Forschungsbericht 2023

Massively Parallel Systems E-EXK5

Leitung: Lal, Sohan
Gründungsdatum: 1.9.2021

Institut auf TORE
Institutswebsite

The institute of Massively Parallel Systems (MPS) investigates and teaches in the field of computer architecture, focussing on massively parallel systems.
These days massively parallel systems are present everywhere – in our smartphones, cars, supercomputers. They help us to do many things which were not possible before. For example, the recent stupendous success of machine learning, especially deep learning, is mainly due to the exponential increase in computational power. As such, machine learning is not new. Machine learning methods are around since the 1950s. What has predominately changed now is the computing power, with many-core processors such as GPUs as the main drivers. If these massively parallel processors are not utilized properly, they are very expensive in terms of power and energy consumption, which is not good as we aspire to reduce our carbon footprint. At MPS, we are working to make computing devices more performance and energy-efficient from the architecture perspective, and improve their programmability from a programmer’s perspective.

Publikationen

• An Efficient Lightweight Framework for Porting Vision Algorithms on Embedded SoCs - Conference Paper
  Ashish, Apurv; Juurlink, Ben H. H.; Lal, Sohan
  6th IFIP TC 10 International Embedded Systems Symposium (IESS 2019)
  Publisher DOI
  
  
• Beyond compression ratio : a throughput analysis of memory compression techniques for GPUs - Conference Paper
  Renz, Manuel; Lal, Sohan
  41st IEEE International Conference on Computer Design (ICCD 2023)
  Open Access