I often meet software developers who say they’ve heard about the awesome power of GPUs but are afraid of venturing into the world of parallel programming. The truth is, it’s much easier than you think.
To help ease wary minds, The Portland Group (PGI) has launched the “2x in 4 Weeks. Guaranteed.” Program. It’s designed to help developers achieve up to 2x speedups in their applications on GPUs in a month or less of development time.
PGI has been working on what you might call a “magic” compiler, called the “PGI Accelerator.” Technically speaking, it’s an auto-parallelizing compiler. It enables programmers to realize immediate acceleration of their application code, without requiring them to modify or adapt the underlying code itself. The compiler only requires the programmer to provide simple hints (known as “directives”), about which portion of an application’s code needs to be parallelized. The compiler then does the heavy lifting — automatically parallelizing that portion of the code.
This directives-based approach is extremely easy to use. It’s also much faster than manually changing the code. And it provides impressive resolution. In fact, many PGI Accelerator users are seeing 5x to 10x speedups, sometimes in as little as two days. And the best part is that the updated code can run on CPUs or GPUs, or both.
We are so confident in this directives based approach to parallel programming, that we are guaranteeing your results. Simply sign up to get a trial copy of the PGI Accelerator compiler and try it out.
We guarantee you will get 2x application speed-up in less than a month of development time. If you don’t, NVIDIA, with PGI, will offer four hours of expert consultation to accelerate your code.
In less than a day since the program’s launch, we have seen nearly 50 developers registered for the program. One developer, working in image processing, submitted his story of how he saw 10x speedup in less than three weeks. Another developer, working in robotics, saw 2x speedup on his application in less than two weeks.
With PGI Accelerator, these developers were able to focus on results, without spending months (or years) tweaking their CPU-based code, only to realize minimal performance increases.
So, give it a try and do come back and tell us your story! We’ll post all the submitted stories in a few weeks so stay tuned.
Tagged: CUDA, GPGPU, gpu computing, high performance computing, hpc, Parallel Programming, Tesla
