The performance of AI models is typically limited by the memory wall. In this project, we aim to (re)move this wall by designing memory systems tailored to the requirements of AI models. We will explore systems based on HBM, DDR, GDDR devices, and Compute Express Link (CXL) memory expanders. Additionally, we will consider near-data processing architectures that offload part of the CPU/GPU processing to the memory system.
Most of the development will take place on high-end hardware products and industrial prototypes. We strongly encourage applications from candidates interested in joining a research team with over a decade of experience in industrial projects with major hardware companies from the US, China, and Korea.
The funding for these actions/fellowships and contracts comes from the European Union Recovery and Resilience Facility - Next Generation, within the framework of the General Invitation by the public business entity Red.es to participate in the talent attraction and retention programs within Investment 4 of Component 19 of the Recovery, Transformation, and Resilience Plan.
For more information, please check: https://www.bsc.es/join-us/excellence-career-opportunities/ai4s
Key Duties
- Profiling and performance modeling of HPC applications running on high-end servers with advanced memory systems: HBM, DDR4/5, CXL memory expanders.
- Exploration of Processing in Memory for HPC and AI applications.
- Prototyping of hardware accelerators and performance tools.
- Benchmarking and performance analysis of high-end hardware products and industrial prototypes.
Data de tancament: Dilluns, 30 Setembre, 2024
Més posts de Recerca
.jpeg)
Cap comentari:
Publica un comentari a l'entrada