CUDA 11.2 is introducing improved user experience and application performance through a combination of driver/toolkit compatibility enhancements, new memory suballocator feature, and compiler enhancements including an LLVM upgrade.
This new 11.2 release also delivers programming model updates to CUDA Graphs and Cooperative Groups, as well as expanding support for latest generation operating systems and compilers.
We describe some of these innovative feature introductions with more detail in a new blog Enhancing Memory Allocation with New NVIDIA CUDA 11.2 Features, and we will publish additional blogs on compiler enhancements shortly. Follow all CUDA Developer Blogs here.