|
Author: Tom Williams
Source: Computer Design Vol. 27, No. 16 01 Sep 1988 (pages 36-40 physical)
With the inexorable force of flowing lava, the 8514/A graphics standard
oozes from the depths of IBM-regardless of whether the world wants it. The
questions "Is it good? Is it appropriate?" pale next to the fact that it's
there and third-party graphics hardware and software manufacturers are rallying
to support it. They're also working to improve its performance and price.
Conceived by IBM as an option that would allow the high resolutions needed
by CAD applications to run on its Personal System/2 family, the 8514/A offers
1,024 x 768 pixels at either 4 bits (16 colors) or 8 bits (256 colors) per
pixel. The 8514/A chip set makes up a drawing processor that takes commands
from the host CPU to update pixels in its frame buffer. This is a departure
from earlier IBM graphics modes-Color Graphics Adapter (CGA), Enhanced Graphics
Adapter (EGA) and Video Graphics Array (VGA)- in which the host had direct
access to the frame buffer's memory map.
In addition, the 8514 display specification has two modes: 1,024 x 768
pixels interlaced, and the VGA display mode of 640 x 480-pixel non-interlaced
scan. Although the 8514/A controller doesn't run VGA graphics software, it can
input VGA signals via a special-feature connector and display them on a
multiple-frequency monitor at the VGA resolutions.
IBM hasn't released specifications on its controller hardware but has
published a document describing the Adapter Interface, or AI, the software
interface to the 8514/A silicon. The AI is the interface to which IBM
recommends all third-party software developers write their code. IBM's stance
forces hardware and software developers who want 8514/A functionality to make
an important decision: do they follow IBM's recommendation and write to the AI,
or do some daring hardware companies try to reverse engineer the silicon so
that software can be written directly to hardware registers? Additional issues
revolving around 8514/A involve devising ways that let the earlier applications
written to EGA and VGA work with 8514/A hardware and lowering the cost of high
resolution monitors needed to meet the display requirements.
"Hardware Compatibility-is it useful? Maybe, but probably not."
– Walt Penny, Media Cybernetics
Reverse Engineering Tempting
The question of writing directly to the hardware is probably the most
crucial. Reverse engineering worked well with EGA and VGA, and a large number
of companies are building boards using register-compatible chips supplied by
five or six IC manufacturers. It appears tempting on the surface. "In all
cases, you'll get better performance by writing to the hardware," says Walt
Penny, vice-president of engineering for Media Cybernetics (Silver Spring, MD),
maker of a graphics toolkit called Halo, which has found wide acceptance on
current graphics hardware. The only problem is that IBM has warned that it may
change the silicon underlying the 8514/A AI at some time in the future, in
which case software written to earlier silicon wouldn't work.
"IBM said that about everything CGA, EGA and VGA," asserts Greg Reznick,
director of marketing for Video Seven (Fremont, CA), whose company currently
provides a VGA board using a register-compatible chip and is now developing an
8514/A chip set. Video Seven is banking on the assumption that IBM's warnings
will turn out to be as empty as they've been in the past.
Others, however, aren't so sure. Media Cybernetics' Penny, for example, says
that this time he's taking IBM seriously, noting that the Entry Systems
Division-producers of PCs and PS/2s - has recently been more closely tied to
the corporate center. And IBM's corporate tactic for dealing with
plug-compatible manufacturers that are trying to emulate its hardware has been
to keep changing the hardware.
Verticom (Sunnyvale, CA), which is readying a family of 8514/A-compatible
products, is also being cautious. "We're probably at a point where the warning
is going to go teeth in it," says Robert Dickenson, president and chief
executive officer at Verticom.
AI Has Serious Flaws
Despite disagreement over whether IBM will change the silicon underlying the
8514/A AI, just about everyone agrees that the AI as it currently stands is
seriously wanting. Video Seven's Reznick calls it a "not-too-far- removed
plotter driver." Indeed, while the 8514/A AI specification does include bit
block transfer (bitblt) commands, it's heavily oriented toward line drawing and
has quickly gained a reputation for being poor at large block transfers.
