Mitigating the Memory Bottleneck With Approximate Load Value Prediction

Amir Yazdanbakhsh, Bradley Thwaites, Hadi Esmaeilzadeh, Gennady Pekhimenko, Onur Mutlu, Todd C. MowryIEEE Des. Test 2016 · DOI · OpenReview

Abstract

Limited memory bandwidth and long access latency are serious GPU performance bottlenecks. For emerging applications that are inherently error resilient, this paper proposes to predict the values of safe-to-approximate loads and drop a certain fraction of the cache misses with predicted values to improve performance and energy efficiency.
BibTeX
@article{yazdanbakhsh2016mitigating,
  author = {Amir Yazdanbakhsh and Bradley Thwaites and Hadi Esmaeilzadeh and Gennady Pekhimenko and Onur Mutlu and Todd C. Mowry},
  title = {{Mitigating the Memory Bottleneck With Approximate Load Value Prediction}},
  journal = {IEEE Des. Test},
  year = {2016},
  doi = {10.1109/mdat.2015.2504899},
}

Loading related content...

Figure