Acceleration of calculations using the GPU, now in Java!
IBM Technical Director John Duimovich stated this in his speech at the JavaOne 2013 conference. He noted that GPU accelerators have incredible computational power and using them for computing in Java will speed up complex calculations and processing large data arrays. Using GPU for computing in Java opens up many new perspectives for both Java developers and GPU manufacturers.
Millions of developers use Java in Web 2.0 applications to handle large data sets and scientific calculations. Java is widely used in frameworks for distributed computing, such as Apache Hadoop, due to its ease of use, modularization and cross-platform. As an example, Duimovich gave an example of standard sorting using the standard Java library, and sorting using the NVIDIA CUDA library:
The performance increase is phenomenal - from 2x to 48x depending on the amount of data! And all this is possible with the help of Java JDK 8 and the existing library of NVIDIA CUDA.
According to Duimovich, IBM will make computing acceleration possible using existing APIs. This will allow millions of Java developers to use computation acceleration to increase performance and extend the capabilities of various applications. In addition, the acceleration of computing using GPU will pave the way for a new generation of corporate applications that could not be implemented without a GPU. The number of possible applications for speeding up computing is almost unlimited: from high-performance distributed applications for financial analysis and fraud detection, to image analysis and wideband video applications and modern scientific applications.
Duimovich did not forget to mention the new OpenPOWER ecosystem built around the POWER hardware architecture, which takes advantage of free software and is built on the principles of openness between partners such as NVIDIA. Transferring computing to a GPU is an excellent example of openness between manufacturers of servers, components, and software.
')
Unfortunately, the IBM Java and NVIDIA Developer Zone sites have not yet said a word about it and there are not even approximate dates.
Millions of developers use Java in Web 2.0 applications to handle large data sets and scientific calculations. Java is widely used in frameworks for distributed computing, such as Apache Hadoop, due to its ease of use, modularization and cross-platform. As an example, Duimovich gave an example of standard sorting using the standard Java library, and sorting using the NVIDIA CUDA library:
The performance increase is phenomenal - from 2x to 48x depending on the amount of data! And all this is possible with the help of Java JDK 8 and the existing library of NVIDIA CUDA.
According to Duimovich, IBM will make computing acceleration possible using existing APIs. This will allow millions of Java developers to use computation acceleration to increase performance and extend the capabilities of various applications. In addition, the acceleration of computing using GPU will pave the way for a new generation of corporate applications that could not be implemented without a GPU. The number of possible applications for speeding up computing is almost unlimited: from high-performance distributed applications for financial analysis and fraud detection, to image analysis and wideband video applications and modern scientific applications.
Duimovich did not forget to mention the new OpenPOWER ecosystem built around the POWER hardware architecture, which takes advantage of free software and is built on the principles of openness between partners such as NVIDIA. Transferring computing to a GPU is an excellent example of openness between manufacturers of servers, components, and software.
')
Unfortunately, the IBM Java and NVIDIA Developer Zone sites have not yet said a word about it and there are not even approximate dates.
Source: https://habr.com/ru/post/195182/
All Articles