Lab Computing Servers:

• image: 6-core CPU (i7-6800K) with 128GB memory + 4 GPUs (4 x GTX 1080, 8GB memory each)

• text: 6-core CPU (i7-5820K) with 64GB memory + 4 GPUs (4 x TITAN X, 12GB memory each)

• video: 6-core CPU (Xeon E5-1650) with 64GB memory + 3 GPUs (1 x Tesla K40c* + 2 x TITAN X, 12GB memory each)

• language: 6-core CPU (Xeon E5-2620) with 64GB memory + 2 GPUs (2 x TITAN X, 12GB memory each)

• music: 6-core CPU (Xeon E5-1660) with 64GB memory + 2 GPUs (2 x TITAN X, 12GB each)

• voice: 6-core CPU (Xeon E5-2620) with 64GB memory + 2 GPUs (GTX 1080 + GTX 1070, 8GB memory each)

• audio: 4-core CPU (i7-950) with 16GB memory + 2 GPUs (2 x GTX1070, 8GB memory each)

• speech1: 4-core CPU (i7-6700) with 16G memory

• speech2: 4-core CPU (i7-6700) with 16G memory + 1 GPU (GTX 1070 8GB)

• speech3: 4-core CPU (i7-6700) with 16G memory + 2 GPU (GTX 1070 8GB + GTX 780 Ti 3GB memory each)

• speech4: 4-core CPU (i7-6700) with 16G memory + 2 GPUs (GTX 1070 8GB memory each)

• speech5: 4-core CPU (i7-6700) with 16G memory + 2 GPUs (GTX 1070 8GB memory each)

• speech6: 4-core CPU (i7-6700) with 16G memory + 2 GPUs (GTX 1070 8GB memory each)

• speech7: 4-core CPU (i7-6700) with 16G memory + 2 GPUs (2 x GTX 1070 8GB memory each)

Check the GPU assignments before you submit your jobs to GPUs. Check here for how to check the workloads. Never overload any machines. Storage Space: * The shared network disk (with backup) is located at /cs/research/asr. * Each of the above machines has a local disk space (several TB's each) at /local (without any backup). In addition, more local disk space can be accessed from Linux directory /tmp. - text: /local/scratch (6 TB) - video: /local/scratch (6 TB); /local/scratch1 (6 TB) - music: /local/scratch (2 TB); /local/scratch1 (6 TB) - language: /local/scratch (1 TB) - voice: /local/scratch (1 TB); /local/scratch1 (2 TB) - audio: /local/scratch (1 TB) - speech1 to speech7: /local/scratch (500 GB each). Software: * TensorFlow: see here (provided by Mingbin) for how to configure TensureFlow. Alternative, you may configure TensorFlow as here (provided by Chao). * TensorFlow Tutorials: - Tensorflow and deep learning - without a PhD by Martin Görner - TensorFlow Tutorial (Sherry Moore, Google Brain) - RNNs in Tensorflow, a Practical Guide and Undocumented Features – WildML - Tfrecords Guide * Acknowledgement**: the Tesla graphical card was donated by NVIDIA Inc.