Performance Analysis of Intel Gen9.5 Integrated GPU Architecture

06/19/2018

CPUs have since the last years come equipped with integrated GPUs. Such coupling has the potential to offer performance improvements if said GPUs are used as accelerators. However, integrated GPUs can hardly reach the performance levels obtained with discrete GPUs due to having lower core counts. Nevertheless, their proximity to the CPU allows for sharing of data with less overhead. Furthermore, the advantages of sharing the same memory hierarchy and the lower power consumption at the cost of performance, allows for high energy efficiency gains. This thesis aims to characterize the Intel GPU micro-architecture and to benchmark its performance upper-bounds to study the limits of its energy efficiency. Through exposing the Intel GPU's performance counters, greater insights can be gleamed about how to extract peak performance from the architecture and how to best exploit the shared memory hierarchy.

Authors in the community:

• 

  H

  Hélder Francisco Pereira Duarte

  ist426276

Supervisors of this institution:

• 

  A

  Aleksandar Ilic

  ist166430

electrical-engineering-electronic-engineering-information-engineering - Electrical engineering, electronic engineering, information engineering

eng - English

Instituto Superior Técnico