The CUDA 6.5 production release is now out and it supports Ubuntu 14.04! Now we can all enjoy the latest CUDA and Ubuntu releases. Register for the upcoming webinars about the new features in CUDA 6.5: Aug. 26, 2014 “CUDA 6.5 Overview and Features“, and Sept. 17, 2014, “Performance Overview“. Meanwhile, it is time to upgrade from that older Ubuntu 12.04 release.
BTW: The Ubuntu upgrade and CUDA 6.5 install worked nicely on my system. CUDA 6.5 also restores the performance in my deep-learning code, which dropped since CUDA 5.5 due to a compiler performance regression.
CUDA 6.5 features 64-bit ARM support and the simplified programming model enabled by Unified Memory.
- Support for 64-bit ARM-based systems.
- Microsoft Visual Studio 2013 compatibility.
- Expanded support for Block Sparse Row (BSR) matrix format in cuSPARSE routines.
- cuFFT callbacks for higher performance custom processing on input or output data.
- Improved debugging for CUDA FORTRAN applications.
- New profiler Application Replay mode reduces the time required to collect multiple metrics when profiling a complex application.
- Updated CUDA Occupancy Calculator API provides optimal kernel launch configurations.
- New nvprune utility optimizes the size of object files by removing code not needed for target GPU architectures.
For more information:
- Parallel For-All: Ten Ways CUDA 6.5 Improves Performance and Productivity