MSA-Seminar: Introduction to GPU Computing with CUDA C++

Modular Supercomputing Architectures Seminar

Referent: Jiri Kraus, Senior DevTech Compute, NVIDIA

This seminar will explain why throughput-optimized GPUs can process massively parallel tasks faster than latency-optimized architectures like CPUs. To utilize GPUs it is required to expose sufficient parallelism. How this can be done is explained by the example of CUDA C++, also covering the basics of data movement and most important GPU Kernel performance characteristics.

Folien: Introduction to GPU computing with CUDA C++