IBM has announced CPU, server for enterprise deep learning.

The server reminds of NVIDIA DGX, only with IBM’s own processor and “state-of-the-art” I/O bandwidth.

      • 2 or 4 NVIDIA Tesla V100
      • 2x IBM POWER9™ Single Chip Module (SCM) processors
        • 16-core, 2.6 GHz (3.09 GHz turbo) or
        • 20-core, 2.0 GHz (2.87 GHz turbo)
      • DDR4 256G … 1 TB
      • [latest NVIDIA NVLink latest]
      • 4 PCIe Gen 4 slots, some 3 CAPI enabled


    • The POWER9 processor markedly improves on I/O bandwidth, including communication with GPU card:
    • next generation NVIDIA NVLink
    • PCIe Gen4
    • OpenCAPI

Leave a Reply

Your email address will not be published. Required fields are marked *