F. Stephen Andes (center), president of Enertronics Research,
discusses an image generated by the company's Aurora 1024 graphics board that
implements the IBM 8514/A Adapter Interface (AI). Enertronics is circumventing
the AI's limitations by writing its own hardware drivers for popular graphics
interfaces.
Furthermore, some criticize AI's lack of a full suite of graphics commands.
"AI isn't full-featured," claims F. Stephen Andes, president of Enertronics
Research (St. Louis, MO), producer of the Aurora 1024- one of the first 8514/A
boards for the AT bus. "Even in CAD/CAM applications, we wish it had an arc
routine and a circle routine." And Media Cybernetics' Penny, comparing the AI
to Hitachi's HD63484 advanced
CRT controller and Intel's 82786,
says, "The AI doesn't do a lot of the things that are native instructions in
other graphics processors. We wrote the Halo toolkit to the AI because we had
no choice."
Given such dissatisfaction with the features and functions of the AI, it's
no wonder that third-party vendors are divided into two camps. One camp is
trying to improve the AI while staying within the safe fold of striking out to
reverse engineer and write directly to the hardware, the risks be damned.
Indeed, Microsoft (Redmond, WA) itself has gotten the 8514/A hardware
specifications from IBM. Its sole purpose is to write Windows and Presentation
Manager drivers-not drivers for application programs-to the 8514/A silicon
because the performance simply wasn't acceptable when the drivers were written
to the AI.
But what happens if IBM does change the hardware? "The driver for
Presentation Manager [and Windows] will have to change," says Verticom's
Dickenson.
Video Seven, which claims to have successfully reverse engineered the 8514/A
chip set, will be offering the specifications under nondisclosure to selected
software companies so they can write both to IBM's present hardware and to
Video Seven's yet unannounced hardware. This news hasn't exactly set off a
chorus of cheers, however. "Can Reznick prove it's 100 percent compatible? It
can't be 99 percent," cautions Kathleen Hunter, marketing communications
director for Enertronics. And Media Cybernetics' Penny wonders, "Hardware
compatibility-is it useful? Maybe, but probably not."
Working Around Compatibility
Graphics-based user environments, such as Windows and Presentation Manager,
and toolkits for building graphics applications, such as Media Cybernetics'
Halo, have already adapted to a lack of hardware compatibility. Media
Cybernetics has written a version of Halo to the AI, but is also preparing a
version that directly accesses not the 8514/A hardware but the
TMS34010 graphics processor from
Texas Instruments (Dallas, TX). This is significant because the initial
third-party 8514/A products appearing on the market such as the Aurora 1024
from Enertronics, the Cobra/2 from Vermont Microsystems (Winooski, VT) and the
MX series from Verticom - implement a version of the AI but use the 34010 as
the under lying hardware.
Large software vendors who provide system-level graphics environments such
as Halo and Windows want to get as close to the hardware as possible because
most application developers will want to use a toolkit of some kind rather than
writing in terms of low-level graphics primitives. This makes it vital that
such toolkits be as efficient as possible. Media Cybernetics, for example,
worked hard to write Halo to the AI because there were certain things that Halo
does that aren't directly supported by the AI's functions. These Halo
operations had to be worked out by the CPU in terms of the limited AI command
set rather than by direct calls.
According to Penny, Media Cybernetics is also preparing a version of Halo
that directly accesses the 34010 hardware, which would mean that applications
written with the Halo toolkit could run on 8514/A boards based on the 34010
without having to go through the AI.
Even if software vendors cover both 8514/A and 34010 bases, hardware
compatibility may win in the long run. For instance, Verticom has let it be
known that while its initial hardware offerings will use the 34010, the company
will be offering a higher performance version of its boards, which it will call
the HX series. The HX series will incorporate the AI but will use proprietary
VLSI—not a reverse-engineered 8514/A chip set — to carry out 8514/A
functions. Verticom will supply drivers for Windows and several other programs
written directly to this hardware, but will rely on both its AI implementation
and its special chips to increase performance of other software applications
written to the AI.
