Cloud and Edge provider Gcore has launched an AI Cloud cluster in Newport, Wales.
This is Gcore’s third such deployment, having established an AI Cloud point of presence (PoP) in the Netherlands and Luxembourg.
The cluster has been built with Graphcore Intelligence Processing Units (IPUs), a processor designed for compute parallelism and large memory bandwidth making them useful for machine learning and AI applications.
The Gcore AI cluster is made up of Graphcore Bow-Pods, IPU-Pods, and Bow-vPods. During performance testing, a Bow Pod16 was able to train EfficientNET-B4 in less than 14 hours, while the latest GPU technology took 70.5 hours. There was also a 10x higher performance in processing Dynamic Temporal Graph Networks. The overall performance capability of these AI clusters has not been shared.
Read more: datacenterdynamics.com
