filmov
tv
Lightning Talk: A New, Open Standards, Open Source, Software Project for... - Charles Macfarlane
Показать описание
Lightning Talk: A New, Open Standards, Open Source, Software Project for All Accelerators - Charles Macfarlane, Codeplay Software
The AI and HPC space is seeing exponential growth, with projects like ChatGPT and GROMACs attempting to squeeze more and more performance from accelerators such as GPUs. The challenge for software developers is that they are currently constrained by proprietary programming models, and this is stifling the choices they can make about their hardware platforms. Developers today are often locked into Nvidia GPUs because their software uses a proprietary programming model called CUDA, and it’s time to build an open alternative. The oneAPI project has been created to bring together an open specification and open source software that today enables developers to write software using standard C++ code and deploy to multiple vendors of GPUs. This project is evolving rapidly and the whole community of hardware and software developers are invited to contribute to this project. This presentation will introduce how it is possible to write multi-target software using the oneAPI project, and how everyone can contribute to making this project available for all accelerators in the future.
The AI and HPC space is seeing exponential growth, with projects like ChatGPT and GROMACs attempting to squeeze more and more performance from accelerators such as GPUs. The challenge for software developers is that they are currently constrained by proprietary programming models, and this is stifling the choices they can make about their hardware platforms. Developers today are often locked into Nvidia GPUs because their software uses a proprietary programming model called CUDA, and it’s time to build an open alternative. The oneAPI project has been created to bring together an open specification and open source software that today enables developers to write software using standard C++ code and deploy to multiple vendors of GPUs. This project is evolving rapidly and the whole community of hardware and software developers are invited to contribute to this project. This presentation will introduce how it is possible to write multi-target software using the oneAPI project, and how everyone can contribute to making this project available for all accelerators in the future.