Clear Speed CSX 620 Overview References Clear Speed

  • Slides: 68
Download presentation
Clear. Speed CSX 620 Overview

Clear. Speed CSX 620 Overview

References • Clear. Speed Technical Training Slides for Clear. Speed Accelerator 620, software version

References • Clear. Speed Technical Training Slides for Clear. Speed Accelerator 620, software version 3. 0, Slide Sets 1 -6, Presentor: Brian Summers (senior engineer), December 2007 – Acknowledgement: Many slides used here are from Slide Set 1. • Clear. Speed Introductory Programming Manual, January 2008

Topics • Overview of Clear. Speed Board – Clear. Speed Technology Company – Accelerators

Topics • Overview of Clear. Speed Board – Clear. Speed Technology Company – Accelerators – Clear. Speed and HPC – Hardware Overview – Performance – Software Development Kit (SDK) – Application Examples – Help and Support • Topics omitted from Clear. Speed Overview – Installing Hardware and Software – Most topics in SDK overview - Some will be covered later • E. g. , Cn Language, Cn Libraries, compiler, debugging Cn, assembler, linker, simulator, graphics profiler, libraries. – Moving Data – Tuning Tips

Clear. Speed CSX 600 Accelerator Board • A PCI-X card equipped with two Clear.

Clear. Speed CSX 600 Accelerator Board • A PCI-X card equipped with two Clear. Speed CSX 600 coprocessors

Performance Specifications of CSX 600 • Sustained double-precision performance of 25 GFLOPS on DGEMM

Performance Specifications of CSX 600 • Sustained double-precision performance of 25 GFLOPS on DGEMM • 10 W max power consumption • 250 MHz clock speed • Transfer speed of internal memory: 96 Gbyes/s • Transfer speed of external memory: 3. 2 Gbytes/s

Multi-threaded Array Processing (MTAP) architecture of CSX 600 • Mono execution unit - process

Multi-threaded Array Processing (MTAP) architecture of CSX 600 • Mono execution unit - process non-parallel data - handle program flow control • Poly execution unit - 96 PEs - 6 KB SRAM - dual 64 -bit FPU - integer ALU - 32/64 -bit floating-point multiplier & adder - 128 B register files

Cn language • Similar to standard C • Main difference is poly variables •

Cn language • Similar to standard C • Main difference is poly variables • Example code: #include <stdiop. h> // Output support #include <lib_ext. h> // Extra functions to support features of hardware int main() { poly int n; n = get_penum(); // individual PE number printfp("PE number: %dn", n); // Output different message per PE return 0; } • poly short get_penum(): number of current PE • mono short get_num_pes(): number of PEs on CSX processor

Note: Do not contact Clear. Speed about a homework problem, answering a question, etc.

Note: Do not contact Clear. Speed about a homework problem, answering a question, etc. They expect these questions to be professional level questions from owners of their CSX 620 boards – not student questions about their class or homework.