EDN — 05.23.96 editorial

Editorial: May 23, 1996

Steven Leibson,
Editor In Chief

Zigging instead of zagging

Being successful at design is very much a task of picking technological winners—but sometimes you zig when you should zag. A case in point is the display-card technology I used in my All-Star-PC article series published in 1990. Back then, Microsoft Windows 3.0 was on the market, and Windows 3.1 had yet to appear. Everyone knew that graphical user interfaces (GUIs) were important—they were standard on Unix workstations and Apple Macintosh computers—but GUIs were not yet common on PCs. Consequently, no standardization had yet occurred for PC graphics accelerators.

On the All-Star-PC project, I looked at what I thought was a sure bet for graphics acceleration: TIGA. This technology was Texas Instruments' (TI's) solution to the problem of slow PC graphics. TI developed a line of graphics processors, the 340 series, to satisfy a wide range of graphics tasks on PCs and other applications. Lacking Windows' standard interface to graphics, each application required a special software driver. NEC and Hewlett-Packard both offered TIGA-based graphics cards for the PC, and I tested both during the All-Star-PC project.

As we all know by now, processor-based graphics acceleration in the Windows environment did not win the technology derby. As Windows 3.1 became the dominant PC GUI, it pushed out all other processor-based graphics accelerators, because Windows' graphics routines weren't written to give a graphics processor much responsibility. Windows 3.1 assumed that the host CPU would perform almost all of the graphics processing. Graphics coprocessors became an expensive overkill.

The technology that did win was "Windows acceleration." Graphics-card vendors took a careful look at the functions Windows would allow an accelerator to perform, and the vendors implemented just those functions in their accelerator silicon. Windows acceleration evolved in stages, starting with bit-block line transfers and absorbing the color-lookup DACs. It then progressed to video acceleration, 3-D acceleration, and texture mapping. These evolutionary steps occurred roughly in lock-step with development of the software required to exploit the benefit of hardware acceleration.

The point of this story is that all technological developments take many branches (ac vs dc, AM vs FM, VHS vs Beta, etc). Some branches die, some coexist, and others take off. Your success at system design often depends on choosing the right branch, or switching to the right branch as soon as the need becomes apparent. In 1990, the Windows-accelerator branch had not yet started to bud, so I couldn't pick it. However, if I had planned to remain in the PC-based workstation-design business, I would have been well-advised to switch branches as soon as I perceived the change in direction.

That switch would have been hard for me, because I really prefer the graphics-coprocessor approach. It makes sense to me to offload the main CPU as much as possible to let it execute application code. However, in the overall design of today's PC-based workstations using Windows, graphics coprocessors don't make sense. So, don't let your emotional preferences mislead you. Stay on top of developments, and coldly re-evaluate your choices to keep on the right course.

Steven H. Leibson
Editor in Chief