AI training on LUMI: the largest supercomputer in the Nordics
"Half a year ago it would have been the largest supercomputer in Europe, but I guess I was to slow with this talk." - Gregor Decristoforo
This talk provides a brief introduction to high-performance computing (HPC) with a specific focus on training AI models on LUMI, the largest supercomputer in the Nordics. The talk will cover LUMI's architecture, focusing on its GPU nodes equipped with four AMD MI250X GPUs each which are primarily used for AI tasks. Gregor will include some information on installing and setting up machine learning libraries like PyTorch on LUMI, as well as strategies for scaling AI projects across multiple GPUs and nodes. The talk will also discuss the differences between using LUMI's AMD GPUs with ROCm and the more widely used NVIDIA GPUs with CUDA.
This talk will be streamed on Discord. Please make sure you have an account or are logged in beforehand. For the best experience, we recommend using the Discord Desktop application. Link to our server
