OpenCL as the CUDA killer
In case you missed it, Apple announced a project called "OpenCL" a couple weeks ago.
OpenCL seems to seek to solve 2 out of 4 of the problems with CUDA:
(3) hard to program
(4) not future proof (hugs hardware too closely)
OpenCL, if successful, looks to fix (1) and (2). The first by being open, the second by mixing in use of CPUs including features to allow task parallelism.
Will history show that Apple effectively announced the death of CUDA?