AMD to release Stream Processor Posted on 15/11/2006 at 16:13 by Damien
Following ATI's announcements about Stream Computing (CTM, client Folding@home for GPU…), AMD decided that it was time to release a specific product for this market: the Stream Processor.
With the CTM, developers have a low level access to the GPU that allows them to highly optimise their code. NVIDIA chose a different path with the CUDA project and the G80. The manufacturer uses the C language and has decided not to unveil too many details about the functioning of their calculation units. AMD and NVIDIA's solutions respectively favour a better optimisation of the code and an easier portability.
What are the improvements of the Stream Processor compared to CTM that works with all Radeon X1000 and future GPUs? Because of the functioning of the CTM, drivers can't be used to differentiate the two products like the FireGL. The difference will be on the validation and support level but also on the product level itself as it will be optimised for general calculations. The 1GB of memory will be of GDDR3 type since it allows lower granularity for memory accesses and probably lower timings than the GDDR4. It is hard to tell if it will really makes a difference compared to a Radeon X1900 that will be ten times less expensive.
