Graphics chip maker Nvidia has released version 6.0 of its CUDA Toolkit development software. This update is said to bring "some of the most significant new functionality in the history of CUDA," and it is free to download now.
Nvidia first announced version 6 of its CUDA parallel computing platform and programming model last November just ahead of the annual International Conference for High Performance Computing, Networking, Storage, and Analysis in Denver (also known, in short, as SC13). At that time we learned that the new toolkit would offer developers simplified programming “by enabling applications to access CPU and GPU memory without the need to manually copy data from one to the other, and makes it easier to add support for GPU acceleration in a wide range of programming languages”.
So, to recap, this development kit just simplifies things for the programmer, it makes memory management simpler. Memory copies between the system and GPU memory still need to be done and the latencies remain, however CUDA 6 handles the programming for developers transparently. Meanwhile, for comparison sake, AMD's HSA architecture provides your system and its developers with truly unified memory access.
Preceding this toolkit release, last month Mark Harris wrote on the Nvidia Developer's Blog to highlight the five most powerful new features in CUDA 6. As well as, of course, support for Unified Memory, he highlighted:
- CUDA on Tegra K1 mobile/embedded system-on-a-chip;
- XT and Drop-In library interfaces;
- remote development in NSight Eclipse Edition;
- many improvements to the CUDA developer tools
CUDA on the Tegra K1 is interesting as this chip has just been released as part of the Jetson TK1 developer board, late last week. It's already been benchmarked and found to be raw performance competitive with an Intel Celeron Bay Trail processor.
The (Extended) XT Libraries are also a great new addition to help make the most of multiple GPUs, offering automatic scaling and offer "drop-in speed-up of existing applications".
The CUDA 6 Production Release is available now for Windows, Linux and Mac OSX. You will also find the release notes, licensing info and online documentation available from that link.