Application-specific requirements for Maximo Visual Inspection

Review the Maximo Visual Inspection recommended hardware configuration and I/O performance.

Recommended hardware configuration and I/O performance

Worker nodes
Maximo Visual Inspection requires the following configuration for each worker node:
  • At least one 64-bit x86-compatible processor.
  • 60GB of memory.
GPUs
Maximo Visual Inspection needs GPUs to perform deep learning model training and requires the following GPU configuration:
  • NVIDIA Ampere, Turing, Pascal, and Volta devices, such as the NVIDIA A10, A100, T4, P40, P100, and V100. Architectures from other vendors are not supported.
  • At least 16 GB of GPU memory is required during model training.
Storage
Maximo Visual Inspection uses a persistent volume claim (PVC) to manage storage and requires the following storage configuration and I/O performance:
  • A PVC storage class that supports ReadWriteMany access mode.
    • NFS based, file-storage based, or IBM Cloud File Storage implementations that offer ReadWriteMany access modes are supported.
    • Storage provider implementations that offer ReadWriteOnce access mode, such as most block storage implementations, are NOT supported.
  • A PVC storage size of at least 40GB for data sets and models.
    • Increased numbers for data sets and models lead to increased storage requirements.
    • Most storage providers allow storage allocations to be increased.

    • If you cannot increase your allocation, use the following rule to determine a rough estimate of your storage utilization:

      (100 KB per image) x (number of images) x (number of data sets) +

      1GB x (number of models) +

      10GB

      Note: Images and video clips vary greatly in resolution and compression ratios. This rule might not be appropriate for your workload.
  • At least 10,000 input/output operations per second (IOPS) of random read/write performance for data set uploads and downloads.
    • Less storage bandwidth means that certain actions are visibly slower.
    • As more storage IOPS are allocated, application performance scales up to approximately 50,000 random read/write IOPS.
    • Configuration of IOPS and storage quotas varies from provider to provider. See how to manage IBM Cloud File Storage IOPS or refer to your storage provider's documentation.