
138 IDiscovering
9.3 Graphics and Compute Interoperability on
a System Level
There are many situations when the application requires distributing graphics and
compute portions across two or more GPUs. In other words, one GPU will be
dedicated to the computation, while another is dedicated to rendering the scene. In
these situations, communication between the compute and graphics contexts turns
into communication between devices in a system (Figure 9.2).
Motivations for multi-GPU system architectures:
• Increase in system processing power. Offloading intensive computa-
tion from the main-display GPU will always result in overall system interac-
tivity