-
Notifications
You must be signed in to change notification settings - Fork 40
MaTEx Performance (Old)
Charles Siegel edited this page Mar 28, 2017
·
1 revision
Intel 20-core Haswell, InfiniBand FDR, and OpenMPI 1.8.1
MLP consists of two layers of 800 features each.
CNN consists of two convolution layers of 16 and 32 features, with 5x5 windows, stride of 2, and after each a 2x2 max pooling layer, followed by a fully connected layer of 500 features.
Getting Started on MaTEx-TensorFlow
- Required Software
- Installing MaTEx-TensorFlow on CPU Clusters
- Installing MaTEx-TensorFlow on GPU Clusters
- MaTEx-TensorFlow on Older glibc(v<2.19)
- DataSet Reader
- Testing Scripts
- Performance
- Running on PNNL Systems
- Running on NERSC Systems
- Restarting the MaTEx TensorFlow environment