A company is building a network fabric for an AI training cluster that will train large language models. The cluster includes 256 GPUs distributed across 32 servers. Security requirements mandate isolation between different training projects and allowing shared access to a central storage system.
Which approach provides the required security isolation and maintains optimal performance for GPU-to-GPU communication?
What does workload distribution offer in an AI infrastructure with local and external resources?
A network architect designs connectivity for a new Cisco AI POD operating in Intersight Managed Mode. To ensure high availability and aggregate bandwidth for external network traffic, a port channel must be established using multiple 100 GbE uplink ports on the fabric interconnects to connect to the upstream core switch infrastructure.
Which specific policy, applied through the domain profile, must be modified to define and configure this port channel on the fabric interconnects for the desired uplink connectivity?
Which type of Cisco Intersight profile, when deployed to fabric interconnects, includes the configuration settings for ports, VLANs, and VSANs?
When considering a rail-only network design, when would adding spine switches make sense?
Currently there are no comments in this discussion, be the first to comment!