14 DAY TRIAL // The CUDA toolkit is one that many programmers use when developing applications capable of taking advantage of GPUs' massive parallel computing capabilities. NVIDIA has just improved it.
Basically, NVIDIA has announced CUDA 6, a new version of the development kit that takes some of the load off programmers.
To actually use the parallel computing abilities of GPU cores, a system has to copy the memory to the GPU first.
This will still have to happen, but programmers won't have to implement it in their applications anymore. Instead, CUDA will do it itself, thanks to the new unified memory system that simplifies management and addresses disparate x86 and GPU memory pools in the same space.
Basically, it doesn't improve CUDA abilities, but simplifies programming and, thus, makes it accessible to more people.
Good news for any corporation thinking of setting up a supercomputer with CUDA GPU accelerators.