GPU Cluster 2 ACCRE Vanderbilt（GPU集群2 ACCRE范德比尔特）.pdf

GPU cluster computing Advanced Computing Center for Research and Education 1 What is GPU Computing • Use of GPU(s) to solve scienfic compung problem. Due to their massive-parallel architecture, using GPUs enables the compleon of computaonally intensive assignments much faster compared with convenonal CPUs. Hybrid CPU-GPU system: • GPU: Computaonally-intensive (massively parallel) part • CPU: sequenal part 2 Why is the GPU So Fast? • The GPU is specialized for compute-intensive, highly data parallel computaon (owing to its graphics rendering origin) • More transistors can be devoted to data processing rather than data caching and control flow • Where are GPUs good: high arithmec intensity (the rao between arithmec operaons and memory operaons) 3 Generic Multicore Chip Processor Memory Processor Memory GlobalMemory • Handful of processors each supporng ~1 hardware thread • On-chip memory near processors (registers , cache) • Shared global memory space (external DRAM) 4 Generic Manycore Chip Processor Memory Processor Memory • • • GlobalMemory Many processors each supporng many hardware threads On-chip memory near processors (register, cache, RAM) Shared global memory space (external DRAM) 5