subject: Nvidia Announces Cuda 5 For High Parallel Programming [print this page] Nvidia has now officially announced CUDA 5 that come up with promised and improved performance with easier coding and a new resource center for those users who are looking to accelerate highly-parallel tasks.
The simple reason that it will sell more graphic chips and with that way it spent much of its launch presentation in which programmers are discussing parallelism for spawning new parallel work within GPU code, GPU Direct, GPU callable libraries for high-performance and low latency for direct memory access between GPU's and PCI Express -connected devices to optimize from a single interface.
This new platform features its ability to spawn new threads from GPU threads which means that this is possible for GPU to automatically adapt to the content at hand where the necessity of communication was previously required. It seems that the performance is quite improved by eliminating and reducing the CPU interference on the GPU's operations.
The callable libraries of GPU are a part of Nvidia attempt to harbor a wider third-party eco system that let users to access CUDA parallelism through their own libraries. Nvidia suggests that coders can write plug-in APIs to let other programmers to extend the functionality of their kernel allowing people to implement callbacks on the GPU to customize the functionality of third-party libraries. This company is hoping that developers will take benefit of the new object linking capabilities to develop larger and more complex CUDA-powered applications.
This minimizes the system memory bottlenecks which are designed to allow GPU's to communicate with other PCI express-connected devices without any involvement of CPU and RAM. GPUDirect is claimed to significantly reduce latency exists between different nodes in GPU cluster as well as improved and overall performance where external hardware is accessed.
The Nsight plug-ins used for Eclipse offers developers with the ability to write, debug and compile their CUDA code within the Popular IDE available on Linux and OS platforms. Those users who are using Eclipse will find a new automatic refracting tool to fatly port the existing code to CUDA combined with customized syntax which are highlighted to differentiate between CPU and GPU code segments.
The CUDA resource Centre offers instantaneous access to all the different things developers could want to begin taking benefits of parallelism. You can get every information related to programming be it Programming guides, API references, library manuals, code samples, tools documentation or any other platform specifications which are required.
