@amaedit: wow I think you're getting ahead of yourself here... You need to understand what GPGPU is first before you go into Stream, Cuda performance boosts etc...
GPGPU us as the acronym suggests "General processing on GPU". This can come in any way shape of form, but it relies primarily on a supporting API (DirectX10+/ OpenGL3.0+ or another one which I may not be aware off) and the underlying hardware being capable of some simple or rather basic arithmetic functions that older graphics hardware (DX9 class etc...) did not have.
DX10 class GPUs are massively parallel even the lowly ones with 32 stream processors and such. They are going to be better at some simplistic calculations than a CPU running at a much higher clock speed but one that can only process 2 or 4 instructions a cycle. These tasks are obviously graphics, physics, video transcoding and other repetitive tasks that need parallelism to get to a desired result than any thing else...
E.g
compare a quad core CPU at 2000MHZ and a 32 stream GPU at 500MHZ.
Assume we have a "simple" mul+add instruction to perform. This instruction take one cycle to complete. The CPU can do 2 Billion of these instructions every second in a perfect scenario in single thread mode. when using all 4 cores this number could be 8 billion. Now the GPU can only ever perform 500 million of these calculations in single thread mode (use one stream processor). However, since you have 32 streams this number goes up by a factor of 32 which means the GPU can do 16 Billion of these calculations per second. The GPU is then twice as fast.
ATI Radeon Graphics Cards Running Nvidia PhysX Are Faster, Stronger, Awesomer
gizmodo link you provided says nothing about NVIDIA but states that a 3DMark Vantage score increased on an HD3800 based card when using hardware physics.The only NVIDIA mention in the article, and I quote ,
"Nvidia's efforts to spread the love to graphics cards "
ATI shows off PhysX architecture
you then link to an article written in June 2006 comparing Havok physics on RV530 and G7X based cards... GPGPU was not available at the time on any desktop platform... Hardware accelerated Physics has not been available since then. Comparing a custom implementation of Havock physics on ATI hardware by ATI employees to what the G7X card can do is meaningless... the article you link says later on
But these figures were provided by ATI, and were not independently verified.As no games run across Havok and PhysX right now, any numbers appear to be theoretical.
Then in the video link you provided, you can clearly see the demo say "Software", believe it or not you can run that demo on just about any graphics card, even on Geforce 7 cards... Since it's software its running on the host CPU....
So in closing what exactly are you saying?
(BTW stream much like CUDA is GPGPU solution. It doesn't boost AMD CPUs it takes away the work load from the host CPU regardless of it being Intel or AMD, the first truly joint effort by engineers and designers of both ATI and AMD will be Fusion in 2010)
