I remember Tim Murray posting else where on forums stating that WDDM limits the size of individual cudaMalloc calls to 256MB. Does this also mean because the call is going through the WDDM that the GPU Memory is Virtualized and supports paging?
I ask because I remember reading somewhere that WDDM v2.1 supports page faulting.
And as a follow up question, does anyone know if the Tesla Compute Only Drivers is supported by Parallel Nsight? (Driver Workaround to WDDM)