Canopus got into the business of selling 3Dfx based cards with a real bang. While the competition for Voodoo Graphics-based cards was running hot, most vendors could only compete on the basis of price and software bundle. The Pure 3D entered the fray very late in the game, but with 4MB of texture memory and a TV-Out connector, it was more revolutionary than anything short of the super exotic (and super expensive) Quantum boards.

The Pure3D2 also enters the Voodoo2 market a little later than the competition, and the hardware isn't nearly as revolutionary, but it's numerous little design touches indicate that the engineers at Canopus give a lot of thought about how to make a good design better.

(Please note that "Pure3D" and "P3D" refer to the original Voodoo Graphics Pure3D. References to the Voodoo2 based card will explicitly use "Pure3D2" or "P3D2")

In the box for the Pure3D2, you'll find: the Pure3D2 itself, the requisite installation guide and registration card, a driver/utility CD, a game demo CD, cables for SLI, passthrough, S-Video, composite video, and even an audio cable to run your sound card out to the TV, should you desire. Oh yes, and a nicely made "I Got a Canopus" sticker that you can use to cover up that unsightly Compaq or Dell logo on your case!

The Pure3d2 hardware is very different than the competition from Diamond and Creative Labs in a number of important ways. While others essentially just boxed 3dfx's reference design when developing their Voodoo2 cards, Canopus took the time to look for better ways of doing things.

The first thing that becomes immediately obvious upon first look at the card is that it's actually about the same dimensions as the Pure3d. This is no small feat considering that it carries an extra TMU and twice as much memory. By comparison, the Creative Labs Voodoo2 12MB card is pretty long - at least an inch longer according to Canopus - but it feels like the difference is even greater than that.

The Pure3D2's reduced length can make installation much less of a headache. It keeps its size under control by moving much of the memory and the GENDAC chip to the backside of the card. Technically this is a violation of PCI specifications, but unless the card in the next PCI slot uses very large components, fitting it should be less of a problem. All the same, you'll probably be well advised to try to keep TV cards and other large-component RFI sources as far away on the bus as possible.

The next most obvious difference in the Pure3D2 is the inclusion of an onboard fan - possibly the first offered as standard equipment in a consumer-level graphics card. The fan is self-contained, drawing power straight from the card and blowing directly on a TMU. This improved cooling was sufficient that the default clock setting on the card is 93MHz, just a hair above the typical Voodoo2 board. The board runs reasonably well at 100MHz, too - totally stable on a non-overclocked system, but crashes were frequent with the test card when running the CPU far into overclocked territory. Pulling back to 97MHz fixed the problem, and the card appears to run stable indefinitely at this speed regardless of even extreme CPU overclocking.

An interesting note about the fan: there are no heatsinks used anywhere, which makes you wonder how effective can the unprepared surface of the PixelFx chip be used for convection. Second, the fan blows on the PixelFx chip directly and just a little on one TexelFx chip, but really not at all on the other. Perhaps this is because one normally does most of the work in current games. As a side note, the special Voodoo2 aftermarket heatsinks designed to fit over multiple chips are meant for the reference design and thus might not fit over the chipset configuration of the Pure3D2.

The passthrough on the Pure3D2 deserves special recognition. Any time a passthrough is used on an analog signal, the signal quality gets degraded somewhat, making normal 2d images - especially text - lose sharpness. At the very low frequencies used by 640x480, 800x600, or 1024x768 desktops this degradation is not very noticeable. However, these days many gamers are running on 17, 19, and even 21-inch monitors, making resolutions up to 1600x1200 not at all uncommon.

At these resolutions, even a small signal loss will result in poorer picture quality. Out of the various 3dfx-based passthrough cards, the Pure3d2 has a remarkably low loss - an important factor to consider for the rest of your non-gaming computer use. For the record though, perception of signal loss varies from person to person, so your mileage may vary.

On the driver side, the Pure3D2's control panel is very efficient and nicely thought out, giving easy access to commonly used beginner features, refresh rates, driver version numbers, TV-Out controls, clock speed, Glide and D3D V-Sync wait controls, even the 2MB texture limit feature to help you run older games. In short, it is comprehensive enough that only the real lunatic fringe should feel compelled to resort to the old-fashioned "SST" environment variable settings.

A couple of side applications give substantially increased functionality. Quick Control is a very handy system tray program that lets you adjust the gamma brightness, color settings, and various TV-out settings instantly while you're playing the game. The Application Launcher is like a bookmark holder for your favorite games and software. Just drag the program's icon to the button bar and it will automatically create a new button (complete with thumbnail icon) that you can use to start your program.

