Why do most of examples in Streaming SDK use pixel shader (ps), not compute shader (cs)? What are the advantage or limitation of CS mode?
CS mode available only from HD 4000, PS can be used even on HD 2000, SDK examples were written ages ago.
As for limitations -- CS doesn't supports color buffers but supports LDS.
And no point at all to use PS with HD 5000.
SDK implements matrix multiplications with PS and CS, respectively. I run the programs on HD4870X2. It seems that PS outperforms CS....
that CS allows all registers (AFAIK 1024) to be used whereas PS only allows max 160 registers because of graphics this and that which really limits the use of register, I read it somewhere in this forums moments ago
that CS example is not the what so called "optimized" one