关于多卡使用CUDA并行运算的问题
一直对于多卡使用CUDA有疑惑,最近看了篇文献,是这么说的:
By doubling the computing resources we observed that the overall performance of the application also doubles. This shows that the solution can benefit from a nearly linear speed improvement when adding more graphic boards to the system. It must be mentioned that the Nvidia SLI option, available for multi-GPU systems, is designed for OpenGL. Therefore, SLI must be disabled for CUDA, which requires direct programming of every installed GPU.
既然SLI方式不能用于设置CUDA,那对于同一主板上的两块显示卡如何设置才行呢,是不是装了驱动,装了CUDA一切必须的环境,就可以了呢(两卡默认可以承担单独运算)?
还有如果成立,依次类推,三卡也是如此吗?
另外,现在有单显卡双核心,使用一个显卡插槽,这样在进行大数据量的传输(极端)时,是不是比两块单核心的显卡使用单独的显卡插槽要慢(默认都是PCI-E 2.0)?
谢谢大家指教。