HPC system expert Penguin Computing has just announced that the company has finalized the installation of the world's first HPC cluster powered by AMD Accelerated Processing Units (APUs) at Sandia National Labs in Albuquerque, New Mexico.
The system is comprised out of 104 servers that are interconnected through a QDR InfiniBand fabric capable of delivering a peak performance of 59.6 TFLOPs.
At the heart of the HPC cluster stands Penguin's new Altus 2A00 compute platform that relies on consumer AMD Fusion APUs based on the Llano architecture
As many of you know, AMD's Llano chips include two or four x86 computing cores, Radeon HD graphics cores with a maximum of 400 stream processors as well as a dual-channel DDR3 memory controller and a PCI-E interface.
Penguin Computing hasn't revealed any additional information regarding the hardware used in the Altus 2A00 machines, but it did point out that AMD's Fusion APUs feature a simpler programming model than conventional GPU server architectures.
The reason has to do with the fact that parallel multiprocessors share the same physical memory space with CPU cores, which removes data movement bottlenecks between GPU and the main system memory as well as data duplication.
"We are interested in research on next generation computer architectures and look forward to collaborating with Penguin and AMD
to advance power-efficient computing strategies," said James Ang, Manager of the Scalable Computer Architectures department at Sandia National Labs.
"This first of a kind cluster of Altus 2A00 servers will support our exploration of advanced programming models like OpenCL, which seamlessly map MPI applications to the CPU and GPU cores, and research into system software support for advanced data movement capabilities," concluded the company's rep.
Penguin Computing hasn't mentioned if it was contracted by other organizations to build similar HPC systems
based on AMD's Fusion APUs.