ttgLabs, a Russian software company that focuses on porting and optimising applications on GPGPUs and hybrid systems, is making available a pre-release version of its TTG Apptimizer toolkit. According to the company, the software facilitates the optimisation of computationally-demanding applications on graphics accelerators and heterogeneous HPC systems enabling developers to easily work around the more tedious programming tasks.
In contrast to traditional optimisation techniques, TTG Apptimizer actively uses the concept of software autotuning. After a short self-learning cycle, an application developed with TTG Apptimizer tailors itself to the hardware platform on which it is running and to the data that it has to process, thus approaching the highest performance for the particular computational environment and data. Such autotuning is dynamic by its nature - the application permanently monitors the hardware configuration and data parameters and readapts itself to any changes. Currently, TTG Apptimizer is available in CUDA edition for Microsoft Windows and Linux and can be used with graphics processors from Nvidia.
The pre-release version, which is freely available from the company by request, supports only one processing node and has limited functionality. A commercial version of TTG Apptimizer with added support of OpenCL and of GPUs from Intel and AMD will be available by the end of the year. The company also plans to release mini-cluster and supercomputer versions of the toolkit supporting up to 20 and up to 1,000 nodes, respectively.