Skip to main content
. 2022 Nov;36(5-6):587–602. doi: 10.1177/10943420221121804
Performance attribute Our submission
Category of achievement Time-to-solution, scalability
Type of method used Machine learning
Results reported for Whole application with and without I/O
Precision reported Mixed precision (FP16 and FP32)
System scale Measured on full-scale system (summit)
Measurement mechanism Internal timers, DeepSpeed FLOPS profiler