Future PCI Express And AGP Solutions
We should see the first "real" PCI Express graphics solutions by the summer. So why is NVIDIA putting so much effort into HSI chip if the next generation of chips, which will support PCI Express natively, are literally on the horizon?
For one thing, because this strategy allows NVIDIA to have a full line of PCI Express graphics cards right from day one, catering to every need from the entry level to the high-end enthusiast. Secondly, and much more simply - because they can. It makes perfect business sense: AGP will not vanish overnight with the introduction of PCI Express. There will still be a demand for AGP cards, and it will probably only fade very slowly, because everyone wants to make a buck in the transition to the new bus. After all, simply upgrading to a new motherboard isn't the end of the line - you'll almost certainly need a new processor, new RAM, a new graphics card, probably a new power supply and perhaps even a new case.
For many prospective buyers, a simple graphics card upgrade with a AGP model will therefore surely be the preferred (read: less expensive) solution. Since this is the case, the chipmakers would suddenly have to design, manufacture and advertise two versions of the same chip, one each for AGP and PCI Express. Considering that a chip tape-out is anything but cheap, it would be logical to focus development of new chips for he PCI Express versions. Should there be demand for an AGP version, it could be implemented by using a bridge chip, which would be far less expensive than creating a second version of the same chip. NVIDIA has all its bases covered, since it had the foresight to design its bridge to work in both directions. This way, it can either link a PCI Express chip to an AGP bus or vice versa. In effect, the HSI bridge allows NVIDIA to present a full complement of PCI Express cards right from the start with very little financial input.
A PCI Express prototype card from ATI.
It stands to reason that future ATi cards will also use some kind of bridge chip for its AGP models - it would simply be too expensive to create two versions of the same chip. Since ATi's bridge will only need to translate in one direction (PCI Express to AGP), it can be incorporated into the chip. It is unclear how many transistors and consequently how much precious die space such a bridge requires and how much heat it develops (NVIDIA's mounts a heatsink on its bridge chips). ATI is also forced to introduce their new cards in time. Any delay means loosing the starting hype of PCI Express.
Which brings us back to square one - right now, all of this is just theoretical. PCI Express is not yet available. We shouldn't expect to see any performance improvements in 3D games as a result of the migration from AGP to PCI Express at first - at least in theory. Just as theoretically, NVIDIA' PCX line of cards should be faster than their AGP counterparts, due to the higher internal AGP transfer rate supported by the HSI bridge. We will have to wait for months at least to get any answers to our questions and proof to back up our hypotheses. Who knows, maybe the next generation of graphics chips will arrive sooner than anyone expects, and all the speculation about how NVIDIA's PCX cards with their HSI bridge perform will be irrelevant.