EDN Access

 

September 11, 1998


Schizophrenic in Sacramento

Brian Dipert, Technical Editor

In a previous column (Editorial Taoism, EDN, August 17, pg 40), I talked about the balance that I strive to achieve between differing perspectives on vendors, products, and design approaches. To hammer the point home, I'd like to attempt a split-personality discussion of one of the hottest debates on my editorial beat—whether the geometry and lighting stages of 3-D graphics belong in software that runs on a PC's host CPU or in floating-point hardware that's part of the graphics controller. I could have chosen any one of countless other examples: application-specific or more generic flash memory for data storage; commodity versus differentiated DRAM; FPGAs or CPLDs; ASICs versus programmable logic versus hybrid devices; schematics or HDLs. The point is, there are no easy answers in this business, and if you think otherwise, you haven't done your research.

Personality A:
Wintel is good for PC 3-D graphics

I'm pleased by Microsoft's decision to delay support for hardware-accelerated 3-D geometry and lighting at least until the release of DirectX 7, due next year. Microsoft and Intel (plus AMD, Cyrix, and IDT) understand that not only system performance, but also system price, is important to consumers. Software running on the host CPU is ultimately the cheapest means of performing a given function. It happened with soft modems, it's happening with soft DVD, and it should continue to happen with software-based geometry and lighting.

Of course, high-end workstations do perform geometry and lighting in hardware; they use either custom chips or off-the-shelf components, such as 3DLabs' Glint Delta and Gamma or Fujitsu's FGX-1. But look how much those systems cost! There's no way that the complex floating-point engines required to do hardware acceleration could hit a price point low enough to penetrate the mainstream PC market.

Even if a vendor could squeeze hardware geometry and lighting onto its chip or into a multiple-chip set, today's end user probably wouldn't see a performance boost. Most of today's 3-D software doesn't even come close to saturating a Pentium-II processor's ability to generate processed polygons. By the time such software exists, both the AMD/Cyrix/IDT 3DNow! and the Intel Katmai MMX2 instruction set-based processors will be in production. The SIMD floating-point engines on these CPUs will be ideal for geometry and lighting tasks. Anyway, if not 3-D, what else would the software use 3DNow! and MMX2 for?

A number of silicon manufacturers have seen their PC business evaporate as hardware functions move to software. Some have adapted and survived; others have disappeared. Nobody promised them an easy ride. 3-D geometry and lighting is already a software-based function; why buck the trend and move it onto hardware? Microsoft has more important things to work on.

You can reach Personality A (Technical Editor Brian Dipert) at 1-916-454-5242, fax 1-916-454-5101, edndipert@worldnet.att.net.

Personality B:
Wintel is bad for PC 3-D graphics

I'm disappointed by Microsoft's decision to delay support for hardware-accelerated 3-D geometry and lighting at least until the release of DirectX 7, due next year. This delay is clearly a result of paranoia within Intel (plus AMD, Cyrix, and IDT) that software developers will run out of uses for increasing CPU MIPS. Cheap PCs won't sell if they don't provide a robust end-user experience. Think about it: soft modems, soft DVD, and now software-based geometry and lighting. Wintel has an interesting interpretation of the words "balanced PC."

Of course, open GL-based workstations (and even some high-end Wintel PCs) do perform geometry and lighting in hardware, and just look at how incredibly well they perform! I don't buy the cost argument; Fujitsu's FGX-1 is less than $20 in volume in a market with very few players. Combine several dozen aggressive competitors with a few 0.25-micron-lithography-capable foundries, and amazing things will happen. Intel, the originator of Moore's Law, should know this.

Developers write to the largest possible installed hardware base; that's why software always lags behind hardware. What motivation does a software vendor have to write a 3-D-rich application if the Pentium-II processor is the bottleneck? Add some hardware assist, and then let the Pentium II hand off unprocessed polygons and focus its muscle on higher-leverage tasks; then everybody wins.

SIMD floating-point engines may work well with certain audio- and image-processing tasks, but it's not clear they're ideal for 3-D geometry and lighting. Put too many software-based functions together (3-D plus video plus modem-based Internet access) and the PC will choke. Anyway, host CPUs are supposed to be there for multiple-application flexibility, not to execute standardized, repetitive tasks.

Continued reliance on software-based geometry and lighting is nothing more than CPU vendors' attempts to gobble up more system silicon. Workstations do full hardware 3-D acceleration, yet somehow software developers keep finding uses for high-end CPUs. Microsoft has plenty of resources; it could pull off DirectX hardware support if it wanted to.

You can reach Personality B (Technical Editor Brian Dipert) at 1-916-454-5242, fax 1-916-454-5101, edndipert@worldnet.att.net.


| EDN Access | Feedback | Table of Contents |


Copyright © 1998 EDN Magazine, EDN Access. EDN is a registered trademark of Reed Properties Inc, used under license. EDN is published by Cahners Business Information, a unit of Reed Elsevier Inc.