cuda.core 1.3.1 Release Notes#
Highlights#
New
minimal-sysctk12andminimal-sysctk13extras install onlycuda-bindingsat or above this release’s floor for that CUDA major, without the CUDA Toolkit wheels thatcu12andcu13add. Thecu12andcu13extras include them. Packages that depend oncuda-coreshould request one of these extras, so that pip and uv see the floor; with a plaincuda-corerequirement, an oldercuda-bindingsis installed and fails at import. See cuda-bindings Version Requirements. (#3459)cuda-core1.3.0 is yanked from PyPI. It declared its floors only in thecu12andcu13extras, so a resolver that cannot use 1.3.1 could fall back to it next to an oldercuda-bindings.
Fixes and enhancements#
Buffer.from_ipc_descriptor()no longer unmaps a newly imported memory pool buffer when another thread releases the same descriptor at the same time. The import now waits for the pending release to finish. (#3450)PinnedMemoryResourceconstruction without options no longer invalidates a CUDA graph capture in progress on another thread.cuda.core1.3.0 introduced acuMemPoolSetAccesscall in the constructor (#3405); the driver refuses that call, and invalidates any concurrent global-mode capture, unless it runs in relaxed capture mode. The fix applies the same relaxed-mode wrapper already used for pool attribute reads and writes andtrim(). Captures started with the defaultbegin_building()mode are unaffected; global mode is the default in several external libraries (for exampletorch.cuda.graph). (#3451)