Question regarding CUDA 8 and Unified Memory

Accelerated Computing CUDA CUDA Programming and Performance
1

Hello,

in [url]https://devblogs.nvidia.com/parallelforall/cuda-8-features-revealed/[/url] it was mentioned that cudaMemAdvise and cudaMemPrefetchAsync will be added in CUDA 8. Will these two features be usable with a Pascal GPU only or are they available for all devices that support Unified Memory?

And, are those features available in the RC for CUDA 8 or only in the final release? (The download page [url]https://developer.nvidia.com/cuda-toolkit[/url] does talk about Unified Memory data migration that will be available at release only, but I am not sure whether that refers to the page-faulting capability on the device or the API extensions I mentioned in the above paragraph.)

Thanks in advance for your reply!

2

quick scan over cuda8rc reveals that they are implemented and documented here. moreover, they are referenced by various exe/dlls, such as cufft64_80.dll and bandwidthTest.exe (but not mentioned in bandwidthTest.cu)