The real treat, though, is that you can save custom Pure3D2 or Desktop settings for each icon. This way, you can run each and every game you have with it's own display settings. Putting off playing an older Glide game because it can't handle more than 2MB of texture RAM? Are you unhappy with the shade of your helicopter in Longbow 2? Want to avoid the "click of death" in M1TP2? Set it up on the Application Launcher and each time you click on it, the custom settings for that particular game will be loaded before the game starts.

When you use this method, remember to save your default settings on the provided "Desktop" icon and click that after you're done to normalize things. These apps really fall under the category, "Why hasn't anyone thought of this before?" Canopus has clearly invested some thought in "how to build a better mousetrap".

Click to continue . . .

The game software bundle is, well, curious. You get a demo disc from "Boot" with a number of 3dfx-enhanced game demos, which is certainly preferable to spending long hours downloading or wondering if the download time is worth the effort. But for a card that is so expensive, one would expect to get a little more.

Compare this to Diamond, who throw in games virtually with a shovel. Creative originally went with no software on their 12MB card, but later added a gamepack. Canopus finally joined in, but in this case, the game is an expansion pack for Quake 2 (The Reckoning) and a mail-in rebate for players that decide to buy Quake 2 to make use of it. A rather unusual choice that perhaps reflects the hardcore market that are drawn to Canopus products.

While Winbench3D may be widely used and recognized, some question as to its validity in gaming situations has been raised. "Final Reality" is the hot new benchmark on the 3D scene. Co-developed by VNU European Labs and Remedy Entertainment, this benchmark is designed from the outset to model real -world game performance as closely as possible. Indeed, the figures it provided meshed very closely with those from Jane's F-15.

Tests were conducted on a Celeron system on an Abit BX6 motherboard with PC100SDRAM and run through a wide variety of speeds. Because of this the performance figures may not be totally identical to a conventional Pentium II at the same clock speed - although the critical floating-point performance figures will be close enough for our purposes.

All tests were run with current drivers, 70 or 75MHz screen refresh rates, and no waiting on the V-sync. Final Reality tests were conducted with the soundless advanced benchmark. Tests were repeated at various CPU and 3d chipset speeds, where the 3d cards were run at default, low, and high speeds. (Note that for the Pure3D, the default and low speed are one and the same). Where possible, graphics settings were set to maximum.

Also of note, the tests pitted the Pure3D2 against the Pure3D. However it is worth mentioning that the Pure3D is not your typical 3Dfx card. Carrying twice as much texture memory as other cards in it's class, it is important to remember that it's performance is not exactly typical of the "classic" Voodoo cards such as the Monster 3D, Righteous 3D, and so on.

The 25 pixel test is designed to measure real-world polygon performance. As such, this is closely tied to CPU speed rather than the fill rate, but the Voodoo2 architecture includes both a higher fill-rate and a 50% better geometry setup engine than Voodoo Graphics. The result is that there is a real, though slim difference between P3D2 and P3D. The difference stays around 10% until one gets into the really fast machines. Note that at 400MHz and beyond, the gap widens to roughly 20 to 35%. At these higher speeds, the CPU finally catches up to the P3D and the extra power of the P3D2 firmly establishes its dominance.

The "Robots Scene" is a good baseline test designed to behave a lot like a real game engine that doesn't use much in the way of costly effects - the most dramatic effect used is a little environment mapping on the mechs themselves. Here, the advantage to the Pure3D2 is fairly small, starting at about 7% at 266MHz and just shy of reaching 12% at 416 MHz.

However, it is important to remember that these figures are average framerates over time. The differences are much more pronounced in the most demanding parts of the scene - so momentary stuttering when the hardware demands peak is much less of an issue with the faster CPU or 3D card. An interesting note is that framerates are virtually unaffected by overclocking either the Pure3D or Pure3D2 - in most cases the difference was a matter of a tenth of a frame per second.

The "City Scene" is more complex than the Robots Scene, filled with all kinds of effects, transparencies, lighting, and so on. As such, the advantage is a little more narrow - roughly an 8 to 9% improvement with P3D2. Here it would seem that CPU speed is a larger factor in determining performance.

Now here's where the Pure3D2 really shows off it's colors - fillrate. Filling at an average of 2.5 times faster than P3D, the fillrate stays consistent regardless of CPU or 3D card clock speeds.

Go to Part II