CUDA Administration Training in Bangkok, Thailand

This course provides system administrators with the knowledge and skills needed to install, configure, manage, and monitor NVIDIA CUDA-enabled GPU systems. It focuses on preparing environments for CUDA development and execution, managing GPU resources, and optimizing system performance for high-compute applications such as deep learning, HPC, and data analytics.

INTRODUCTION TO CUDA PLATFORM AND SYSTEM SETUP

Overview of CUDA and GPU Architecture

  What is CUDA and how it fits into GPU computing

  GPU vs CPU for parallel processing

  Understanding the NVIDIA ecosystem: drivers, CUDA toolkit, cuDNN

Installing and Verifying CUDA on Linux

  Installing NVIDIA drivers and CUDA toolkit

  Verifying installation with nvidia-smi and sample builds

  Setting up environment variables and PATH configuration

GPU Monitoring and Device Management

  Using nvidia-smi to monitor GPU stats

  Configuring power states, fan settings, and persistence mode

  Checking compatibility and runtime issues

ADVANCED CONFIGURATION AND MULTI-GPU MANAGEMENT

Multi-GPU and Resource Management

  Identifying and isolating GPUs for specific processes

  Managing multi-GPU workloads

  Using MIG (Multi-Instance GPU) on supported hardware

CUDA Toolkit Management and Versioning

  Supporting multiple CUDA versions

  Managing library dependencies and symbolic links

  Using modules and environment managers (e.g., Conda)

CONTAINERIZATION, MONITORING, AND TROUBLESHOOTING

CUDA in Containers and Virtualized Environments

  Using Docker with NVIDIA Container Toolkit

  Setting up GPU access in containers

  Managing container images with CUDA dependencies

Troubleshooting and Performance Optimization

  Diagnosing driver and kernel issues

  Handling runtime errors, segmentation faults, and memory leaks

  Profiling and optimizing resource usage

Final Hands-On and Wrap-Up

  Deploy a containerized CUDA environment

  Configure and monitor a multi-GPU workload

  Review of common issues and admin best practices