PROFET

The PROFET tool (PROFiling-based EsTimation of performance and energy) is designed to profile memory system performance, quantify the application pressure to the memory system and estimate application performance on hardware platforms with novel memory systems.

Institution:

Research Group:

BSC Group: Computer Sciences Memory Systems for HPC and AI

Researcher/s:

Mariana Carmin, Victor Xirau Guardans, Pouya Esmaili Dokht, Isaac Boixaderas, Valéria Soldera Girelli, Milan Radulovic, Petar Radojkovic, Eduard Ayguadé Parra

Description:

PROFET analyzes application behavior on current systems and predicts how it would perform on future memory technologies, such as high-bandwidth memory or new architectures. It uses collected performance data to estimate speed, power, and energy under different configurations. This helps designers and operators compare options before building new systems, supporting more efficient and informed decisions. It works with standard data formats, making it compatible with many applications, and contributes to energy-aware planning in modern computing environments.

Value Proposition:

Predicts performance-and-energy on future memory systems from one profile

Aplication areas:

Memory-system co-design, HPC node roadmap planning, CXL fabric sizing, DRAM vs. HBM trade-offs, AI inference TCO studies, academic architecture research

Target market:

Semiconductor & system vendors evaluating HBM/NVRAM/CXL options; Super-computing centres planning next-gen nodes; Cloud providers optimising memory tiers for AI inference; Academic labs teaching performance-energy modelling

Protection:

BSD 3-Clause License

More information

if you want to know more about this project do not hesitate to contact us