An HPC cluster on AWS runs a tightly coupled workload that produces many shared files on Amazon EFS. Performance was acceptable with 100 EC2 instances but dropped significantly at 1,000 instances. Which set of design choices will maximize cluster performance? (Choose three.)

For tightly coupled HPC workloads, minimizing network latency is crucial. Launching the entire cluster in a single Availability Zone reduces inter-AZ network latency, which can be significant. EFA provides OS bypass and low-latency communication, essential for scaling tightly coupled applications. Amazon EFS, while scalable, has higher latency and lower throughput limits compared to specialized HPC file systems. Amazon FSx for Lustre is designed for high-performance computing, offering significantly higher throughput and lower latency, making it ideal for shared file systems in large HPC clusters. Distributing across multiple AZs would increase latency. Attaching multiple ENIs is not a primary performance driver for tightly coupled HPC. Replacing EFS with EBS RAID is not suitable for a shared file system across 1,000 instances.