I know this post is long, sorry for that, but I think I have some good points worth of debating. Please read and tell me whatyou think, guys.
However, in what way have we've seen ULi and 3dfx in G80 or G92 video card?
I really don't see it.
In what way has Nvidia used Uli and 3dfx? That's your question?
Well in G80 and G92 very few things if at all, I guess.
But IIRC before Nvidia bought Uli they only offered one chipset and it was an enthusiast one. Later on Nvidia offers the high-end chipset along with his smaller brother.
3dfx: They designed the GeForce FX series entirely. Any doubts on why we don't see their "touch" on later GPUs? I'm sure that there are some features in them developed by the people who belonged to 3dfx, it's just not apparent on the surface. Not sure, but I think that Lumenex engine has something to do with them. I think that I have read that somewhere, but can't find a link to it.
BUT if you want to see 3dfx into Nvidia you don't have to go too far. SLI is a 3dfx invention and it was first developed by 3dfx. It's there where 3dfx influence is most clear.
Now on topic. IMO they have bought Ageia because they had already integrated some kind of support for physics in their future designs, maybe in conjuntion with Havok and now without them, they had hardware support without any API or compiler to work with. This support would come in the shape of GPU's instruction set and maybe how the SPs work internally. Of course both GPU and PPU's internal units are no more than Floating Point Units, but the difference relies on how they opperate. As an example, think of AMD's Firestream stream processor based on RV670, it does double-precision (FP64) at 1/4 of the speed of single-precision. Or you can think of SSE extensions on the CPU.
There are some posts in this thread claiming that physics calculation on the GPU would cripple graphics performance. That's not exactly true. I'm of the opinion that G80 was bottlenecked by SPs, but this doesn't happen with G92, meaning that there's some spare shading power, specially on the GTS. Ageia Physx has a peak of 50 GFlops, current GPU are around 500 GFlops. You can see where I'm going. And if we ahve to believe leaked info about 9800GTX my point is much more feasible:
-384 SP and over 1 TFlop.
-96 TMU
-32 ROP
-512 bit
-GDDR4
That translates to:
- Double the SP/TMU ratio compared to current gen: 384/96 = 4 Vs. 128/64 = 2
- 50% increase over G92 in SP/ROP ratio and 125% increase over G80: 384/32 = 12 Vs. 128/16 = 8 Vs. 128/24 = 5,33
- A hell of a lot higher memory bandwidth.
With those (if true) it's easy to conclude that Nvidia was expecting to see a high increase in SP usage and data traffic. Are games in the near future going to see such an increase in shading usage for it's graphics alone? I don't think so, because they would render current generation useless, a big base for their games. But what about a feature like Ageia's physics? I mean something that would be more like an added feature, a bonus, just like now happens with Physx?
Let me know what you think of this guys...