LFU:online

Students will be able to evaluate for which problems GPUs are appropriate. They are able to write programs in CUDA and confidently use libraries for GPU computing. They should understand the performance characteristics of accelerators and understand how to optimize code in this context.

The topic of the lecture is the use of accelerators (in particular, graphic processing units or GPUs) in scientific computing and related fields such as data science and machine learning. In particular, we consider

• How GPUs work and their performance characteristics.
• For which problems such architectures are appropriate.
• Programming GPUs in CUDA.
• Numerical libraries for GPUs (e.g. cuBLAS)
• Advanced topics such as shared memory, multiple GPUs, etc.
• Best practice in developing GPU codes.
• Use of GPUs in Julia and Python (if time permits)