NVIDIA and Havok Bring SLI Physics to Life
At the most basic level, NVIDIA and Havok's partnership is very similar to AGEIA's PhysX PPU and Novodex API solution. NVIDIA has several advantages that may put a damper on AGEIA's plans for 2006.
First, NVIDIA's solution does not require the user to buy a separate piece of hardware in order to accelerate physics processing in-game. The same GPUs that NVIDIA sells for graphics processing inherently excel at physics processing as it is very similar to the highly parallelizable graphics processing. And by not requiring a user to add another piece of hardware to their system, they can save an expansion slot and some money.
Now, according to NVIDIA's own information, the minimum requirement for Havok FX support is the GeForce 7600 GPUs and they recommend the 7900 series of cards. The 7900 GTX sells for around $500 or so, nearly double the rate that AGEIA has planned their PhysX card will sell for. Users that have 6-series cards will not be able to take advantage of the physics processing capabilities that the Havok FX engine will utilize and the slower 7-series cards won't run them as fast.
This NVIDIA and Havok partnership also has an advantage with hardware availability and a pre-existing user base. Though the GeForce 7600 and above series of GPUs do not have an extremely significant market share compared to the 6-series parts, it is better than zero which is where AGEIA is starting from with their PhysX PPU. This may mean that game developers are more likely to take the time to program specialized physics code for a separate GPU if they know that it will be used more often in end-user's systems. Only time will tell if the Havok FX engine can be more successfully integrated in games than the Novodex API from AGEIA.
Another advantage that NVIDIA will have, at least in theory, is a lack of latency or communications issues getting data back and forth between the physics processor, the graphics processor and the CPU. All three components need to share data often and thus having both the graphics and physics data on the same card (or two cards, in SLI mode) should offer some kind of advantage. What kind though, is impossible to tell, as we haven't seen either implementation in any detail.