Complete AMD GPU Programming for Beginners 12+ Hours

Trace AMD GPU architectures from GCN to RDNA and Vega, including seven-nanometer tech and GF versioning, and learn kernel execution on compute units with wavefronts and memory coalescing.

Learn hip programming basics, including compiling on Nvidia systems, migrating Cuda code, and using hip kernels with memory management, templates, and printf debugging; plus matrix-vector multiplication assignment context.

Learn hip kernel programming, using global, device, and host qualifiers to launch workloads from the host and manage thread, block, and grid indices.

Explore hip memory models and the gpu memory hierarchy, from global memory to registers and shared memory, with locality, caching policies, and warp behavior to optimize amd gpu code.

Explore hip synchronization and atomic operations, race conditions, and memory fences, while learning how to compile for Nvidia with hip platform and nvcc to support reductions.

Learn why mastering tools matters for performance and how rokom info and the system management interface bridge high-level apps with hardware, then profile and debug with profiler, tracer, and debugger.

Explore HIP performance tuning by examining debugger and profiler workflows, understanding AMD GPU ISA, and mastering occupancy, memory access, and divergence concepts for efficient kernels.

Learn how hipify port cuda code to hip using heapify pearl and heapify clang, with demos like vector add, k-means, and CNN, and understand portability across AMD and NVIDIA platforms.

Learn to harness gpu parallelism with local sums, reductions, and memory coalescing through tiling and barriers. Use rock, rocblas, and hip libraries to achieve high-performance matrix operations and optimized convolutions.

Analyze a simple matrix multiplication kernel, show how tiling with shared memory reduces memory traffic, and compare naive versus tiled approaches emphasizing standard library benefits.

Master multi-GPU AMD programming by querying device counts, setting devices, and launching kernels in parallel streams; implement vector add across GPUs and explore thread, MPI, and peer-to-peer memory patterns.