Fundamentals of Accelerated Computing with CUDA C/C++ Training Course

GPU Computing and CUDA Architecture

• CPU vs GPU architectural differences
• NVIDIA GPU streaming multiprocessor model
• CUDA programming model overview
• Heterogeneous computing and the host-device paradigm

Setting Up the CUDA Development Environment

• Installing the CUDA Toolkit 13.x
• NVCC compiler and build workflow
• Verifying the environment with device queries
• IDE integration and development tools

Writing and Launching CUDA Kernels

• Kernel function syntax and qualifiers
• Launch configuration and execution
• Vector addition and basic data-parallel patterns
• CUDA error checking macros

CUDA Thread Hierarchy and Execution Model

• Grid, block, and thread organization
• Thread indexing and global ID calculation
• Warp execution and SIMT model
• Occupancy and resource utilization

GPU Memory Architecture and Management

• Memory types: global, shared, constant, registers
• Allocating and freeing device memory
• Host-to-device and device-to-host transfers
• Shared memory for intra-block collaboration

Unified Memory and Data Migration

• Unified memory model and managed allocations
• Page migration and on-demand paging
• Asynchronous prefetching with cudaMemPrefetchAsync
• Memory advice hints for access patterns

System-Wide Profiling with Nsight Systems

• Nsight Systems timeline analysis
• Identifying CPU-GPU synchronization points
• Visualizing kernel execution and memory transfers
• Interpreting system-level performance data

Kernel Optimization with Nsight Compute

• Nsight Compute interactive kernel profiling
• Memory throughput and bandwidth analysis
• Compute utilization and warp state statistics
• Guided analysis and optimization rules

Concurrent Streams and Asynchronous Operations

• CUDA streams and the default stream
• Overlapping kernel execution with data transfers
• Stream synchronization and CUDA events
• Multi-stream pipeline design patterns

Error Handling and Debugging Tools

• CUDA API error codes and recovery strategies
• Compute-sanitizer for memory access checking
• cuda-gdb for kernel debugging
• Assertions and synchronous error detection

Profile-Driven Optimization Workflow

• Iterative profiling methodology
• Bottleneck identification and prioritization
• Performance regression testing
• Documenting optimization decisions

End-to-End Accelerated Application Project

• Designing a complete GPU-accelerated solution
• Integrating profiling throughout development
• Performance benchmarking and reporting
• Deployment considerations for production