PARTANS
Thibaut Lutz, Christian Fensch, Murray Cole · ACM Transactions on Architecture and Code Optimization · 2013
GPGPUs are a powerful and energy-efficient solution for many problems. For higher performance or larger problems, it is necessary to distribute the problem across multiple GPUs, increasing the already high programming complexity. In this article, we focus on abstracting the complexity of multi-GPU programming for stencil computation. We show that the best strategy depends not only on the stencil operator, problem size, and GPU, but also on the PCI express layout. This adds nonuniform characteristics to a seemingly homogeneous setup, causing up to 23% performance loss. We address this issue with an autotuner that optimizes the distribution across multiple GPUs.
Cited by 65 works. doi:10.1145/2400682.2400718 Open-access PDF
Authors
References
- Patterns for parallel programming
- PATUS: A Code Generation and Autotuning Framework for Parallel Iterative Stencil Computations on Modern Microarchitectures
- An auto-tuning framework for parallel multicore stencil computations
- Stencils and Problem Partitionings: Their Influence on the Performance of Multiple Processor Systems
- Concurrent Processing for Scientific Calculations.
- Physis
- Implementing the Himeno benchmark with CUDA on GPU clusters
- Achieving a single compute device image in OpenCL for multiple GPUs
- Mint
- Exploring the multiple-GPU design space