NVIDIA has announced and released its CUDA 5 platform today to a hive of optimism from industry experts. It looks like the CUDA 5 platform comes with a bunch of new and exciting features that will make programmers lives so much easier. Some of the features listed by NVIDIA’s official release documentation include Dynamic Parallelism, GPU Object Linking and GPUDirect.
Personally I found GPUDirect the more interesting with RDMA (Remote Direct Memory Access) meaning that GPUs can directly access memory on other GPUs in other servers which means less memory bottle necks and allowing direct access between GPUs. This will improve overall performance where external hardware is accessed.
To me it looks like NVIDIA is positioning its self through CUDA to become the market leader in highly-parallel programming and why not as they will sell more graphics cards that way (with over 375m CUDA enabled GPUs in the market place). You may be wondering what parallel programming and CUDA can be used for well nVidia list such things as Identify hidden plaque in arteries and analyze air traffic flow, or one you might know better Folding at Home (FAH).
NVIDIA have also stated that in a future release of CUDA they will be looking at including support for ARM processors which shows just how important mobile computing is becoming for NVIDIA.