I have been wanting multi-device CUBLAS for a long time. Along with shared memory, this is going to make a lot of applications much more powerful, quickly. A

I have been wanting multi-device CUBLAS for a long time. Along with shared memory, this is going to make a lot of applications much more powerful, quickly. A major hindrance to getting started with CUDA can be dealing with memory allocation in the device. That is also made more complicated when dealing with threads and multiple devices, especially when workloads have to be somewhat tweaked depending on the number of devices available in a given box. Nvidia did good things here.