Article Addendum: 'Balancing In Three Dimensions', April 27, 2000

By Brian Dipert, Technical Editor -- EDN, 4/27/2000

This'll be a 'living' document; I'll periodically update it with new hardware and software that I evaluate, plus additional links that I come across. So occasionally check back, and drop me an email with any feedback. Thanks!

April 27, 2000

Hands-On Project

Unfortunately, I've no results to share with you at the moment; Intel's got me at a standstill. There's a statement in the iPeak Graphics Performance Toolkit licensing agreement that precludes users from publicly disclosing analysis results. Intel has historically been somewhat lenient about enforcing this clause, but apparently they've had a change of heart and adopted a tougher stance of late (I'd like to think that they're not just picking on me personally). Not wishing to have a flock of Intel lawyers descend on my front porch, I'd decided to override my normal 'it's better to seek forgiveness afterwards than to seek permission beforehand' style and request formal authorization to use iPeak as my analysis platform.

My request is still winding its way through a maze of Intel lawyers, marketeers and PR types. Frankly, my motivation is pretty low to do yet another 3D WinBench, 3D GameGauge or 3DMark study; plenty of folks have already done this, and by following the links below you can see the results. Plus, iPeak is a much more meaningful analysis tool, for the reasons mentioned in my article. Stop back here periodically and you might yet find some iPeak-derived numbers and commentary (yes, I know, I'm a perpetual optimist). My article included a list of the graphics boards I've got in-hand and hope to test, with more to follow (based on 3dfx's VSA-100, ATI's Radeon 256 and Nvidia's GeForce 2, among others). You might also be interested in the test systems' hardware and software specifications below.

Before you ask; my choice of the VC820 and OR840 motherboards had little to do with their Rambus main memory (though some of the workstation graphics companies had expressed concern that PC100 or PC133 SDRAM might polygon-, texture- and command-starve them). More important, though, was stable, full-featured AGP 4X support in the chipset, and I'd heard quite a few horror stories about Acer Labs, AMD, SiS and Via chipsets. These problems are likely solved, or at least being solved, but at the time I had to make my decision the i820 and i840 chipsets were the only games in town. This reality also precluded use of Athlon processors.

I'd like to thank the companies who provided hardware and software for this study, particularly IBM and Maxtor for the hard drives, and Kingston Technology and NEC for the currently-very-rare PC800 RIMMs. I'll also be using this computer platform for future hands-on projects, such as my late-May memory card benchmarking study and late-July digital audio codec analysis. And note the multiple operating systems; I'm currently running only Windows 98 SE but thanks to PartitionMagic, I plan to soon also put the other operating systems on the hard drive and will let you know my results here, too.

Benchmarking Programs

Should you want to take a stab at doing your own analysis, here's a few good places to start:

• 3DMark
• 3D WinBench
• GameGauge
• Indy3D
• iPeak Graphics Performance Toolkit
• SPECglperf
• SPECviewperf

While on the subject of benchmarking, I want to clarify a point I made in my article. Triple-buffering postpones, but doesn't ultimately eliminate, the artificial performance-limiting quandry that turning Vsync off obviates. The problem, briefly, is that a fast-enough graphics accelerator may complete rendering of the next frame into its back buffer before the RAMDAC and monitor have finished displaying the front buffer. At that point, the graphics chip stalls, waiting for the vertical sync signal that signifies front buffer end-of-display so that it can flip next-scene back buffer contents to the front buffer and begin rendering the next frame.

Triple buffering (creating multiple back buffers) gives the graphics accelerator more work to do per each frame displayed. But, especially at low resolutions and with low quality level settings, today's highest-performance chips are still Vsync-limited even with triple buffering turned on! Benchmarking with Vsync off, thereby removing the display refresh rate bottleneck, gives the most accurate picture of a chip's frame rate performance. However, the display 'tearing' it creates usually makes it an invalid setting in normal usage.

Conferences

The best places to find out what's new and popular in graphics:

• 3D Conference & Expo
• ACM Siggraph
• Comdex
• Game Developers Conference
• Intel Developer Forum
• Microsoft Windows Hardware Engineering Conference

Newsgroups

Scan the newsgroups below and I warn you, you'll wade through a lot of junk; wacky rumours with no basis in fact, tedious debates on the relative strengths of Quake III Arena versus Unreal Tournament or 3dfx versus Nvidia, and long-winded off-topic rants on the issues of the day, not necessarily even related to computing. Occasionally, though, there'll be a nugget or two of wisdom. The excited ramblings of some 13-year old game afficienado who's been able to squeeze an extra 0.1 frame per second out of his latest graphics card (already running over a hundred frames per second, of course) via some obscure registry hack also make for a good humour break at the end of a long work day.

Quite a few fanatical computer users regularly post, and you'll also encounter analysts, journalists and even a few lurking graphics hardware and software developers (particularly in the comp.graphics and microsoft.public.win32.programmer.directx newsgroups, which in general have a more professional tone than the others). Keep in mind, when you review the benchmarking results you'll find here, that they sometimes come from overclocked PCs. Performance gains may derive in part or in total not from the graphics subsystem itself but from non-standard system configurations, such as faster-than-PC133 SDRAM or an 89 MHz AGP bus (resulting from the popular SE440BX chipset being run at 133 Mhz). You'll also frequently see postings based on beta driver releases, which may or may not ever make it to production.

• 3dfx.products.voodoo2
• 3dfx.products.voodoo3
• alt.comp.periphs.videocards.ati
• alt.comp.periphs.videocards.matrox
• alt.comp.periphs.videocards.nvidia
• comp.graphics.algorithms
• comp.graphics.api.opengl
• comp.graphics.misc
• microsoft.public.directx
• microsoft.public.win32.programmer.directx

Publications

Don't already have enough to read? Here's some more journals to add to your pile:

• ACM Siggraph Computer Graphics
• 3D Design
• Computer Gaming World
• Computer Graphics World
• Game Developer and Gamasutra
• IEEE Computer Graphics
• IEEE Multimedia
• IEEE Spectrum
• Multimedia System Design archives
• PC Magazine (particularly the yearly December 1 graphics roundup issue)

Websites

And finally, here's a smorgasboard of websites, some hit-and-miss, others very solid, that'll further expand your graphics knowledge

• 3dfx developer site
• 3Dlabs developer site
• 4th Wave
• AnandTech
• ATI Technologies' developer site
• Cahners MicroDesign Resources, specifically Peter Glaskowsky's 'A consise review of 3D technology'
• Jon Peddie Associates
• InQuest Market Research, specifically Bert McComas' '800Mhz platform analysis'
• Matrox Graphics developer site
• Mercury Research's The Meter, specifically their '3D graphics technology primer'
• Nvidia developer site, with a whole lotta demos, code samples, documentation......
• Quantum3D's 'Selection criteria for 3D graphics chipsets in advanced real-time graphics subsystems'
• Sharky Extreme
• Thresh's FiringSquad
• Tom's Hardware
• Tweak3D
• Voodoo Extreme and The Reverend's Pulpit, which they host.