PPT-CPUs, GPUs, accelerators and memory

Andrea Sciabà On behalf of the Technology Watch WG HOW Workshop 1822 March 2019 Jefferson Lab Newport News DRAFT Introduction The goal of the presentation is to give a broad overview of the status and prospects of compute technologies. Avg Access Time 2 Tokens Number of Controllers Average Access Time clock cyles brPage 16br Number of Tokens vs Avg Access Time 9 Controllers Number of Tokens Average Access Time clock cycles brPage 17br brPage 18br ITS Research Computing. Mark Reed . Objectives. Learn why computing with accelerators is important. Understand accelerator hardware. Learn what types of problems are suitable for accelerators. Survey the programming models available. 2010 Turing Award Recipient. Chuck Thacker. Improving the future by examining the past. Thesis. Computer architecture has hit a wall.. This will require us to build future systems in new ways.. New systems will require changes in the way we program them.. Sathish. . Vadhiyar. List Ranking on GPUs. Linked list prefix computations – computations of prefix sum on the elements contained in a linked list. Irregular memory accesses – successor of each node of a linked list can be contained anywhere. Proposed Work. This . work aims . to enable efficient dynamic memory management on NVIDIA GPUs by utilizing a sub-allocator between CUDA and the programmer. This work enables Many-Task Computing applications, which need to dynamically allocate parameters for each task, to run efficiently on GPUs.. ECE . 751. Brian Coutinho. ,. David Schlais. ,. Gokul Ravi. &. Keshav . Mathur . Summary. Fact. : Accelerators gaining popularity - to improve performance and energy efficiency. Problem. : Accelerators with scratchpads require DMA calls to satisfy memory requests (among other overheads). Supercomputing. The Next wave of HPC. Presented by Shel Waggener. HP Materials from Marc Hamilton. June. , . 2011. © Copyright 2011 Hewlett-Packard Development Company, L.P.    . GPUs – changing the Economics of Supercomputing. Historical/Conceptual. Intel. AMD. ?. CPU Core Components. The Man in the Box. External Data Bus. Registers. AX, BX, CX, DX. Code book. Clock. System crystal. Back to the External Data Bus. Memory. Memory and RAM. Introduction. Multiprocessing.  is the use of two or more . central processing units.  (CPUs) within a single . computer . system. . The . term also refers to the ability of a system to support more than one processor and/or the ability to allocate tasks between them. ECE . 751. Brian Coutinho. ,. David Schlais. ,. Gokul Ravi. &. Keshav . Mathur . Summary. Fact. : Accelerators gaining popularity - to improve performance and energy efficiency. Problem. : Accelerators with scratchpads require DMA calls to satisfy memory requests (among other overheads). Twiss. Parameterization (left out yesterday). The transfer matrix over any period (. s. . s+C. ). must be stable over infinite passes. Must have purely imaginary eigenvalues. Any such matrix can be represented as. printatemenhilebuggingmighhangiminxecutionenough to me race disappearThypical way to avois is to uococnsurutuxclusiothnl can executeimhiakecenariovpossibleorrectlockeersioovodusineotnumbered8void1017W 2. Accelerators. for . Medicine. Maurizio Vretenar, CERN. Academic. Training, 12 . June. 2018. Accelerator and Society. 12/6/2018. M. Vretenar, Accelerators for Medicine. 3. Research. 6%. Particle. Performance portability: An implementer’s Perspective. Versatility of HPE Cray Programming Environment (CPE). Platforms from HPE/Cray. Supported by CPE. 2. AMD CPUs and NVIDIA GPUs. AMD CPUs and AMD GPUs.