The original Unreal Tournament (as well as the games based off that engine, at least by editing the config file) had tru-form support. Counter-strike also, I think, and probably a few others. I think there were actually quite a few but it wasn't heavily marketed. I only know of TF2 that used the later tesselator (via a command line), although there may be some other obscure ones ou there.
But really, the answer is that Nvidia spent more money, and marketed more heavily. It doesen't really matter in the long run though; games that used the ATI tesselator are like a drop in the bucket compared to games which use PhysX which is again a drop in the bucket compared to the kind of market penetration Glide had... and we all know how that ended. As soon as a strong industry standard emerges that everyone can use, the few developers who use PhysX will dump it en-masse.
You seem to draw you "conclusion" on a confusion of terms.
Glide was an API.
So is DirectX, OpenCL, CUDA, DirectCompute ect.
PhysX is a physics middleware.
Just like Havok or OpenBullet
But unlike Havok/Bullet which only runs on the CPU (Empty PR means nothing) Physx runs on both the CPU...and the relevant API's for the GPU:
CUDA
OpenCL
DirectCompute.
There is nothing stopping PhysX from running on OpenCL...except why should NVIDIA do that right now?
The have full controll over CUDA and both OpenCL and DirectCompute are far away in the review mirror in temrs of usage, tools-set and a general "enviroment".
So PhysX runs via CUDA for now (on the GPU) and the old SDK ran on AGEIA's old code, that they got from NOVODEX...one of the best CPU physics middlewares of that time.
Som dosn't understand this and post grabage like this:
http://www.realworldtech.com/page.cfm?ArticleID=RWT070510142143
Which even had the not so flattering fate of being debunked at Beoynd3D..and ATI haven...so sad really.
But that is what happens when drinking buddies (DKanter and Char-lie) gooff up *shrugs*
In the new PhysX SDK it's not up to the developer anymore to make everything multithreaded, it's done by auto now...so I wonder waht the excuse will be when people find out the CPU cannot compete with the GPU in SIMD calculations?
Physx isn't going anywhere, infact it's market share has being growing ever since the AGEIA days...facts don't lie.
Stating that PhysX will die because it's "proprietary" is like saying OpenGL will kill DirectX, because OpenGL is open and free and DirectX is proprietary.
And not looking at market shares, API features and developers relations.
So even is OpenBullet (a physics middleware) came out on OpenCL...not much would change.
Intel is hugging Havok tight...they won't let AMD get in that way.
OpenBullet....well...there has been a lot of TALK.
Like ATI has TALKED about GPU-physics since 2006.
Keyword: TALK.
Only one that has brought something to the table is NVIDIA.
And PhysX could run on AMD GPU's to...if AMD would so...but they won't.
They will rather TALK and TALK...did I mention TALK?
But deliver nothing.
So PhysX isn't going anywhere, quite the opposite in fact...
PS. I have heard this:
"PHYSX WILL DIE AS SOON AS...." since 2006.
PhysX has never been more used than it is today.
Find a new fail-manta and move on...please?