Splitting one physical GPU into isolated slices that several virtual machines or containers can use at the same time.
GPU Virtualization raises the share of an expensive GPU's capacity that is in use, which improves the return on the hardware. Instead of one AI model holding an entire GPU while idle, virtualization divides the VRAM and compute cores among workloads, so multiple isolated Agentic Workflows and inference services can run concurrently on the same physical chip.
See also Agentic Workflows, Data Mobilization, Data Lakehouse and Vector Database.