Every vertical blank, the hypervisor DMAs this buffer to the GPU memory, and then flips the GPU's visible page to the new data synchronously with the vertical blank signal.
对于垂直方向上的每个空格来说，管理程序都会采用 DMA 方式将自己的缓冲区读到 GPU 的内存中，然后将 GPU 的可视页刷新为与垂直空格信号同步的数据。
Each CUDA-capable GPU node includes local DDR3 SDRAM as well as a 16-lane PCI Express? gen2 interface to the system backplane, providing maximum data throughput direct to GPU memory.
每个CUDA GPU节点包括本地的DDR3 SDRAM以及一个16通道PCI二代系统的背板接口，直接向GPU内存提供最大的数据流量。
This might sound simple, but there are a lot of design decisions around making sure that GPU memory is not allocated without the user asking for it and still allowing people to build what they want.
这听起来貌似很简单，但是这背后有着大量的设计决定以确保GPU内存没有分配给不需要之人，且仍然能够允许人们构造他们之所想。
应用推荐