GPU Programming in Java Using OpenCL, Part 1

General Purpose Computing on Graphics Processing Units (GPGPU) allows you to make use of your video card (GPU) to perform general purpose computing. The typical GPU contains hundreds of cores capable of performing mathematically intense operations. However, harnessing the power of the GPU is much different than the traditional CPU programming that most programmers are used to. This presentation will show how to make use of the GPU from Java using OpenCL. OpenCL abstracts the differences between competing GPU architectures.

GPU's are not suited to every task, so the criteria for a good GPU task will be reviewed. To demonstrate the power of the GPU in Java I will show how to use the GPU to speed a Big Data process over a large volume of financial data. This can greatly speed up certain data mining and predictive Modeling applications. I will also show how GPU's can be used in an Amazon EC2 cluster.

[dodgydave]

Interesting stuff Jeff! Thanks for this

Nice video. Im using Jocl and Lwjgl to use cl-gl-interop and doing cloth animation and I need some variables to be persistent in gpu-registers(private vars) between different kernel calls but its not possible. Will this be applicable in the future?

[zacharystevens]

A small note that I would add to anyone interesting in using opencl. Like mentioned above, Nvidia uses a more specialized version of opencl named CUDA. Therefore they don't have the highest focus on opencl, therefore they are still shipping opencl 1.1 instead of 1.2 like intel, amd/ati. This might save you some pain trying to compile on NVidia cards with 1.2. :)

[abaybektursun]

Is the floating point precision and the transfer speed going to be same for Pascal generation?

[CsabaTothMr]

Another question: what is your opinion about Xeon Phi? Especially that you port neural network and machine learning algorithm to GPGPUs, I'm really curious.

[andy]

You can pick up Tesla, Quadro and GeForce cards at most online PC parts retailers. The Tesla cards cost a few thousand but they're certainly available to consumers.

[CsabaTothMr]

I like your presentation. Do you have any information/link when a Quadro has real hardware difference (for example improved DP FP capabilities) as opposed to a GeForce counterpart? As far as I saw usually you can find the GeForce counterpart with the same core, same hardware (but different driver).

[hellowill]

Kek used this in parallel programming class and my program was much faster than others.