Adding 8514/A to the AT
IBM intended that the 8514/A display adapter be used only on its PS/2 line
of computers with the Micro Channel bus. Third-party vendors are producing
8514/A display cards for both the Micro Channel and the AT bus. This clearly
shows the intention of third-party vendors to preserve the ability of the
installed base of 80286- and 80386-based, AT-bus machines to run graphics
applications intended for the PS/2 and its OS/2 operating system.
The Video Graphics Array (VGA) either occupies its own IBM
PC AT bus slot or, increasingly, is incorporated on the system mother board. In
Personal System/2 systems, it's always on the mother board. A monitor attached
to the 8514/A can display graphics from either the VGA or the 8514/A. If the
VGA is being displayed on the 8514 display monitor, its palette contents are
loaded into the 8514/A's palette via the mode switch. It's also possible to
attach a lower resolution VGA monitor directly to the output of the VGA
palette.
But this raises a partitioning issue. The PS/2 implements VGA on its mother
board, and future models are expected to have the 8514/A migrate to the mother
board as well. And the newer versions of AT clones and 80386-based AT-type
machines are beginning to show up with VGA on the mother board too. For such
machines with VGA on the mother board, adding 8514/A capability is a matter of
plugging in the card and cabling between the VGA feature connector on the
mother board and the 8514/A card. The 8514/A card contains a mode switch that
can pass VGA graphics through to the monitor, where they're displayed at 640 x
480-pixel resolution.
For older machines, though, it will usually be necessary to take up two bus
slots: one for the VGA card and one for the 8514/A. Enertronics offers an
optional VGA card bundled with its Aurora 1024 for those who don't already have
VGA. And Video Seven is looking at the possibility of a VGA daughter card that
could attach to the 8514/A without taking an extra slot. Verticom's entry-level
MX cards for the AT bus will include VGA capability on the card, but are viewed
as transition products.
Another issue regarding the 8514/A is that of cost for the required high
resolution monitor. When one moves from 640 x 480 to 1,024 x 768 pixel
resolution, the price of the monitor needed for display takes a quantum jump
from less than $1,000 to around $3,000. This price differential represents a
barrier to wider acceptance of 8514/A, although there are indications that
prices will soon come down. For this reason, says Verticom's Dickenson, "For
the next 12 to 24 months, the high-resolution market is still a vertical
one."
It's possible to use a multisynchronous monitor such as the NEC Multiscan
with an 8514/A card and get the graphics performance it offers (a definite
improvement over VGA) but at the VGA screen resolution. This may provide an
impetus for the three-element feedback loop that Dickenson sees as essential
for lowering prices: standards, software and low-cost monitors. "The critical
thing for something being mainstream is that it has to be accepted as a
standard, and then it feeds on itself," he says. Software developers have
something to write to, and the existence of applications presents a market to
which monitor manufacturers can sell at higher volumes and, hence, reduced unit
prices.
"The 8514/A isn't the ideal controller for shading operations, and it's not
very fast."
– George Krucik, Autodesk
Whatever criticism one may have of the 8514/A, it still represents a defined
standard, if only because it comes from IBM. Large software vendors such as
Autodesk (Sausalito, CA), which produces the AutoCAD drafting program, are
supporting it, but they're also trying to influence IBM to extend the AI. "The
8514/A still isn't the ideal controller for shading operations, and in its
current incarnation, it's not very fast," says George Krucik, manager of future
development for Autodesk. But he adds that IBM has solicited comments from
Autodesk about improving performance characteristics and may implement some of
those suggestions.
The big question is, Will those improvements come in the form of extensions
to the AI, changes in the silicon or both? Until an answer begins to emerge,
the prudent software developer would do well to stick with the AI, for all its
present faults.
|