Reducing Job Completion Time in AI/ML Clusters with Broadcom DNX