Deprecated features in the latest/next Cuda toolkit

[Claggy]

For anyone who hasn't seen it, this came up on the Boinc dev board yesterday:

http://boinc.berkeley.edu/dev/forum_thread.php?id=9588

(Mac OS X) Support for 32-bit CUDA and OpenCL Applications on Mac OS X
 Developing and running 32-bit CUDA and OpenCL applications on Mac OS X platforms is no longer supported in the CUDA Toolkit and in the CUDA Driver. Legacy 32-bit CUDA and OpenCL applications will not run on this version of the CUDA Driver on Mac OS X platforms.

and:

Targeting sm_10 (G80) for CUDA Applications
The CUDA Toolkit no longer supports the sm_10 target architecture (the G80 architecture) for CUDA and OpenCL applications.

and:

http://boinc.berkeley.edu/dev/forum_thread.php?id=9588&postid=55766

Well, reading through there, the problems are a little bigger than just broken applications for Mac users. It'll be broken applications for A LOT of GPUs, as Deprecated Features writes:
The following features are deprecated in the current release of the CUDA software. The features still work in the current release, but their documentation may have been removed, and they will become officially unsupported in a future release. We recommend that developers employ alternative solutions to these features in their software.

Tesla and Quadro Products and CUDA Toolkit on 32-bit Windows Platforms
Support for the CUDA Toolkit on 32-bit Windows platforms is deprecated, as is support for Tesla and Quadro products for the CUDA driver on 32-bit Windows platforms. Additionally, on 64-bit Windows platforms, support for the following features for 32-bit CUDA and OpenCL applications is deprecated from the CUDA driver and CUDA toolkit, as appropriate:

Tesla and Quadro products
CUDA Toolkit scientific libraries, including cuBLAS, cuSPARSE, cuFFT, cuRAND, and NPP
Thrust
CUDA samples

This deprecation notice doesn't impact any 64-bit components.

Linux RHEL 5 and CentOS 5 Support
Support for CUDA on the RHEL 5 and CentOS 5 Linux distributions is deprecated in this CUDA release and will be dropped in a future release. We recommend that users and developers migrate to RHEL 6, which is supported in the current and future releases.

Support for sm_10, sm_11, sm_12, and sm_13 Architectures
The sm_10 architecture is deprecated within the CUDA Driver, and the sm_11, sm_12, and sm_13 architectures are deprecated within the CUDA Toolkit and the CUDA Driver. Support for these architectures will be removed in the next major version of the CUDA Toolkit and Driver. Note that support for the sm_10 architecture has already been removed from the CUDA Toolkit.

The latter means that any GPU with compute capability 1.0, 1.1, 1.2 and 1.3 on https://developer.nvidia.com/cuda-gpus will get deprecated in the future. This includes the GTX2xx series.

http://docs.nvidia.com/cuda/cuda-toolkit-release-notes/index.html#unsupported-features

Claggy

[Raistmer]

Are they crazy

[Urs Echternacht]

Are they crazy

They have quite a long breath ...

[arkayn]

Considering we know that 64-bit CUDA apps are slower than 32-bit CUDA apps, that is just plain ridiculous.