Hardware and software requirements

The following hardware and software requirements must be met before installing IBM Spectrum Conductor Deep Learning Impact 1.1.0.

Hardware requirements

The following hardware requirements are applicable to IBM Spectrum Conductor Deep Learning Impact 1.1.0.
Table 1. Hardware requirements for IBM Spectrum Conductor Deep Learning Impact 1.1.0
Requirement Management host Compute host
RAM 64 GB 32 GB
Disk space to install 11 GB 11 GB
Hardware x86 hardware that leverages GPUs or IBM POWER8 or POWER9 servers with NVLink and NVIDIA GPUs x86 hardware that leverages GPUs or IBM POWER8 or POWER9 servers with NVLink and NVIDIA GPUs

Software requirements

The following software requirements are applicable to IBM Spectrum Conductor Deep Learning Impact 1.1.0.

Note: IBM Spectrum Conductor Deep Learning Impact on Linux for POWER LE (64-bit) with IBM PowerAI 1.5.2 is only available with IBM PowerAI Enterprise 1.1. For more information, see Release notes for IBM Spectrum Conductor Deep Learning Impact 1.1.0.
Table 2. Software requirements for IBM Spectrum Conductor Deep Learning Impact 1.1.0 for Linux for POWER LE (64-bit)
Hardware Operating system GPU software IBM PowerAI IBM Spectrum Conductor with Spark Shared file system
POWER8 with PowerAI 1.5.0 Red Hat Enterprise Linux (RHEL) 7.4 (ppc64le) NVIDIA CUDA 9.0 and CUDA Deep Neural Network (cuDNN) 7.0 library 1.5.0 2.2.1 IBM Spectrum Scale 4.2.3, 4.2.2, 4.2.1, 4.1.1

Network file system (NFS) 2, 3, 4
POWER9 with PowerAI 1.5.1 (Only available with iFix 489874) Red Hat Enterprise Linux (RHEL) 7.5 (ppc64le) NVIDIA CUDA 9.2.64 and CUDA Deep Neural Network (cuDNN) 7.1.3 library 1.5.1 2.2.1 (with the latest fixes from IBM Fix Central) IBM Spectrum Scale 4.2.3, 4.2.2, 4.2.1, 4.1.1

Network file system (NFS) 2, 3, 4
Table 3. Software requirements for IBM Spectrum Conductor Deep Learning Impact 1.1.0 for Linux (64-bit)
Operating system GPU software IBM Spectrum Conductor with Spark Shared file system
Red Hat Enterprise Linux (RHEL) 7.3 (64-bit) NVIDIA CUDA 8.0, CUDA Deep Neural Network (cuDNN) 6.0 library and the NVIDIA Collective Communications Library (NCCL) 2.2.1 IBM Spectrum Scale 4.2.3, 4.2.2, 4.2.1, 4.1.1

Network file system (NFS) 2, 3, 4
Ubuntu 16.04 NVIDIA CUDA 8.0 and CUDA Deep Neural Network (cuDNN) 6.0 library and the NVIDIA Collective Communications Library (NCCL) 2.2.1 IBM Spectrum Scale 4.2.3, 4.2.2, 4.2.1, 4.1.1

Network file system (NFS) 2, 3, 4