Europe's AMD-Powered LUMI Supercomputer Continues With Code Porting, Open-Source Tuning

The LUMI supercomputer in Finland is still being assembled with its 2,560 nodes consisting of a 64-core AMD Trento CPU and four AMD Instinct MI250X GPU accelerators per node. This 375+ PFLOPs was supposed to come online by the end of 2021 but was challenged by the supply chain crisis and is now aiming for general availability by the middle of the year. While the hardware is still coming together, their HPC engineers have been hard at work optimizing the open-source Linux software stack.

Last February there was a great presentation by Georgios Markomanolis who is the lead HPC scientist at CSC. He talked about the state of the Radeon Open eCosystem last year and all their porting efforts involved in getting traditionally NVIDIA centered HPC workloads working in their AMD test hardware at the time -- originally MI100 while waiting on the MI250X to be used in LUMI.
Georgios Markomanolis presented last weekend at FOSDEM 2022 to share more about their latest open-source software efforts over the past year.

The HPC engineers continue becoming more experienced with AMD's ROCm open-source software stack for GPU computing, continue hammering on HIP for porting more CUDA codebases into AMD GPU supportive software, and then working to fine-tune the codes for maximum performance once running on the AMD hardware.

One of the new software components they have added to their toolbag over the past year is AMD's open-sourced GPUFort for helping to move OpenACC and CUDA Fortran code into AMD's architecture.

CSC LUMI

It still is a rather complex jungle moving complex codebases into the AMD ROCm space for optimal GPU performance.

CSC LUMI

Those wanting to learn more about the open-source porting and tuning work going on at CSC to prepare for LUMI coming online can see Georgios Markomanolis' great virtual presentation at FOSDEM.org with the slides and video recordings. Those wanting to learn more about LUMI can do so at lumi-supercomputer.eu.