Hardware and software requirements
IBM® PowerAI Enterprise requires the following hardware and software.
Hardware requirements
- IBM POWER8® with NVLink and NVIDIA GPUs
- IBM POWER9™ with NVLink and NVIDIA GPUs
The following tables list the minimum system requirements for running IBM PowerAI Enterprise in a production environment. You might have extra requirements (such as extra CPU and RAM) depending on the Spark instance groups that will run on the hosts, especially for compute hosts that run workloads.
| Requirement | Management hosts | Compute hosts | Notes |
|---|---|---|---|
| RAM | 64 GB | 32 GB | In general, the more memory your hosts have, the better performance is. |
| Disk space to extract install files from the PowerAI Enterprise install package | 16 GB (First management host only) | NA | |
| Disk space to install IBM Spectrum Conductor™ | 12 GB | 12 GB | |
| Disk space to install IBM Spectrum Conductor Deep Learning Impact | 11 GB | 11 GB | |
| Additional disk space (for Spark instance group packages, logs, and so on.) | Can be 30 GB for a large cluster | 1 GB*N slots + sum of service package sizes (including dependencies) | Disk space requirements depend on the number of Spark instance groups and the Spark applications that you run. Long running applications, such as notebooks and streaming applications, can generate huge amounts of data that is stored in Elasticsearch. What your applications log can also increase disk usage. Consider all these factors when estimating disk space requirements for your production cluster. For optimal performance, look at tuning how long to keep application monitoring data based on your needs. |
Software requirements
The following software is required:
| Hardware | Operating system | GPU software |
|---|---|---|
| POWER8 | Red Hat Enterprise Linux (RHEL) 7.5 (ppc64le) |
|
| POWER9 with this security fix: RHSA-2018:1374 - Security Advisory | RHEL 7.5 (ppc64le) |
|
- Supported GPUs: NVIDIA P100 and V100
- Shared file system:
- IBM Spectrum Scale 4.2.3, 4.2.2, 4.2.1, or 4.1.1
- Network file system (NFS) 2, 3, or 4
Deep learning frameworks
By default, all of the frameworks included with PowerAI are installed. At least one supported framework must be installed. However, it is recommended that you install both TensorFlow and Caffe or IBM Caffe. If either framework is missing, the option for the missing framework will not work in the cluster management console. If both Caffe and IBM Caffe are installed, IBM Caffe is used by default.
To determine which frameworks are included with PowerAI Enterprise, see What's included.
Required third party software components
- Anaconda 5.2
- NVIDIA CUDA 9.2.148 and Patch 1
- NVIDIA GPU driver 396.44
- NVIDIA cuDNN v7.2.1 for CUDA 9.2
- NVIDIA NCCL v2.2.13 for CUDA 9.2
Required additional repositories
- 'optional', 'extras', and 'common' from Red Hat Enterprise Linux (RHEL)IBM
- Fedora Project EPEL (Extra Packages for Enterprise Linux)