by Shui-Che Lim

Recently, I’ve been fielding scores of questions regarding what sort of upgrades users should be making for 1998. This really increased after my last column "Core Wars: Socket 7 vs. Slot-1" appeared. As if choosing between Socket 7 and Slot-1 weren’t complicated enough, add to this consideration the imminent arrival of 3Dfx Voodoo2 graphics cards, and the road to Upgraders’ Purgatory has widened considerably.

Before I begin this month’s Tech Corner column, I want to tell you what this article is not about. It is not about putting together the Ultimate Gaming System as this particular topic has been done to death already. I decided on a slightly different angle. This article is about how to make informed value decisions to get the most bang for your gaming dollar.

Let’s face it, with enough money you could get the best that current technology has to offer. The problem is that not many gamers carry around the type of pocket change that Bill Gates and Andy Grove sport in their wallets. The challenge becomes how to get a reasonable balance of cost vs. performance.

Also, I’d like to give credit where credit is due. Thanks to Thomas Pabst, of Tom’s Hardware Guide, for the images of Diamond’s Voodoo2 video card and the performance benchmarks that I’ll be using in this month’s Tech Corner. If you want to see what else Tom has to say about Diamond’s Monster 3D II you can go to his First Look article.

Riding the CPU Performance Curve

Prior to the advent of 3D accelerators, the only way to get better performance from the simulations we enjoy was to upgrade to faster CPUs. Even as 3D accelerators first appeared, performance and quality were dubious at best. However, recent hardware based on 3Dfx’s Voodoo, Verite’s V2x00, and nVidia’s Riva 128, to name a few, have shown that a good 3D accelerator is capable of better 3D graphics without sacrificing framerates.

The CPU today still remains one of the most costly components in a system. Consider that in a current Pentium II-300 system, which can be had (on average) for about $2700, the Pentium II-300 CPU accounts for 20% of the total system cost. Meanwhile, the appearance of newer and faster CPUs is occuring at shorter and shorter intervals pretty much assures CPU obsolescence within a year’s time.

Adopting the strategy of staying ahead of the performance curve by using the most expensive consumer CPU that Intel has to offer will put you anywhere between $500-$750 (or more) in the red every time you decide to upgrade. Clearly, using the CPU to ride the performance curve can quickly put you in the poor house.

The CPU is, in fact, too expensive from both a cost and functional standpoint to be used to perform work best left to dedicated 3D hardware. Freeing up this processing power would allow game designers to use the CPU to model more believable, realistic physics-based 3D environments. Thankfully, recent 3D hardware gives the user a strategy to break this vicious cycle once and for all.

Voodoo2 to the Rescue

Perhaps the most talked about 3D accelerator due out by the first quarter of 1998 is 3Dfx’s Voodoo2. Voodoo2 ups the performance ante just as other companies are catching up to Voodoo performance.

Voodoo2 will run 2-3 times faster than the current Voodoo graphics chipset. It mates dual TexelFx chipsets to the PixelFx to improve rendering performance. Furthermore, in SLI mode (Scanline Interleave), you can connect two Voodoo2 cards to roughly double your performance and also get up to 1024x768 resolution with z-buffer enabled (depending upon the amount of total framebuffer memory). These specific performance characteristics will allow end users to escape the seemingly endless cycle of upgrading to the latest and most expensive CPUs.

The Benchmarks

The following benchmarks are taken directly from Tom’s Hardware Guide. These benchmarks are among the most talked about on the Net, and I’ll be using them to make certain points.

The first benchmark shows Glide performance. The new Glide API supports the triangle setup engine that is new to the Voodoo2 chipset. Another nifty feature is that the new version of Glide will allow Voodoo2 to be used with any 2D or 2D/3D card, including cards based on 3Dfx’s Voodoo Rush. This is welcome news for all the people who wished that Voodoo could have been used with the Voodoo Rush architecture.

With triangle setup enabled in the native Glide API, you see that though there is a difference in performance between the Pentium II-300 to the other CPUs, the difference is much smaller than in Direct3D mode. I’ve been told that D3D supports triangle setup, however, the difference in performance between native Glide and D3D suggests that D3D is not utilizing the full power of the Voodoo2 triangle setup engine.

Glide Performance of Voodoo2

The performance of Voodoo2 slides more in Intel’s favor under Direct3D since the hardware triangle support is not up to par with native Glide. The Pentium II-300 exhibits about 75% better performance over the AMD-K6/233 and the Pentium 200 vs. approximately 30% under Glide. Against the Cyrix 6x86MX PR200, the Pentium II-300 better than doubles the performance under both Glide and D3D.

Direct3D Performance of Voodoo2

Under GLQuake, the performance advantage of the Pentium II-300 shows that Quake is much harder on the FPU and would appear to be even more geometry intensive than Turok.

GLQuake Performance of Voodoo2

With Voodoo2’s triangle setup engine enabled, why is there still a difference in performance between the Pentium II and the rest of the processors? The reason is that triangle setup is actually the second part of the 3D rendering process. Before the coordinates making up the vertices of the triangles can be passed to Voodoo2, geometry transforms must first take place to determine perspective, clipping and lighting. Geometry transforms are also FPU intensive. Since Voodoo2 does not handle geometry setup, the host CPU still must do this work.

So far all the benchmarks appear to disagree with my premise of not upgrading to the latest and most expensive Intel CPUs. But then again, appearances can be deceiving.

Go to Part II

© 1997 - 2000 COMBATSIM.COM, Inc. All Rights Reserved.