Lies, Damn Lies and Benchmarks
Getting accurate and meaningful benchmarks is a problem with many videocards. Back in 1998, a representative from Matrox defended its Millennium G200's laggardly performance on the Final Reality benchmark, by claiming that it was a bug in the Final Reality software. The test program, she explained to me, always defaults to the highest bit depth when performing its 2-D tests. Thus, a card that supports 32 bits-per-pixel, such as the G200, has to push around 4 bits per pixel, whereas a 24-bit card only has to move 3bpp. Thus, Matrox maintained, an 8-bit card would always perform best in that test. And, once they found out that we planned to include FutureMark's benchmark results in our test suite, they refused to send the test lab of the largest computer magazine in the country (of which I was the editorial director at the time) a card for review. In short, they only sent review products to reviewers they already know would give them favorable press.
Another problem occurred a few years ago, when manufacturers deliberately designed the video BIOSes of some cards specifically to excel at ZD's often-touted WinBench benchmark tests -- sometimes even going as far as hard-coding the text strings the tests used into the ROMs of the cards. This produced absolutely no real-world benefit, but it made the card look good in numbers comparisons.
More recently, as noted on Slashdot, it has become clear that people who benchmark by redrawing faster than the display refresh rate are missing the point. The question is how much you can draw per frame time, not how many times you can redraw the same simple scene in a second. That measurement stopped being meaningful when display boards became fast enough to draw useful screens in less than a frame time.
Therefore, it is much more important to evaluate cards in terms of real-world performance. We value -- as you should -- trouble-free setup and compatibility, at least as much as we do raw performance.
AGP and Older Systems
Our experiences during 1997 and 1998 using AGP graphics cards in Super7 motherboards (noted in our motherboard archives, elsewhere on this site) running AMD K6-2 chips and the like suggested that AGP just barely worked in the Socket 7 environment. The display from the Permedia II-based Leadtek WinFast 2300 card we tested sometimes stayed blank after a reset or reboot, and i740-based cards on non-Intel chipsets such as the Via Apollo MVP3 and ALi Aladdin sometimes didn't work at all. (A seemingly endless stream of "4-in-1" updates at Via's website sought to address this issue -- see our Q+A report on Socket7 and AGP for details.) Even the most popular cards of the era, including the best cards manufactured by ATI suffered from laggardly 3D performance and poor rendering of advanced graphics modes that made them less attractive for those with game-playing interests. Fortunately, companies such as the now-defunct 3dfx were selling add-on accelerator cards that greatly boosted 3D performance. Of course, now everybody cared about 3D speed. The business world embraced ATI and Matrox as safe choices for general-purpose use.
3D Goes Mainstream
But with the dramatic price drops and computing power progress made during the last few years, entertainment titles became more demanding of raw 3-D horsepower and average users came to expect a better degree of 3D acceleration in their home systems of mainstream systems. Nvidia steadily gained market share by intently focusing on 3D performance and, it should be noted, resorting to some pretty dubious marketing tactics.
A prime example was NVidia's pre-release claims about its TNT chip. Before it delivered the graphics accelerator, the company's marketing dept. touted its ".25 micron" die-size and a 125MHz clock speed. After it was released without these attributes, the company had already benefitted from plenty of breathless hype from news-hungry hardware enthusiast sites. And average consumers, it seemed, didn't notice that it was actually a hotter and slower .35 micron fabrication, running at 90 to 100 MHz. And what about that fabulous 250 million pixels-per-second speed that early marketing brochures boasted about? Post-release, the company admitted it only pushed about 190 Mpps. Nevertheless, the TNT and its successors went on to become the preferred 3-D gaming cards for the remainder of the nineties and, although ATI has kept itself in the game with its powerful Radeon 8500 series of cards (and Matrox is making a bit of a comeback with its Matrox Parhelia 512), Nvidia cards continue to be the preferred by many as the preeminent gaming graphics solution.
Send comments about this article, disagreements, or your own picks and pans to Gra...@PCBuyersGuide.com
For Further Reading
Post new comment