Unit 6 Quiz

InfiniBand is a networking technology known for its:

• High throughput and low latency.

• Low bandwidth and high latency.

• Limited scalability.

• High processing overhead.

Parallel and distributed file systems offer advantages in:

• Data security.

• Scaling capacity and performance across multiple storage units.

• Simplified data access compared to other file systems.

• Reduced complexity in data management.

Storage performance for AI applications is often measured in terms of:

• IOPS (Input/Output Operations Per Second) and bandwidth.

• Network latency.

• CPU clock speed.

• GPU memory capacity.

The primary function of compute nodes in an AI cluster is to:

• Run computations on CPUs and GPUs.

• Store data used for training.

• Monitor and manage the system.

• Connect compute, storage, and management services.

Object storage systems are best suited for:

• Caching frequently accessed data.

• High-performance computing tasks.

• Storing data with a complex directory structure.

• Creating large-scale data repositories (petabytes or exabytes).

The NVIDIA DGX POD reference architecture is primarily designed for AI clusters with:

• Unlimited number of nodes.

• Exclusive use of Ethernet networking.

• A single type of GPU server.

• Up to eight nodes.

NVIDIA reference architectures offer the following benefits:

• Increased complexity in system design.

• Proven designs for reliable and robust systems.

• Reduced system performance.

• Limited flexibility for customization.

RDMA (Remote Direct Memory Access) offers the following benefits:

• Increased CPU overhead during data transfer.

• Complex data access protocols.

• Efficient data transfer with minimal CPU involvement.

• Limited operating system support.

Which type of shared file system is most common and based on a standard protocol?

• Object storage systems.

• Parallel file systems.

• Network File System (NFS).

• Local file systems.

GPU Direct RDMA technology enables:

• Slower communication between GPUs.

• Reduced ROI (Return on Investment) for AI infrastructure.

• Decreased scaling efficiency for multi-GPU systems.

• Direct communication between NVIDIA GPUs in remote systems, bypassing CPUs.