Larrabee
The other fly in the ointment for NVIDIA is that Intel already owns the other main physics technology - Havok - and AMD has decided to cozy up to them on the matter of physics. "I'm getting frustrated with people comparing Havok and PhysX and saying they're the same thing," says Del Rizzo. "Havok can only run on the CPU, whereas PhysX can run on the GPU and CPU."
Can anyone use CUDA? "If AMD wanted to they could produce CUDA stuff; it's not a proprietary or closed interface," says Del Rizzo. "AMD has never come to us and said ‘We need help doing CUDA.' And they don't even need to because it's open - no strings attached. If AMD wants to use CUDA they can and it's the same with PhysX. If they're using CUDA there's nothing to pay us for. If I was Intel, I'd make Havok work on Larrabee."
Ah, Larrabee. Surely that's got NVIDIA quaking in its collective boots. "Larrabee is interesting because it doesn't exist," says Stam. "All it really does is legitimise the importance of the GPU. You should ask Intel: ‘Won't Larrabee jeopardise their CPU operations?' Also, look at Intel's history with software. There is so much software IP with NVIDIA and ATI; does Intel have the software expertise?"
Not necessarily, but it can buy it. Surely it's just a matter of time before Intel catches up. We leave the final word to Stam: "The overall pie is going to get bigger. Our belief is that we're going to grow along with the expanding market and so will AMD if they're still around."
He just couldn't resist.
