Merrimac: Supercomputing with Streams

By William (Bill) J. Dally, F. Labonte, Arindam Das, P. Hanrahan, Jung Ho Ahn, J. Gummaraju, Mattan Erez, Nuwan Jayasena, Ian Buck, T. Knight and 1 more. International Conference on Software Composition, 2003

347 citations recorded by the source corpus.

Subfields: Computer Networks and Communications, Hardware and Architecture

View Paper: https://www.semanticscholar.org/paper/5ab9976fed0f05d1c894c0cc1c33b54cc76d3bb5

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Networks and Communications

In Hardware and Architecture

Citations in the corpus

Dynamic Warp Formation and Scheduling for Efficient GPU Control Flow (2007, background): “However, .nding effective ways to leverage process technology scaling for improving the performance of real world applications has become increasingly challenging as power limitations restrict clock frequency scaling [13].”
Analyzing CUDA workloads using a detailed GPU simulator (2009, background): “Merrimac [8] and Imagine [3] are both streaming processor architectures developed at Stanford.”
PyCUDA and PyOpenCL: A scripting-based approach to GPU run-time code generation (2009, background): “Graphics Processing Units (GPUs) [6, 28, 44] promise tremendous advantages in throughput over conventional processor architectures, ideally resulting in a large reduction of execution time for suitable compute-or bandwidth-bound algorithms.”
Nodal discontinuous Galerkin methods on graphics processors (2009, positive): “tations on streaming hardware. Building on this work, Barth et al. [1] already predicted promising performance for two-dimensional DG on a simulation of a the Stanford Merrimac streaming architecture [6]. Nowadays, compute abstractions are becoming less encumbered by their graphics heritage [17, 18]”
Brook for GPUs: Stream Computing on Graphics Hardware (2004, methodology): “Dally et al. [2003] explain how stream programming encourages the creation of applications with high arithmetic intensity , the ratio of arithmetic operations to memory bandwidth.”
High-throughput sequence alignment using Graphics Processing Units (2007, background): “These applications are well suited for running on graphics hardware because they have high arithmetic intensity – the ratio of time spent performing arithmetic to the time spent transferring data to and from memory [ 19 ].”
Conservation cores: reducing the energy of mature computations (2010, background): “Specialized designs exist for applications and domains such as cryptography [39], signal processing [12, 16], vector processing [3, 10], physical simulation [2], and computer graphics [4, 28, 30].”