AMD today announced it is enabling its accelerated processing units (APU) for next-generation servers through advancements in software tools developed by AMD and in collaboration with technology partners and the open source community. In his keynote address at APU13, AMD Corporate Fellow Phil Rogers highlighted the progress AMD has made in both developing software internally and assisting other developers to take advantage of the capabilities of AMD APU technology, which combines the AMD Radeon graphics processing engines with x86 computational processing on a single chip.
Heterogeneous System Architecture (HSA), maintained by the HSA Foundation, is a system architecture that allows accelerators and graphics processors to operate at the same processing level as the system's CPU. HSA allows programs to use the graphics processor for floating point calculations without separate memory or task scheduling, leading to better low-level system integration and a corresponding increase in speed. The technique is commonly used in mobile devices using the "System on a Chip" (SoC) concept.
AMD has developed tools for software developers to take advantage of the tangible benefits that HSA provides. HSA enables the CPU and GPU to work in harmony on a single piece of silicon, seamlessly moving the right tasks to the best-suited processing element with no data transfer speed penalties, and makes more memory available to the GPU so that complex processing tasks can fit in a single node.
"Servers must be efficient, scalable and adaptable to meet the compute characteristics of new and changing workloads. Software applications that leverage server APUs are designed to drive highly efficient, low-power, dense server solutions optimized for highly parallel and multimedia workloads," said Suresh Gopalakrishnan, corporate vice president and general manager of the Server Business Unit of AMD. "We have evolved our processor roadmap to support this opportunity, and now we are showcasing how the APU software ecosystem is gaining momentum and what developers can do to participate."
New tools being made available using the new technologies are joint Oracle and AMD Project Sumatra, the GCC/HSA project bringing the technology to the GCC compiler for Linux, integration of AMD clMath libraries to the Open Source community, and a fast math library by AccelerEyes that utilizes clMath for GPU computing and offers an easy-to-use API for Windows or Linux developers called ArrayFire 2.0. Perhaps most important is the release of CodeXL 1.3, a developer's tool suite for Windows and Linux that features remote debugging and profiling to enable server application developers to use the new libraries.