
420 V Transfers
GPUView trace is shown in Figure 29.4, and shows that upload and render are han-
dled sequentially and the additional GPU hardware queue is unused. This graph
also shows that the memcpys are in fact interspersed with the transfer, causing the
spikes and gaps in the CPU and GPU command queues. Ideally, we would like the
execution timeline of a packet to be solid until completion to show that the GPU is
kept fully busy, as is shown by the render.
29.3.2 CPU Asynchronous Texture Transfers
The OpenGL PBO [ARB 04] mechanism provides for transfers that are asynchronous
on the CPU if an application can schedule enough work between initiating ...