Table 2 Time benchmarks for simulation of training data, training, inference, and model estimation with the 2D-Fit
From: A deep learning approach to real-time Markov modeling of ion channel gating
Task | Hardware | Time consumption |
|---|---|---|
Simulation of training data 1 million 2D-histograms time series length: 10 million samples | HPC-cluster (Intel Xeon Platinum 8360Y) | 100 node-hours (72 cores per node) |
Training 1 million training samples (2D-histograms) global batch size: 1024 | HPC-cluster (NVIDIA A100 GPU) | 2 node-hours (8 GPUs per node) |
2D-Fit21 time series length: 10 million samples ensemble size: 64 population size: 800 | HPC-cluster (Intel Xeon Platinum 8360Y) | 16–48 node hours (72 cores per node) |
Inference Single 2D-histogram | Desktop Computer (12th Gen Intel® Core™ i5-12600 @ 3.30 GHz with 6 cores) | 67 ms ± 16 ms |
Inference Single 2D-histogram | GPU (NVIDIA RTX Titan) | 120 ms ± 40 ms |
Inference 10,000 2D-histograms | GPU (NVIDIA RTX Titan) | <11 s (<1.1 ms each) |
Model prediction and complete analysis calculating \({\bar{V}}_{{{\rm{D}}}}({{\bf{G}}},{{{\bf{H}}}}_{1},\ldots ,{{{\bf{H}}}}_{100})\), \({\bar{V}}_{{{\rm{R}}}}({{{\bf{H}}}}_{1},\ldots ,{{{\bf{H}}}}_{100})\), and uncertainty quantification 100 2D-histograms time series length: 10 million samples | (Workstation dual processor) Intel Xeon E5-2697 v2 @ 2.70 GHz | ~30 s |
