Authors: Davide Balzarotti (Eurecom), Roberto Di Pietro (Bell Labs), and Antonio Villani (University of Roma Tre)



In this paper, we assess the impact of GPU-assisted malware on memory forensics. In particular, we first introduce four different techniques that malware can adapt to hide its presence. We then present a case study on a very popular family of Intel GPUs, and we analyze in which cases the forensic analysis can be performed using only the host’s memory and in which cases it requires access to the GPU’s memory. Our analysis shows that, by offloading some computation to the GPUs, it is possible to successfully hide some malicious behavior. Furthermore, we provide suggestions and insights about which artifacts could be used to detect the presence of GPU-assisted malware.