NVIDIA has announced the availability of CUDA 5 that has been updated to support the new features of the Kepler GPU. The company also unveiled its new online CUDA Resource Center that is a one-stop shop for developers to take advantage of the benefits of GP-GPU computing. The company has also made its recently announced Nsight Eclipse Edition, an integral part of the CUDA 5 framework by making it part of the CUDA 5 toolkit.
The Tesla K20 GPU supports Dynamic Parallelism, making it easier for GPU developers to speed up all parallel recursive loops, resulting in the Kepler GPU being able to create its own jobs without having to go back to the CPU for such instructions. The new CUDA 5 framework has been updated to take advantage of this feature of the Tesla K20 that sports a GK110 graphics core.
GPU Callable Libraries
A new feature of the CUDA 5 is the ability of developers to utilize the new CUDA BLAS library to write and distribute their own libraries, complete with APIs to extend the ecosystem of the CUDA 5 framework into their own software system. These developers are also able to implement callbacks on the GPU to customize the functionality of third-party libraries. This is akin to object linking and embedded (OLE) that is used to insert data from one software program into another. In this situation, the data inserted or called can be an independently-compiled object from different CUDA source codes and linked to other applications and third-party libraries through a compiler tool chain.
GPUDirect Support for RDMA
The updated GPUDirect technology now supports direct memory access between network interface cards and the GPU while enabling direct communication between GPUs and other devices connected on the PCI-E bus system. GPUDirect is claimed to significantly reduce latency between nodes in a GPU cluster and improves overall application performance.
The Nsight Eclipse Edition has been made an integral part of the CUDA 5 toolkit and is immediately available to developers who have installed the CUDA 5 framework. NVIDIA has also launched its new CUDA Resource Center that provides instantaneous access to all the things software developers require to start taking advantage of the benefits of GP-GPU computing. The resource center provides CUDA programming guides, API references and library manuals. There are also code samples, development tools documentation as well as platform specifications as the Nsight Eclipse Edition has been expanded with support for Linux and Mac OS platforms.