In our previous article GPGPU for Java Programming we showed how to setup an environment to execute CUDA from within java code. However the previous article focused only on setting up the environment leaving the subject of parallelism untouched. In this article we will see how we can utilize a GPU do what is doing best: parallel processing. Through this example we will take some metrics and see where GPU processing is stronger or weaker than using a CPU …and of course as the title suggests there is an ugly part at the end.

Read more at:

Java Code Geeks : GPGPU with Jcuda the Good, the Bad and … the Ugly

This article can be replaced with: "Just Use C++". If necessary, interface it with Java using JNA ( http://jna.java.net/ ). It's simple, it works fast.

That's all. Java is not suitable for GPGPU tasks.

This article can be replaced with: "Just Use C++". If necessary, interface it with Java using JNA ( http://jna.java.net/ ). It's simple, it works fast.

 

That's all. Java is not suitable for GPGPU tasks.

No need to be a hater because people can now easily integrate another part of their system with java instead of using ancient c++ hieroglyphs.

Have you actually checked the article? C++ is infinitely better than manually packing your data into linear arrays of primitives.

Have you actually checked the article? C++ is infinitely better than manually packing your data into linear arrays of primitives.

 

Actually this article is not about if it is better to write a fresh new application in this or that language, but what options do you have if you already have a Java application, or your company has made a huge investement on Java and you need to use CUDA.   My aim was not to see if C++ is better for CUDA than Java. Of course and it is no question about it, although maybe that will not be such a clear win when I check what tools exists for OpenCL and Java, will see...