards (i.e. ET4000, etc..)
with regard to direct video memory access from the CPU. A few months ago
DJ circulated a video RAM access speed benchmark program and asked the
readers of the group to run it on their systems. Check the archives of
the mail group for the results.

If you want the best performance, you probably should not use any of the
graphics libraries in your rendering code. Calling 'GrPlot' 1024x768 (or so)
times can add quite a bit of overhead. Using GO32's linear VGA memory
mapping you can quite easily inline your pixel set routine. (Of course
you cannot run under DPMI.) If you want to use 24 bit color, you also
may want to look for a card which pads the pixels to 32 bits. (It is
much faster to write a single long than three bytes.) One way to find
out is to run the VESAINFO program from the GRX 1.03 distribution on a 
machine which has the card you consider buying and check whether
<bytes per scan line>/<screen width> is 3 or 4.

Csaba Biegl
csaba AT vuse DOT vanderbilt DOT edu