Path: utzoo!utgpu!jarvis.csri.toronto.edu!rutgers!usc!cs.utexas.edu!uunet!zephyr!tektronix!tekcrl!tekgvs!keithe
From: keithe@tekgvs.LABS.TEK.COM (Keith Ericson)
Newsgroups: comp.sys.ibm.pc
Subject: Re: Vga Boards, and Seiko's new monitor
Keywords: vga,graphics,windows,video-7,wonder,fastwrite,ati
Message-ID: <5513@tekgvs.LABS.TEK.COM>
Date: 6 Jul 89 07:04:20 GMT
References: <2565@PEDEV.Columbia.NCR.COM> <[304.1]comp.sys.ibm.pc;1@vpnet.UUCP> <1685@netcom.UUCP> <14867@ut-emx.UUCP>
Reply-To: keithe@tekgvs.LABS.TEK.COM (Keith Ericson)
Organization: Tektronix, Inc., Beaverton,  OR.
Lines: 40

In article <14867@ut-emx.UUCP> nghiem@walt.cc.utexas.edu (Alex Nghiem) writes:
>
>Note that even though the ATI VGA Wonder did not pass two tests, it
>got PC Magazines Editor's Choice award because it processed instructions
>much faster that the other Video cards...

I and some co-workers evaluated the ATU Blunder card a couple of months ago.
We concluded that for our uses it was absolutely unacceptable because of its
slow speed (actually, "speed" is the wrong word here.)

So why does the ATI card seem to be so fast in other peoples tests?
Well, I think it has mostly to do with "Does The Test Use BIOS Calls" or
"Does The Test Access The Screen Directly?"  Our tests access the screen
memory directly.  We have a 386-based Smalltalk "bit-blits" (that we
were hoping to market but has been shelved) for one test; we have an
internal-use-only IC layout package and a 386-based version of a popular
circuit analysis package that also plots to the screen directly.  In
each and evey one of these tests the ATI card showed itself to be one of
the slowest cards we had ever used.

HOWEVER: if a "benchmark" uses BIOS calls, and the video BIOS is written
more efficently than the IBM-clone version (and almost ANYTHING would be
better than some we've seen) the performance could be GREATLY improved
over a competitor's product.  Many INT 10 (I think it is) functions are
MASSIVElLY inefficent (for example, saving each and every register,
doing something that doesn't change a single register, then restoring
all of these unaltered registers).  By doing a GOOD job of writing the
video BIOS routines a particular vendor could achieve significant
improvements in speed.

>The processing speed of the VGA card is significant if you use
>graphics intensive programs, such as Xwindows under SCO Xenix. With >a
slow VGA card, Xwindows under Xenix is practically useless.
   ^^^^
Because of the difference in how "slow" is determined, and
because a UNIX driver may very well manipulate the display memory
direclty, I would be EXTREMELY reluctant to link DOS-based speed
measurements to UNIX, X-Windows or any other performance metrics.

kEITHe