FTGPGPU- Hardware Fault Tolerance Analysis for General Purpose Graphics Processing Units (GPGPU) Applications (TUBITAK 3501 – Career Development Program (CAREER))
FTGPGPU aims to analyze the regional soft error vulnerability of GPGPU applications and help the development of the efficient fault tolerance techniques. FTGPGPU examines the application code level error vulnerability of the GPGPU applications, and contributes with a regional analysis of the code segments in those applications. FTGPGPU is seeking answers to the following research questions: Do the different code segments in GPU applications have different error vulnerabilities? Do the faults in the code segments affect the other code segments? How is the error vulnerability affected if the different GPU memory structures are used for data allocation? Is it possible to generalize the differences between regional error vulnerabilities for similar programs? What are the possible suggestions for the software developers to develop more reliable software? How can a fault tolerance technique be developed by using the differences between regional error vulnerabilities?