Slides for computing platforms lecture

Author: jussienko

intro-day/docs/05-computing-platforms.md

@@ -0,0 +1,60 @@
+---
+title:    Using LUMI and Mahti supercomputers 
+subtitle: Portable GPU Programming
+date:     October 2025
+lang:     en
+---
+
+
+# LUMI and Mahti supercomputers {.section}
+
+# Outline
+
+- Introduction to LUMI and Mahti
+- Connecting
+- Using Slurm batch job system
+- Modules and compiling
+
+# Web resources
+
+- LUMI User guide: <https://docs.lumi-supercomputer.eu/>
+- CSC User guide: <https://docs.csc.fi/>
+- General Slurm documentation: <https://slurm.schedmd.com/>
+
+# LUMI supercomputer
+
+<div class=column style=width:60%>
+- LUMI is pan-European supercomputer hosted by CSC
+    - user support is distributed to LUMI consortium countries
+- For Finnish users access is via CSC
+- Over 10 000 AMD GPUs
+- "small" (about the size of Mahti) CPU partition
+- auxiliary partition for data-analytics with lots of memory (32 TB per node)
+</div>
+<div class=column style=width:38%>
+![](img/lumi2.jpg){.center width=90%}
+</div>
+
+# Mahti supercomputer
+
+- Current national flagship system
+- 1404 nodes with AMD CPUs, 128 cores per node
+- 24 nodes with NVIDIA A100 GPUs
+    - subset can be sliced to smaller GPUs for interactive workloads,
+      code development etc.
+- To be replaced by Roihu (AMD CPUs and NVIDIA Grace Hopper superchips)
+
+# Performance analysis with Rocprof and Nsight Systems {.section}
+
+# Rocprof
+
+- AMD performance analysis and tracing tool
+- Visual analysis via web browser
+    - <https://ui.perfetto.dev/>
+
+# Nsight Systems
+
+- NVIDIA performance analysis and tracing tool
+- GUI for visual analysis
+    - Recommended to install on local workstation for smoother respose
+    - <https://developer.nvidia.com/nsight-systems/get-started>