Troubled 3D graphics hardware developer 3dfx is to buy Gigapixel in a stock swap that values its smaller rival at around $186 million. The deal will see Gigapixel shareholders offered nearly 15.6 million 3dfx shares valued at around $40 apiece - rather more than the $13 or so they're trading at these days.
Senior staffers at 3dfx said the acquisition was made for two reasons: first, to accelerate the company's graphics chip roll-out cycle, which has been knocked way out of kilter by the much longer-than-expected development of the VSA-100, powerhouse of the upcoming Voodoo 4 and 5 cards. At the same time, 3dfx wants to break into the lower-end consumer electronics market: the set-top boxes that so many players are pinning their hopes on, albeit without any real evidence that there's a market for them.
At the high end, Gigapixel role will be to provide technology that will allow 3dfx's post VSA-100 chip to leap-frog a stage in the evolutionary process, in other words to get 3dfx to the point it would have been had the VSA-100 not been so delayed. Gigapixel's approach to 3D graphics is in marked contrast to 3dfx and its main rivals, Nvidia, S3 and ATI. Instead of rendering images directly from a vast number of polygon vertices, Gigapixel's Giga3D system breaks the display frame down into a grid of tiles, and renders each in order, skipping tiles that don't need updating.
The upshot is that graphics cards based on Giga3D need far less memory than current cards, and card manufacturers can get away with slower RAM into the bargain. By working on a tile-by-tile basis, far less 3D data needs be sent across the bus to the card, so the technology's system bandwidth requirements are rather smaller than those of tradition rendering schemes. Processing smaller amounts of data at a time also means there's more available power to throw in complex visual effects without adversely affecting the frame rate.
Gigapixel claims it's GP-1 chip can do full-screen anti-aliasing in 32-bit colour with no frame-rate penalty. The VSA-100 can do that too, but it's a much more complex, more expensive chip. Incorporating Giga3D into VSA should at the very least allow 3dfx to bring its Voodoo 5 cards down from double-chip solutions - they all have two VSA-100s working in parallel - down to single-chip configurations without a performance hit. The cost saving to 3dfx from that move alone would be considerable. Giga3D also holds the possibility of getting future VSA transistor counts down, again leading to cost savings through higher chip yields. Gigapixel also has a low-power notebook-oriented chip design, the GP-LP, which could form the basis for a family of Mobile Voodoo products, a market that ATI is carving up rather nicely and the others are making moves on but 3dfx itself has yet to tackle.
In the meantime, Gigapixel's own GP-1, GP-3 and GL-LP (for Low Power) designs will continue to be offered to third-parties under licence, much as Gigapixel has been trying to do - it has yet to announce any licensees since the GP-1 was first made available last October. Offered as a 3dfx product, the designs will carry far more weight than the did under the Gigapixel brand. The idea is to provide consumer electronics companies with low-cost 3D graphics silicon IP.
Again, ATI and S3 have been pursuing this market aggressively, and now 3dfx can too. The strategy here seems to be to sell a limited, 3D solution on the 3dfx brand rather than by offering a complete 'does everything' set-up to set-top box manufacturers, which is essentially how ATI appears to be approaching the market. Finally, it's interesting that the acquisition has emerged now, only a few weeks after Gigapixel was rejected as Microsoft's X-Box partner in favour of 3dfx's arch-rival, Nvidia. ®