A characterization study showing MLPerf v0.5 training benchmarks are distinct from DAWNBench and DeepBench, scale differently across GPUs, and are sensitive to interconnect topology, mixed precision, and compiler optimizations.
Findings of the 2017 conference on machine translation (wmt17),
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2019 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Demystifying the MLPerf Benchmark Suite
A characterization study showing MLPerf v0.5 training benchmarks are distinct from DAWNBench and DeepBench, scale differently across GPUs, and are sensitive to interconnect topology, mixed precision, and compiler optimizations.