Open Source
Tools & Resources
Open-source frameworks, tutorials, talks, training courses, and datasets from our research.
The Showroom
Our Tools
Mess Framework
Memory System Stress Framework
A unified framework for memory system benchmarking, simulation, and application profiling. The Mess Framework encompasses multiple tools that work together to provide holistic memory system characterization.
Components
Mess Benchmark
v2.0Memory System Stress Framework
Unified benchmarking, simulation and profiling across x86, ARM, RISC-V, and GPU architectures. Now faster and more portable in v2.0
Mess-Paraver
ToolMess Integration with Paraver
Utility to use Mess together with the BSC's Paraver for advanced memory system analysis and visualization.
Mess Simulator
SimulatorMemory System Simulator
Advanced simulation framework for modeling memory systems in HPC and AI workloads.
PROFET
StablePerformance & Energy Prediction
Analytical models predicting application performance and energy changes on future memory systems.
FAiNDER
AI Hardware Explorer
Open-source platform to navigate AI model requirements and optimize hardware choices.
Prediction of DRAM Errors in the field
DRAM Error Prediction and Mitigation
Advanced tools for predicting and mitigating uncorrected errors in DRAM systems, focusing on next-generation HPC and AI requirements.
Components
UEPREDICT
DRAM Error Prediction
Method for predicting DRAM Uncorrected Errors and evaluating its model's performance. Our work focuses on advanced memory technologies and memory requirements of next generation HPC and AI systems.
UERL
Adaptive UE Mitigation with RL
Source code for an adaptive mitigation method designed to address uncorrected errors (UEs) in DRAM. Leveraging Reinforcement Learning (RL) techniques, the method dynamically adapts to the probability and potential cost of encountering such errors, offering a proactive approach to mitigate their impact. This is the supplemental code for the HPDC24 paper 'Reinforcement Learning-based Adaptive Mitigation of Uncorrected DRAM Errors in the Field'.
Learning
Trainings & Tutorials
Courses and tutorials on memory system performance analysis and our tools.
It's the Memory, Stupid! Tutorial
Half-day tutorial at MICRO 2026. BSC, Intel and INESC-ID deep dive into memory system profiling and performance analysis, with the creators of the Mess Framework, Roofline models and TopDown analysis.
It's the Memory, Stupid!
Join BSC, Intel and INESC-ID to deep dive into memory system profiling and performance analysis. Learn from the creators of tools like the Mess Framework, Roofline models, TopDown analysis, and heterogeneous memory systems.
Mess Framework Tutorial
Tutorial on using Mess Framework for memory system benchmarking, simulation, and application profiling.
The Library
Materials
Slides, talks, tutorials, and datasets from our research.
It's the Memory, Stupid! — Full Course Materials
Every session of the February 2026 course in one place: slides, reference papers, repositories, sample traces, and the interactive visualizers built for the course.
Memory Course — Introduction
Opening lecture: motivation, structure, and why the memory wall is still with us after 30 years.
Roofline(s) and the Cache-Aware Roofline Model
Roofline lecture given by INESC-ID, covering the classical Roofline and the Cache-Aware Roofline Model (CARM) with per-cache-level ceilings.
CARM Tool Demo
Walkthrough of the CARM roofline tool: automatic benchmarking of a machine and plotting an application against its ceilings.
CARM Tool Demo — Sample Data
The measurement results used live during the CARM tool demo, ready to load into the tool yourself.
Memory System Performance is Messy
Mess Framework lecture: unified memory benchmarking, simulation, and application profiling across x86, ARM, RISC-V, and GPUs.
PROFET: No-stress Performance Prediction
PROFET lecture on predicting application performance and energy across current and future memory systems without simulating the CPU.
TopDown Microarchitecture Analysis
Lecture by Intel on decomposing CPU pipeline slots into front-end, back-end, speculation, and retirement bottlenecks.
TopDown Reference Guide
BSC internal reference manual (February 2026) with a full breakdown of the TopDown memory metrics and their references. Not official Intel documentation.
Heterogeneous Memory Systems
Lecture by the BSC Accelerators and Communications for HPC group on programming and profiling heterogeneous memory systems.
Course Visualizers — Mess, PROFET & TopDown
Three standalone desktop applications built for the course, for macOS, Windows and Linux on both x86 and ARM. Explore Mess bandwidth-latency curves, PROFET predictions, and TopDown CPI breakdowns interactively.
Mess — Sample Traces & Curves
Mess traces from MareNostrum 5 GPP and HighMem nodes, plus a set of sample bandwidth-latency curves to load into the Mess visualizer.
PROFET — Sample Traces & Configs
PROFET traces from MareNostrum 5 GPP and HighMem nodes together with the memory system configuration files used in the course.
TopDown — Sample Traces
TopDown counter traces collected on a MareNostrum 5 GPP node, for use with the TopDown visualizer.
CARM Roofline Tool
Open-source tool from INESC-ID for automatic roofline benchmarking and application analysis, used in the Roofline session.
Getting Started with Mess 2.0
A brief introduction to Mess 2.0, covering the key changes from version 1.0 and the new features it adds.
Mess-Results
Collection of measured Mess bandwidth-latency curves across a wide range of production memory systems.
PROFET Brief Tutorial
A brief introduction to PROFET presented at MICRO 2025.
A Mess of Memory System Benchmarking, Simulation and Application Profiling
An introduction to Mess Framework for memory system benchmarking, simulation, and application profiling.
Mess Curves
Access the original Mess-benchmark repository containing pre-computed benchmark results and system curves from previous studies.
FAINDER - Symposium Poster
Symposium poster presentation for FAINDER.
No materials match this combination yet.