Nvidia Buys Gaming Software, Hardware Maker Ageia (PhysX)

#1 I hadn't seen mention of it here, but it was over at the DarkBasic and Unity forums. Should be interesting.

#2 The main question is... why would nVidia do that?

I mean, as far as vector processors are concerned, nVidia have by far the best technology out there. Ageia's tech isn't nearly as good, and it does very similar things (hardware wise).

Are they just getting rid of one of their potential hardware competitors? Or are they simply after Ageia's software and user base (which is undoubtely good, but it's probably not worth all the money they're going to spend IMO).

#3 Think bigger, Hadoken.

A GPU is a highly-parallelized, autothreaded vector processor. A PPU is a vector processor. One of the primary problems with GPUs right now is that they are highly specialized for a specific task, but they have all the capabilities of more general tasks, only with much poorer data access. PPUs were mostly DOA. It was a semi-cool idea, to hardware accelerate physics. The problem is that working with someone elses global solver isn't what you want to do, most of the time.

Most likely this is a move twords more generalized vector hardware, and that is quite exciting.

#4 This report has a seemed aproach:

Where next for NVidia and Ageia?
"ANALYSIS: How the recent acquisition could affect developers"

Very intresting report on Magazine/WebSite Develop.



On a side note, is somehow significative to see so much movement on the industry giants, after almost a decade of calm. In a relative short period of time, AMD has bought ATI, Intel has bought Havok, and now Nvidia bought Ageia.

#5 Apparently ATI has worked with Havoc on something similar.

ATI CrossFire and Physics

#6 The analysis paper was a very good read, thanks for posting that link.

I have to say I share their doubts... i've never seen any modern AAA game that isn't GPU-bound. Even if GPU's are getting more and more powerful, visual requirements are also getting higher and higher.
Big studios are always very concerned about making games that *look* great, and that's where all the GPU power ends up. Hopefully that'll change in the future, although I dont know if it's going to be a near future.

Pat, you have a good point there. However, I'm not sure about GPU's having poorer data access... they have excellent data access, as long as your data is organized in a GPU-friendly way :)
Seriously, that's a problem whenever you multithread anything so massively, regardless of the hardware.

You can do almost any general task efficiently with GPU's the way they are now (even without CUDA), you only have to disguise your data as vertices, pixels and textures that fit well into those pipelines.
It's really up to you whether you want to think of it as a limitation, or as a necessity imposed by massive multithreading itself.
(I'm deliberately ignoring the limitations of higher-level APIs such as D3D, that's a different issue).

But I agree, it's probably exciting news. Also, PPU as an acronym was starting to mean too many things, it was time someone did something about it ;)

#7 Well the "poorer data access" bit I guess should have been reworded as, "very restrictive data access." For example, shaders execute on a single entity, like a pixel. Depending on the per-pixel operation that you want to do, it may be faster to do it on multiple CPUs that can fetch clumps of data.

DX10 and GL provide much, much better data access with fewer restrictions and less overhead than DX9 currently. The Vista barrier is an intimidating one, and I think that, with the way that GL3.0 looks, OpenGL could return to being "the API" for games. I've been chomping at the bit for a while to re-write, and unify, the data storage and manipulation in GFX. Right now things are kind of backwards. You say, "Please make me a vertex buffer, I have data for it." Instead of saying, "I have a chunk of formatted, aligned data. Please use it as a vertex buffer."

#8 Makes sense to buy them. I am sure some brilliant staff and of course the IP.

#9 I'm rather interested in this news, my concern would be having two competing and incompatible technologies.

And very interested by Pat Wilson - I have for a long time thought that OpenGL should be the main graphics layer in engines. So hopefully OpenGL 3.0 will make this happen :-).


EDIT: It took nVidia a long time to use SLI (and even then the technology is quite improved over what 3DFX did).

#10 SLI is part of the problem. The retail folks and Nvidia have locked PC gamers into channels such as "enthusiast".. translated that equals someone with Cash flow to afford 2 cards.

SLI does not help the masses get stable performance development client target platform when people like Buy Moore still sell the crap low end gpu's and then wonder why PC games are not selling as well.