InvasIC Seminar: Vorträge im SFB/TRR 89 (HD 1280 - Video & Folien)

Some data- and compute-intensive applications can be accelerated by offloading portions of codes to platforms such as GPGPUs or FPGAs. However, to get high performance for these kernels, it is mandatory to restructure the application, to generate adequate communication mechanisms for the transfer of remote data, and to make good usage of the memory bandwidth. In the context of the high-level synthesis (HLS), from a C program, of hardware accelerators on FPGA, we show how to automatically generate optimized remote accesses for an accelerator communicating to an external DDR memory. Loop tiling is used to enable block communications, suitable for DDR memories. Pipelined communication processes are generated to overlap communications and computations, thereby hiding some latencies, in a way similar to double buffering. Finally, data reuse among tiles is exploited to avoid remote accesses when data are already available in the local memory.

Some data- and compute-intensive applications can be accelerated by offloading portions of codes to platforms such as GPGPUs or FPGAs. However, to get high performance for these kernels, it is mandatory to restructure the application, to generate adequate communication mechanisms for the transfer of remote data, and to make good usage of the memory bandwidth. In the context of the high-level synthesis (HLS), from a C program, of hardware accelerators on FPGA, we show how to automatically generate optimized remote accesses for an accelerator communicating to an external DDR memory. Loop tiling is used to enable block communications, suitable for DDR memories. Pipelined communication processes are generated to overlap communications and computations, thereby hiding some latencies, in a way similar to double buffering. Finally, data reuse among tiles is exploited to avoid remote accesses when data are already available in the local memory.

The race for increasing computing performance implied a dramatic increase in power which represents the main wall for further increase in computing. The “multi-core revolution” has shown a way out. Thus, today’s state-of-the-art architectures in embedded computing are based on heterogeneous multi-core architectures with application specific optimized accelerators. Unfortunately the immense computing power of such multi-core architectures brings as negative effect an increased demand on bandwidth and memory, denoted as bandwidth and memory walls. In this talk we will discuss two topics strongly related to the memory and bandwidth wall. First, the impact of memories and data transfers on metrics to compare different algorithms and implementations in the context of wireless baseband processing architectures. Second, the design space and potential of 3D DRAM architectures and multi-channel DRAM controllers.

The race for increasing computing performance implied a dramatic increase in power which represents the main wall for further increase in computing. The “multi-core revolution” has shown a way out. Thus, today’s state-of-the-art architectures in embedded computing are based on heterogeneous multi-core architectures with application specific optimized accelerators. Unfortunately the immense computing power of such multi-core architectures brings as negative effect an increased demand on bandwidth and memory, denoted as bandwidth and memory walls. In this talk we will discuss two topics strongly related to the memory and bandwidth wall. First, the impact of memories and data transfers on metrics to compare different algorithms and implementations in the context of wireless baseband processing architectures. Second, the design space and potential of 3D DRAM architectures and multi-channel DRAM controllers.

Urs Gleim, Leiter des Parallel Processing Systems Teams bei der zentralen Forschung der Siemens AG (Corporate Technology), diskutiert in diesem Vortrag die technischen Herausforderungen der nächsten Jahre aus industrieller Sicht. Der Schwerpunkt der Betrachtung liegt auf der Nutzung neuer, insbesondere paralleler, Hardwarearchitekturen. Abschließend stellt er ausgewählte aktuelle Arbeiten der Siemens Corporate Technology in diesem Umfeld vor.