The promise of heterogeneous computing: an insider look at the HSA Foundation

There has been a lot of talk recently about the HSA Foundation that AMD, ARM, Imagination Technologies, MediaTek and Texas Instruments have set up. Articles in the Inquirer or PC Mag have looked at the implications of this union which was announced through the official press release at the AMD Fusion Developer Summit  in Bellevue, WA.

Let’s spend some time thinking about what this really means for the concept of heterogeneous computing.

Patrick Moorhead, principal analyst at Moor Insights & Strategy said: “This [foundation] will place CPUs, GPUs, and other accelerators as equal citizens across computing platforms. If the HSA Foundation gets support from Microsoft with Visual Studio, Google with Android SDK, and Apple with XCode, this would be a real game changer.”

 

Key Founders of the HSA Foundation*

Fighting the right fight for GPU compute

Here at Imagination we’ve always felt that a SoC should be a collection of processors in a peer-based system where each of them can accomplish certain tasks that the others are not architecturally suited for.

The best example of this is GPU compute, where the central processing unit (CPU) simply cannot cope with the large amount of parallel data it needs to process. Intrinsically a sequential design, the CPU is better suited for control and scheduling tasks. This is why the GPU, a true parallel engine, has to step in when intensive data processing tasks are tackled. Its architecture has evolved from the early days of OpenGL 1.0 when it was designed specifically to perform floating point calculations – fundamental to 2D image drawing and 3D graphics – to a fully programmable TBDR (tile-based deferred rendering) unit that uses shaders to calculate rendering effects on graphics hardware. The impact of graphics acceleration has been so remarkable, that we now have photo realistic game engines or smooth, touch-responsive UIs.

Because of their very nature, CPUs and GPUs have traditionally been used for entirely different things: one for step by step control, the other for parallel texture mapping and polygon rendering. We must avoid thinking about heterogeneous computing as tightly coupling the two and, instead, focus on their different use-cases, where their standout features can be best put to use. Simply adding more cores of different sizes together and hoping they will cope is not only area and power inefficient but also unwise in terms of scalability and performance. The same concept applies for other processors that a typical SoC might include, such as VPUs (video processing units) and RPUs (radio processing units).

As we’ve constantly pushed the boundaries of what GPU computing can accomplish in the mobile space, we feel HSA Foundation is a right choice to bring it to a broader audience. The HSA Foundation is fundamentally an open group with a firm set of goals: an expanding body that will reach everyone from IP vendors and foundries to device manufacturers and OS vendors, from software developers to universities and academia, bringing GPU computing at the forefront of modern architectures, improving and expanding use cases of this concept by developing tool chains and libraries and innovating OS runtimes.

Also, it aims to incorporate support for special purpose hardware accelerators such as DSPs and rasterizers and promote the advantages of using GPU compute to develop rich content and applications by establishing a complete ecosystem of developers, SDKs, documentation, training and support.

Finally, as the foundation recognizes the importance of academia in this process, it promises to help universities and schools quickly adopt these new tools so that it can drive innovation and nurture the talent required for this program to be a success.

Heterogeneous computing: it’s all coming together

Just by having a look over what HSA tries to accomplish, one can see how Imagination fits in perfectly. We’ve advocated these concepts for years, by pushing the GPU onto the SoC at a time when this was unheard of. And we are now envisioning a new wave of systems-on-chip where RF processing is integrated in as well. This is beginning to be recognized as one of the leading trends in the industry.

We have continued to be at the forefront of heterogeneous processing architectures because they are the future of computing. We have a broad range of IP which covers everything from graphics and video, processors, radio and communications which enable our partners to deliver the best solutions on the market. Our PowerVR Series6 GPU IP family is the best example for this: it provides the best graphics performance in its class but has also been designed with GPU compute in mind, supporting all major industry standards including: OpenCL, Renderscript Compute and DirectCompute (whereas some suppliers have decided to either go with a limited set or just ignore some of them altogether). By introducing a cluster-based architecture, we are able to offer a scalable platform that can be optimized for best results in performance, power and area without sacrificing anything in terms of features.


 

Just imagine what a developer with access to hundreds of GLOPS (or even TFLOPS, thanks to our revolutionary architecture) can do with such a powerful computing platform.

Moreover, look no further than Meta general-purpose CPU to discover how parallel processing can be successfully implemented by using separate threads to run a fully-featured operating system like Linux or Android and also DSP tasks like audio decoding or digital signal demodulation. Whereas others have gone with an SMP implementation, Meta is a multi-threaded CPU that can really go into overdrive thanks to the AMA (Automatic MIPS allocation). It’s already been proven to be the best solution for a range of home and in-car entertainment solutions from Pure, Sony, Roberts, Alpine and other major brands.

Imagination has had a clear vision of what heterogeneous platforms should be. We’ve taken this concept and implemented it in an entire line of IP cores: PowerVR Series5/5XT and Series6 GPUs, multi-threaded Meta CPUs, PowerVR Series3 video encoders and decoders and Ensigma UCCP Series2 RPUs. And as we continue to make progress in ray-tracing technologies, we intend to bring RTUs (ray-tracing units) to the party as well.

On top of that, Imagination’s IP has been integrated and included in hundreds of millions units shipped across numerous markets. We’ve also been involved in launching and contributing to a number of industry standards such as all open Khronos APIs as well as our own PVRTC and OpenRL standards, to name just a few.

Markets that will benefit from HSA Technology*

The foundation’s role is primarily to make HSA an open standard so that developers will be able to deliver a killer set of apps that will revolutionize the industry once again. But also, as there are several key players involved, it looks to be a democratic organization that is neutral and inclusive at the same time to all computing architectures so that tools will be able to evolve quickly and efficiently.

How the HSA Foundation can help you

There are numerous benefits to being part of this, as members are able to license important IP that will be developed as part of this mutual effort as well as participate in innovative initiatives that will drive new products. Also, members will be able to lower costs by taking advantage of the work that will be put in tool chains and operating systems and have access to a strong ecosystem of developers.

So, for example, if you are a semiconductor company looking for the next great thing, HSA can offer you the ability to lower development costs together with neutral platform governance, meaning you can have a say where heterogeneous architectures are headed into. Also, you can tap into a large and diverse app ecosystem which automatically means technical sustainability through a close alignment with key industry initiatives. For device manufacturers and platform vendors, HSA will be a way to achieve commercial success by connecting with semiconductor members, providing true differentiation and strong software support.

Have a look over their website to find out more and keep coming back to our blog for more information and developments on this exciting partnership.

 

 

*Images courtesy of the HSA Foundation, all rights reserved

, , , , , , , , ,

  • http://metallboard.net.ru metaseeing

    Интересно, а можно у вас брать посты с RSS? Ссылку на источник обязательно поставлю.
    (Edit: Interesting, I wonder if we can take the posts with RSS? We would reference the source.)

  • http://www.imgtec.com Alex Voica

    Hi,

    Sure, as long as you link back to our blog. We are also working on implementing some advanced features to our blog, which will make sharing much easier.
    But in the future, please leave comments in English only, it is somewhat difficult to translate from and to Russian.

    Regards,
    Alex.

    Привет,

    Конечно, если вы ссылку на наш блог. Мы также работаем над реализацией некоторых дополнительных возможностей в нашем блоге, которые сделают совместное гораздо проще.
    Но в будущем, пожалуйста, оставьте комментарии только на английском языке, это довольно трудно переводить на русский.

    С уважением,
    Алекс.

  • http://gamemusic.net.ru sufinish

    Очень красиво, вот бы у нас так сделали
    (Edit: Very nice, we would have done so)