Multicore Computing


The recent 2011 AMD Fusion Developer Summit

posted Aug 5, 2011 6:54 PM by Manu Pillai

Back in June, I attended the AMD Fusion Developer Summit in Seattle, where MCW (our client) unveiled a number of tools for increasing the productivity of parallel programming, with (naturally) special focus on the Fusion APU launch. The MCW tools were also included in hands-on lab demos to great success. 

At a personal and professional level, I was impressed with quality of the conference; each presentation had something for most, and the cosy arrangements by the Meydenbauer Center staff with a constant supply of quality coffee (this was Seattle ) and snacks, along with tons of table and chairs for impromptu meetings created a great atmosphere. Conversations ranged from details of instruction sets to programming models to business models. 

There were a number of great presentations made as well, including a number of though provoking keynote speeches which I really recommend. 

Looking ahead, there is clearly going to be a ton of activity in this area, as OpenCL and other programming methods become more accepted on Windows, MacOS and Linux on the server, desktop and portable application areas. On the mobile handset / tablet side of things, the impact of Android and Google's approach to things will clearly have some impact. It is too early to say what will happen by this time next year - except we all know there will be constant change.

OpenCL or RenderScript?

posted Jul 29, 2011 5:27 PM by Manu Pillai   [ updated Aug 5, 2011 7:15 PM ]

Long ago, NVidia essentially took the bull by the horns, and developed a programming model for their GPUs. They called this CUDA. Its been successful and has delivered results - but the GPU part works only with NVidia. That made sense at the time - NVidia was standing alone in this industry, and no one had their back. They did what they needed to, and did it well. 

Intel continued to support Parallel Studio on Windows and Linux on a CPU-focused strategy, which made sense because their GPU plans weren't going anywhere against NVidia or AMD-ATI. And NVidia continued to do their thing. 

OpenCL came about through Apple; they put in the work, got things going, and handed OpenCL to the Khronos Group for open standards stewardship a few years ago.  This makes sense - Khronos also handles OpenGL and the system works. With Apple's leadership, Intel, NVidia and AMD all came to support OpenCL. There is a ton of activity in OpenCL, especially in the server applications space. OpenCL drivers work on Linux, MacOS and Windows. It's here now.The functional portability of this standard make it appealing to software developers and system houses, as does its programming model. 

RenderScript is from Google, or from !Apple and !Microsoft if you want to think that way. But, due to being the default parallel programming support system on Android, it's getting a lot of attention. As of writing, the approach for GPU support on RenderScript is not yet publicly clear.

The net of all is confusion - while the server-class guys can move with confidence to OpenCL - the multicore market on mobile devices is in a bind. Not many people can do both from ground up, so the logical path would seem to be provide OpenCL support now, and then support RenderScript with small modifications. (This is feasible since both leverage the C99 language model, enabling translators to be built.) Doing this also ensures that a chip supplier could still approach Apple or Windows licensees for business on various mobile platforms, while also having a clear plan for RenderScript support. 

TechRev, through it's partnership with MulticoreWare Inc, can support a CUDA, OpenMP, P-Thread, OpenCL or RenderScript based approach to parallel programming in multicore systems - including CPU-GPU models as well. 

If this is of interest to you, please contact us at info@techrevllc.com

TechRev and Multicore computing

posted Jul 16, 2011 11:56 PM by Manu Pillai   [ updated Jul 29, 2011 5:27 PM ]

With the growth in multi-core SoC's, CPU's and GPU's, we tend to see dedicated-function cores in embedded areas, and GPGPU and CPU-GPU in larger systems. Interestingly enough, with the right tools and methodologies to maximize the efficiency of memory transfers with packed read/writes, thread coarsening and super-queueing combined with global memory spaces, tremendous improvements are possible. 

TechRev is proud to be the global Business Development Partner of MulticoreWare Inc - MCW -  ( www.multicorewareinc.com ) - the leading provider of heterogenous multicore computing tools, IP and related professional services. With several PhD's on staff in a range of disciplines, MCW provides technical, financial computing, video and image processing in a number of domains, in addition to signal processing and bio-informatics. MCW has already made its mark in a number of areas with clients that range from Mil-Aero to Broadcast to Genome research and analytics, where MCW uses the tools it develops, to also provide top notch professional services in software engineering. MCW has also recently delivered a number of software development tools for AMD's Fusion family of APU's that integrate CPU and GPU cores tightly and efficiently. 

1-3 of 3