With the emergence of the autonomous car, the increased processing requirements of the ADAS/AV functionality are driving the need for more powerful processing, including GPUs. At the same time, there is a trend towards more high resolution displays for cluster, infotainment, and HUDs. Some applications, such as dashboard and ADAS, must run securely with guaranteed performance. Others, such as infotainment, must be secure for DRM.
While these disparate applications were historically handled by separate chips, by a single chip running software or paravirtualization or by a single chip with dedicated GPU IPs, Tier1s/OEMs are moving towards more powerful single-chip solutions to reduce costs. Full GPU hardware virtualization goes beyond this to enable total isolation between the various applications for increased security, as well as maximum utilisation of the GPU hardware.
We will share the use cases for GPU hardware virtualization, discuss how full hardware virtualization compares to single chip software/paravirtualization, and outline an architecture in which there is no requirement for hypervisor intervention for task submission. We’ll also discuss additional methods, such as task prioritisation, Quality of Service (QoS) mechanisms, Denial of Service (DoS) recognition/recovery, and OS separation, which can enable further levels of reliability.