Will Direct3D Conquer All?

Steve Platt

If you follow Internet newsgroups such as comp.graphics.api.opengl, you may have noticed the intense and ongoing debate concerning the relative merits of Microsoft OpenGL, Microsoft Direct3D, and Silicon Graphics' Cosmo OpenGL. The debate, between the principals of the Microsoft OpenGL and SGI Cosmo OpenGL camps but also including other knowledgeable parties, has been intense.

The two OpenGL implementations have very different target audiences. Microsoft maintains that its rendition suits all performance levels, from simple Windows 95 VGA systems to high-performance NT workstations. The software giant offers vendors several approaches to create optimized OpenGL libraries that yield fairly impressive performance. SGI, on the other hand, has targeted W indows 95 systems lacking 3-D graphics accelerators in particular, as an adjunct to its Web applica tions. Cosmo is a software-only solution that SGI believes can surpass software-only Direct3D implementations. It is also cross-platform, enabling a greater degree of integration among Web applications based on the Cosmo line of products.

Despite differences in their views on OpenGL's future, both parties agree that OpenGL has multiple advantages over Direct3D, and that Direct3D has no lock on the future of any PC 3-D market. Despite its high-end history, OpenGL need not sacrifice good performance for rendering accuracy. For example, OpenGL implementations don't have to provide subpixel accuracy, and OpenGL applications don't have to use it. Like game APIs such as Direct3D, OpenGL can obtain high performance by sacrificing details that aren't necessary in a real-time video environment. Programs running under Cosmo OpenGL have provided performance similar to that of unaccele rated Direct3D. We can hypothesize that accelerated solutions would also be comparable.

Programmers find OpenGL code easier to write and debug. OpenGL presents a completely abstract environment; the programmer can assume that an appropriate screen exists and that OpenGL will make a program's graphics requests fit that screen. Direct3D presents to the program (and programmer) all the hardware details; a program must examine them all and fit itself to that set of features. In OpenGL, all optimizations are in the device driver; any OpenGL program will run on any hardware at any resolution. In Direct3D, you can tune a program to particular screen resolutions, but a particular graphics adapter might not support those resolutions.

OpenGL is not dead in the high-performance games arena, and Direct3D is not necessarily the API of choice for future programmers. What remains to be seen is where each group will take its approach to 3-D graphics and how users will react.

---

Steve Platt is manager of electronic publications at NSTL.