Cuda 12.6 Release Notes Today Today

: Introduced Range Profiling APIs in CUPTI and Link-Time Optimized ( LTO ) callbacks in cuFFT.

For developers writing complex kernels, 12.6 introduces cudaLaunchKernelEx . This is an extension of the launch API that allows you to explicitly configure launch attributes (like shared memory configuration or cluster sizes) programmatically at the call site, rather than relying on global state or compile-time attributes. cuda 12.6 release notes today

NVIDIA is not afraid to clean house, and developers should take note: : Introduced Range Profiling APIs in CUPTI and