Design Alternatives for Ultraperformance Parallel Computers
Design Alternatives for Ultraperformance Parallel Computers
Jacob T Schwartz
The book Design Alternatives for Ultraperformance Parallel Computers was written by author Jacob T Schwartz Here you can read free online of Design Alternatives for Ultraperformance Parallel Computers book, rate and share your impressions in comments. If you don't know what to write, just answer the question: Why is Design Alternatives for Ultraperformance Parallel Computers a good or bad book?
What reading level is Design Alternatives for Ultraperformance Parallel Computers book?
To quickly assess the difficulty of the text, read a short excerpt:
The 'dimensions' of parallel supercomputer design can be listed, roughly in order of diminishin g importance, as follows: (1) 5/AfD vj. MIMD ' A SIMD (singlc-instruction-multiplc-data) machine consists of multiple processors (or of a few very fast pipelined processing elements), all of which are fed synchronoiisly by a single instruction stream, but which operate in parallel on independent streams of data. Vector machines like the CRAY I or Hitachi 810, and synchronous array machines like the E...liac IV and NASA/Goodyear MPP typify this class. An MIMD computer consists of multiple processors (or fast timeslioed processing elements) driven by independent instruction streams, and capable of branching inde^ndently, but also able to pass data and synchronization signals between processors, perhaps via a shared memory. The Dcnclcor HEP, Maryland ZMOB, Cal Tech Homogeneous Hypercube, and numerous other university machines belong to this class. SIMD machines are generally more efficient for applications characterized by very regular patterns of processing, in part because their 'lockstep' mode of operation can eliminate much of the message-passing overhead needed when all the processors of an SIMD machine must proceed through a computation in dose synchrony.
User Reviews: