AMD (s amd), ARM (S armh), Texas Instruments (s txn) and two smaller chip firms have teamed up to create a nonprofit that will try to unseat Intel’s (s intc) x86 dominance in computing. They have formed the Heterogeneous Systems Architecture Foundation, which will standardize a single architecture for low-power computing as well as simplify the parallel-programming model used with multicore graphics processors and other systems on a chip.
So how many chip makers does it take to unseat Intel? So far this consortium counts five including MediaTek (one of the largest wireless chip makers after Qualcomm) and Imagination Technologies. Notable absentees are Qualcomm (s qcom) and Nvidia, (s nvda) although John Taylor, the director of Product Marketing for AMD, says the consortium is reaching out to those companies as well. The consortium aims to offer standardized boards for the server, mobile and embedded markets, and it has a definite focus on graphics processing units, or GPUs.
Why the GPU and CPU combo makes sense
The creation of the HSA Foundation makes a lot of sense, as graphics have become more-essential elements in our day-to-day computing. The videos we watch, the photos we snap and send, and even the growing importance of a highly visual design in sites such as Pinterest make a GPU essential for everyday computing. Even back in 2008, software makers and PC manufacturers were shifting more and more work to the GPU inside laptops. Outside the consumer realm, for tasks such as video transcoding and certain types of simulations, graphics processors were gaining ground. Even top supercomputers had them.
GPUs not only handle graphics more efficiently than CPUs, which means the batteries can last a bit longer or a supercomputer consumes less power, but also can tackle parallel computing jobs that can be broken into smaller pieces. However, writing software to divvy up those jobs is hard. Nvidia pioneered CUDA, a tool that helped compile traditional C code into something that could work on a GPU, and then OpenCL and other efforts followed. The HSA Foundation wants to reside under those efforts by creating software and a new compute architecture that manages what tasks are right for
and then sent to the CPU and what tasks should be performed by the GPU.
CPUs are sporks, and we’re ready to graduate to specialty flatware
This makes sense, but it’s a fundamental shift in computing because it no longer assumes the CPU is the default player. When most of your computing is visual, using general-purpose CPUs is like using a spork to eat pudding. If there’s a spoon available, wouldn’t that be better? If this consortium is successful, it’s likely we’re going to see it move beyond optimal CPU and GPU architectures and software to other heterogeneous combinations.
For example, Texas Instruments is a big manufacturer of digital-signal processing chips that are really great at math. A DSP combined with a CPU core could become a fork designed for a certain type of job. And the possibility of an ARM Cortex processor combined with something is a possibility as well. The HSA press release mentions both the ARM Cortex line — which acts as the brains inside smartphones — as well as the ARM Mali line of graphics cores. Taylor likens the options not to silverware but to the Avengers, saying, “You used to need just one superhero but now you need a team of them.”
HSA in the real world
Talk of silverware and superheros can help explain the concept, but the results of the foundation’s work won’t hit the streets until 2014, when AMD releases a line of chips that will use the new architecture. Taylor notes that by 2014 AMD will offer APUs (its version of a combo GPU and CPU chip) and GPUs that are HSA-enabled. Already the foundation is working with application providers such as Microsoft (s msft) and Adobe (s abde) to optimize their programs to take advantage of the new chips.
AMD’s focus will initially be on the workstation and PC market, but it will also expand the concept and work to its server division as well. Meanwhile with firms like Texas Instruments involved, the likelihood is high that we will see this concept in smartphones. TI’s OMAP application processors are currently the brains inside the Samsung Galaxy Nexus and the Motorola RAZRs (s goog). The involvement of Imagination and ARM also means the embedded market, which includes video gaming consoles, DVRs and other home devices, could also get HSA-compliant chips.
“We’re not rolling something out there that requires an extreme effort,” Taylor says. “This doesn’t require much from the operating system, it just benefits the apps sitting on top of that OS. The real beauty is that it lowers the development times and lets the developers work with familiar tools.”
Should this work out, for consumers the real beauty would be more-efficient computing and longer battery life on their devices, while in the data center it provides a foundation to move from a CPU-dominated world to one where systems administrators or application architects can use the best processor for each job. We will discuss the idea more at our Structure 2012 conference next week with AMD and others who might end up using the architectures described.