New debugging features in CUDA 4.0

We’ve listened to your feedback and added new features (and bug fixes!) to the CUDA debugging tools in CUDA Toolkit 4.0 at
Please give the new release a try and let us know if you find any problems or would like to see further enhancements in the next release. (Feedback instructions below)

New features in CUDA-GDB and CUDA-MEMCHECK in the 4.0 Release Candidate #2 release include:
1. CUDA-GDB is now enabled on MACOS 10.6.5 in addition to all CUDA supported Linux distros.
2. C++ debugging is now officially supported.
3. Support added for conditional breakpoints.
4. Automatically break on every CUDA kernel launch; no need to explicitly set breakpoints on kernel names or line numbers (unless that’s what you want).
5. Debug applications that use textures on Tesla-architecture GPUs and new support for reading texture memory on all CUDA-capable GPUs.
6. Disassembly view for Fermi architecture GPUs in addition to the existing support for Tesla architecture.
7. CUDA-MEMCHECK now detects a wider range of application error types.
8. Debugger APIs are now part of a separate CUDA SDK installer to help NVIDIA partners integrate into more debugger frameworks.
9. Support has been added for 3-dimensional grid launches.

For additional information, please see the User Guides:
CUDA-GDB User Guide

Please check out the improvements in the CUDA Toolkit 4.0 at and let us know if you discover any problems we missed. The best way to provide feedback is via the GPU Computing registered developer program at using the Bug Report or Feedback links, but you can also send email directly to