Machine learning is being adopted by more and more companies for a wide range of artificial intelligence applications such as image recognition and self-driving vehicles. Machine learning involves running large data sets through a software program in order to train the software to build its intelligence. One of the biggest challenges facing developers building software that incorporates machine learning is in processing the huge amounts of data required for training their programs. By using GPUs, developers can vastly increase the amount of data that can be processed whilst also reducing the overall power consumption by running calculations in parallel.
Working within the EU Horizon 2020-funded LPGPU2 project, Codeplay has made key contributions to the open source TensorFlow™ project, adding OpenCL™ support through the SYCL™ open standard. OpenCL provides a framework for developers to write software that executes across heterogeneous platforms such as GPUs. TensorFlow, a Google-led project, is one of the most popular machine learning frameworks, specializing in computation using data flow graphs with a focus on machine learning. TensorFlow uses the Eigen C++ library for linear algebra and, by offloading parts of Eigen to OpenCL devices, it is possible to accelerate training for the neural networks built by TensorFlow as well as reduce the power usage of applications.
ComputeCpp™, Codeplay’s implementation of SYCL, offers a C++ single-source heterogeneous programming model layer on top of OpenCL. SYCL provides an abstraction layer that simplifies parallel development, giving developers access to the computing power of GPUs and reducing the amount of code required. Developers will be able to use ComputeCpp to take advantage of OpenCL and accelerate their TensorFlow applications through use of GPUs that are capable of processing thousands of calculations in parallel. This will enable them to build huge neural networks, speed up training and execution of complex neural network models and reduce power consumption.
Find out more about the Codeplay ComputeCpp SDK.
Comments are closed.