At the Fusion Developer Summit, AMD announced the formation of the HSA Foundation in conjunction with ARM, Imagination, MediaTek, and Texas Instruments (TI).
This follows a previous announcement in January to rename the Fusion System Architecture to the Heterogeneous System Architecture (HSA). The effort is born from AMD’s processor architecture, referred to as accelerated processing unit (APU), that combines CPU and GPU cores on a single chip and the industry’s efforts to offer GPU acceleration, also referred to GPGPU compute.
GPU acceleration all hinges on the ability to efficiently program the GPU to handle certain workloads better suited for the massively parallel graphics cores.
The announcement of the HSA Foundation follows in the footsteps of many other industry efforts to offer GPU acceleration. The most recent and notable are CUDA developed and promoted by AMD’s chief competitor in discrete graphics, NVIDIA, and OpenCL, an industry standard maintained by the Khronos Group and receiving support in recent operating system releases from both Apple and Microsoft.
In fact, the announcement sounds very similar to previous announcement by the CUDA camp in the effort to support multiple programming languages while offering a superset of OpenCL functions. The HSA foundation announced plans to support more common programming standards, such as C, C++, C++ AMP, JAVA, and Python. Unlike CUDA, however, the HSA Foundation will focus on processor architectures with on-chip GPUs, which are now commonplace in computing, CE, and embedded applications. But, the most notable portion of the HSA Foundation announcement is the partners.
While CUDA and OpenCL have received much of their support from the computing and scientific communities, the HSA is supported by key players in the ARM and mobile CE ecosystem. The mobile community has been exploring the potential for using GPU acceleration for the past several years, but applications have been few because most mobile applications do not require the computing performance of many high-end computing applications.
However, the potential for continuously processing information from a host of sensors combined with processing applications with that data over high-speed wireless networks is likely to create the need to parallel processing and GPU acceleration. For the ARM camp, teaming up with AMD offers to expertise and lessons learned from implementing GPU acceleration. For AMD, the benefit may be even greater as it seeks to extend its boundaries into new markets like embedded applications under a new management team.
Despite the potential synergies of AMD and the ARM camp, the HSA still faces daunting challenges. The first is that there are already standards available, primarily through CUDA and OpenCL in the computing space, and even these efforts have struggled to gain wide use in programming applications within the broader computing ecosystem, not to mention the infancy to GPU acceleration in the mobile environment.
As with the other standards, the success of the HSA Foundation will hinge on the group’s ability to gain support from the rest of the compute and CE ecosystems, including but not limited to the operating systems developers, application programmers, device OEMs, and even service providers.
There is no doubt that there is a growing need for GPU acceleration, but the announcement marks a new chapter in the mobile devices that are bridging the gap between computing and traditional consumer electronics.
David Patterson, known for his pioneering research that led to RAID, clusters and more, is part of a team at UC Berkeley that recently made its RISC-V processor architecture an open source hardware offering. We talk with Patterson and one of his colleagues behind the effort about the opportunities they see, what new kinds of designs they hope to enable and what it means for today’s commercial processor giants such as Intel, ARM and Imagination Technologies.