TunaCode » Blog

GTC 2012 just weeks away and we’re participating!

Salman Ul Haq — Wed, 25 Apr 2012 16:24:54 +0000

NVIDIA GPU Technology Conference 2012 is just weeks away. It’s the biggest gathering of everything GPGPU (General Purpose Computation on GPUs) with over 300 technical sessions, 100+ companies participating and expected attendance of more than 3000 professionals, this is a not-miss event. This year will be different since for the past 3 years, all the focus was on proving that GPUs are a disruptive technology and here to stay and it’s not just another spike in the HPC (High Performance Computing) domain that happens every decade. This year, we’re expecting to see a roadmap being chalked out as well as even more major players announcing support for GPUs.

NVIDIA GTC also offers a great opportunity for startups in the GPU landscape to showcase their products and services at the event. There’s a dedicated “Emerging Companies Summit” and this year 60+ companies have signed up. We are one of them and we are definitely very excited about that.

The event kickoffs from Monday, 14th May packed with keynotes from industry luminaries, technical sessions, round-table discussions, startup summit and if you’re a registered developer you get a free 1536 cores GPU!

So if you’re planning on attending the event, do stop by our booth!

TunaCode Releases CUVI version 0.5

Salman Ul Haq — Wed, 20 Jul 2011 09:10:53 +0000

TunaCode is pleased to announce the release of CUVI (CUDA Vision and Imaging Library) version 0.5 which comes with a new API and amazing new features. With this release, our team of expert GPU developers have made it even more simpler to add acceleration to your existing Imaging applications, without any prior technical knowledge of GPUs whatsoever. This not only removes the technical barrier to gaining massive speedups but also reduces your time-to-market. We believe that CUVI 0.5 will add value to your product.

CUVI v0.5 is built from bottom up with performance and ease-of-use in mind. I would like to take this opportunity to tell you a bit about these 2 aspects:

Performance:

CUVI provides GPU accelerated Image processing and Computer Vision functions which are, on average, 10 times faster then their CPU based versions. CUVI functions provide this acceleration by utilizing NVIDIA GPUs (Graphics Processing Units) which have hundreds of light weight compute cores compared to CPU cores. This will shift the computing load from your CPU to GPU freeing the CPU for other tasks. Take a look at the Features Page for comparison of some of the functions.

Ease-of-Use:

CUVI v0.5 has been tested to work with OpenCV and Intel IPP libraries which means integration is ever more easier. This also means that you do not have to rewrite your code from scratch to utilize GPU’s compute power. Take a look at this blog post that explains how easy it is to use CUVI with OpenCV.

New Features:

The following new functionality has been added into CUVI:

1) Feature Detection: KLT and Harris feature detectors
2) Dense Motion Detection: Optical Flow Lucas & Kanade and Horn & Schunck
3) Image Transform: Haar Discrete Wavelet Transform
4) Image filters: 2D linear filter

The API has been improved to include CuviImage object which holds the image on the GPU. You can read more about it in the Developer Reference (PDF). CUVI version 0.5 is available for download at CUVILib website and is available for Windows (Win32, x64) with planned support for Linux and Mac.

You can also get in touch with our GPU experts to help you set up CUVI and make sure you get the best out of it.

Debugging OpenCL Made Easy: Introduction to AMD gDebugger

Aamir Majeed — Fri, 01 Jul 2011 06:14:51 +0000

Finally, the OpenCL developers have the long awaited tool that takes the mystery out of debugging OpenCL and OpenGL applications. Now, a developer can actually peer into memory objects on the GPU, view their contents as they change from kernel operations and step into the kernel execution for debugging inside the kernel, all in real time. The credit goes to the AMD gDEBugger, as it is called, for bringing all this capability on a single computer with a single GPU! Shocked?

What is AMD gDEBugger?

AMD gDEBugger is an OpenCL and OpenGL debugger and memory analyzer tool that traces application activity on top of OpenCL & OpenGL and provides necessary information for debugging and optimizing application performance. Analytic capabilities of gDEBugger range from locating bugs and memory leaks to optimizing applications by removing redundant calls and errors. The best part is that this tool directly integrates into Microsoft Visual Studio and requires a single GPU.

gDEBugger offers real-time OpenCL kernel debugging which allows the developers to debug inside the kernel and view all the variable values across different work-groups and work-items. All allocated memory objects, like OpenCL Images and/or Buffers, are monitored to detect any memory leaks and the scenarios that cause these memory leaks. gDebugger also create logs for all API function calls that can be saved and/or viewed at run-time. These logs provide information about non-recommended and deprecated functions and behaviors in the kernel and offer best-practice alternatives to help improve application performance.

Key Features

Real-Time OpenCL Kernel Debugging

Minimize the debugging effort as no special configuration or changes to the code are required. Allow stepping through the work-flow of a single work-item and compare values across a work-group.

Seamless Integration with Microsoft VS

Debugging is performed inside Visual Studio Source Editor. When included in the project, OpenCL Kernel files (.cl files) are automatically identified and used for kernel debugging. The debugging information is displayed in Visual Studio Views like Locals View etc.

API Statistics Overview

gDEBugger provides an overview of the OpenCL & OpenGL API usage and detailed views including non-recommended function calls (with alternative suggestions).

Object Visualization

gDEBugger allows developers to export OpenCL Images and Buffers, and OpenGL Textures and Buffers as pictures or as spread-sheet data.

Develop & Debug on Same Machine

With gDEBugger you can Develop and Debug on a single computer with one GPU. So every GPGPU developer can use this tool without any additional hardware requirement. CUDA debugger, Parallel NSight Monitor, requires a dual GPU machine or two machines with one GPU each for the same purpose.

System Requirements

1. Microsoft Win 7, Win Vista (32-Bit, 64-Bit)
2. Microsoft Visual Studio 2010.
3. gDebugger [Download Link]

Conclusion

Whether your goal is to shorten debugging time, improve application quality or optimize application performance, gDEBugger provides the information you‘re looking for.

Welcome to TunaCode Blog

Jawad Masood — Thu, 19 May 2011 06:05:04 +0000

Hello everyone and welcome to our blog. Let’s me do a brief introduction of our company; we are a software company but not just-another software company. We, at TunaCode, write software that runs fast leveraging the power of Graphics card that reside in almost every personal computer. To speak technically we are CUDA and OpenCL experts with domain knowledge in Imaging, Computer Vision and Data Encryption.

Our flagship product is a GPU accelerated software library, CUVILib, that provides numerous Imaging and computer vision functionality. We guarantee at least 5x faster results for the same algorithm when compared to CPU based implementation and at a lesser cost. CUVILib has its own blog where we share tips and guides relating to the field itself. Here (on this blog) we’ll be sharing news about our company as well as the Industry news and trends, so stay tuned.