Online Access Free NCP-AII Practice Test

You are implementing a distributed deep learning training setup using multiple servers connected via NVLink switches. You want to ensure optimal utilization of the NVLink interconnect. Which of the following strategies would be MOST effective in achieving this goal?

• A.Implement a data compression algorithm that can be processed by the CPU before sending data over NVLink.
• B.Disable peer-to-peer GPU memory access within each server to avoid contention.
• C.Increase the batch size to reduce the frequency of inter-GPU communication.
• D.Configure NCCL to use GPUDirect RDMA for inter-GPU communication across servers.
• E.Use a standard TCP/IP socket connection for inter-GPU communication across servers.

A BlueField-3 DPUis configured to run both control plane and data plane functions. After a recent software update, you notice that the data plane performance has significantly degraded, but the control plane remains responsive. What is the MOST likely cause, assuming the update didn't introduce any code bugs, and what is the BEST approach to diagnose this issue?

• A.Driver incompatibility; Downgrade the Mellanox OFED drivers to the previous version.
• B.Resource contention; Use 'perf or 'bpftrace' to profile the data plane processes and identify resource bottlenecks (CPU, memory, cache).
• C.Network misconfiguration; Verify the MTU and QOS settings on the network interfaces.
• D.Firmware corruption; Re-flash the BlueField DPIJ with the latest firmware image.
• E.Power throttling; Check the DPU's power consumption and thermal status via the BMC.

You have a GPU-intensive application that requires the latest features of CUDA 12. However, your host system's NVIDIA driver is only compatible with CUDA 11.8. What steps can you take to enable your application to use CUDA 12 within a Docker container, without upgrading the host driver?

• A.Use a Docker image based on 'nvidia/cuda:12.0-base-ubuntu20.04'. The NVIDIA Container Toolkit will automatically handle the driver compatibility between the host and the container.
• B.Install CUDA 12 inside the Docker container and set the 'CUDA DRIVER VERSION' environment variable to match the host driver version.
• C.Upgrade the NVIDIA driver on the host system to the latest version compatible with CUDA 12.
• D.Downgrade the application to use CUDA 11.8 to match the host's driver version.
• E.Mount the CUDA 12 libraries from a separate Docker volume into the container and configure the accordingly.

Your deep learning training job that utilizes NCCL (NVIDIA Collective Communications Library) for multi-GPU communication is failing with "NCCL internal error, unhandled system error" after a recent CUDA update. The error occurs during the 'all reduce' operation.
What is the most likely root cause and how would you address it?

• A.GPU Direct RDMA is not properly configured. Check 'dmesg' for errors and ensure RDMA is enabled.
• B.Faulty network cables used for inter-node communication (if the training job spans multiple servers). Replace the network cables with certified high-speed cables.
• C.Firewall rules blocking inter-GPU communication. Configure the firewall to allow communication on the NCCL-defined ports (typically 8000-8010).
• D.Insufficient shared memory allocated to the CUDA context. Increase the shared memory limit using 'cudaDeviceSetLimit(cudaLimitSharedMemory, new_limity.
• E.Incompatible NCCL version with the new CUDA version. Update NCCL to a version compatible with the installed CUDA version.

You have installed NVIDIA drivers using the .run' installer on a system running Ubuntu. However, after each kernel update, the NVIDIA drivers stop working. What is the most effective way to address this issue permanently?

• A.Use DKMS (Dynamic Kernel Module Support) to automatically rebuild the NVIDIA kernel modules after each kernel update.
• B.Disable automatic kernel updates.
• C.Re-run the '.run' installer after each kernel update.
• D.Create a script that symlinks the old kernel modules to the new kernel directory.
• E.Manually compile the NVIDIA kernel modules after each kernel update.