Beyond '32/32/32'

According to a tip, the "Voodoo3 4000" will deliver final 32-bit color rendering , a 32-bit Z-buffer, and 32 or 64MB of RAM. We've also heard that this part will be able to run dual in 3dfx's famous SLI, but the AGP/PCI matter lends doubt to this conjecture.

V3 4000 by any other name

Speaking to sources close to 3DFX, we've been able to uncover certain facts. The "Voodoo 3 4000" may not the actual name of the product, when released. The official name has not-yet been determined, and we feel it may indeed be changed in order to reflect any new capabilities.

In regards to speed, what could a simple 4X implementation serve for 3dfx? The 166Mhz V3000 is very close in performance to the 183Mhz V3500, and the most important feature justifying the price increase is the LCDfx flat-panel support. Yields for the 183Mhz part are still uncertain, and 5.5ns RAM will remain expensive for some time. You can consider it a given that 3dfx will not be able to bring products with faster core speeds or memory speeds to market without a significant redesign or fab process modification.

What we expect from the V3-4000

That leaves us with 2 logical possibilities. The first is that the Voodoo3 4000 will be a V3 with crippled AGP 4x, with no support for AGP texturing. It's doubtful that 3dfx would attempt to increase the clock speed past 183Mhz, unless expected yield rates suddenly shoot through the roof. If 3dfx were to take this route, it would be more of a marketing/shelf-space move (Voodoo3 AGP 4X!!) than taking advantage of a performance gain.

The second possibility, specifically mentioned by our sources, indicates that the late-Summer slated Voodoo3 4000 will have some degree of architectural changes, at the very least incorporating such new features as final 32-bit rendering, massively multiple multitexturing, singlepass blending, bump mapping, trilinear and anisotropic and more.

V3-4000 Performance

Many have noted 3dfx's reluctance (or inability) to enter the 32-bit market, commonly citing the speed hit incurred by moving to 32-bit. We're seeing that even with next-generation cards such as TNT2, working in 32-bit does bring a 20-40% speed hit, depending on where your bottlenecks lie.

Our source reiterated 3dfx's position that no speed sacrifice is acceptable for 32-bit performance, though many have cried sour grapes (and the truth is likely somewhere between the two stories). For those of us (everyone) wondering what the performance will be on this new card over Voodoo3, our source states that it should be close to the raw speed increase of Voodoo2 over Voodoo1. No mention was made of whether this was for 16 or 32-bit color, or with what settings. For a more subjective take on this, I first asked whether or not, from a pure performance standpoint, whether or not Thresh would prefer this card over Voodoo3. The answer, without a moment's hesitation, was "definitely."