上一篇:Pre-Petascale同构超级计算机仍然面临未解决的高性能计算挑战
Still Unsolved High-Performance Computing Challenges for up to Pre-Petascale Homogeneous Supercomputers
摘要:
根据HPLinpack基准测试,兆兆级前的高性能计算机(HPC)的实际性能可以达到400 Pflop/s以上。对于纳米科学和量子生物学来说,要求那些基于量子物理算法的程序代码难以理想地并行化。这样的并行代码在万亿级性能集群上达到了它们的极限。标准的Amdahl定律对代码并行化的建议使并行代码开发的下一步的重点和计划变得复杂。在这份报告中,我们集中在三个关键的应用领域,这是高度并行化:HPC基准,量子计算模拟器和汽车Parinello分子动力学。根据计算结果,我们总结了Amdahl定律和并行加速比的性能成就与超级计算机与前千万亿次同构HPC硬件。我们得出结论,作为一个通用的计算机,前千万亿次超级计算性能同构硬件仍然有基本的挑战,必须解决的研究人员或开发人员,以有效地使用它们。
英文摘要:
Pre-exascale High Performance Computers (HPC) can reach more than 400 Pflop/s real perfor-mance according the HPLinpack benchmarks. For nanoscience and quantum biology there are requirements for those program codes based on quantum physics algorithms which is difficult to ideally parallelize. Such parallel codes reach their limitations at terascale performance clus-ters. The standard Amdahl's law suggestions for code parallelization complicates focusing and planning for the next step the parallel code developments. In this report we focused on a three key applications domain which are highly parallelizable: HPC benchmarks, quantum compu-ting simulators and Car-Parinello molecular dynamics. According the results we summarize the Amdahl's Law & Parallel Speedup performance achievements with supercomputer with pre-petascale homogeneous HPC hardware. We conclude as an universal computer the pre-petascale supercomputing performance homogeneous hardware still has the basic challeng-es which must be addressed by the researchers or developer in order efficiently to use them.
机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。