We’re pleased to announce that the CUDA® 7.0 Release Candidate (RC) is now available to all CUDA Registered Developers. As a member of the program, you can now download this preview of the CUDA Toolkit before it becomes available to the public.
Your existing applications can run faster simply by recompiling with this latest version of the CUDA Toolkit. Give it a try and share your experience via the Feedback link on the downloads page. And, as always, if you encounter any problems please file a bug!
Check out the Release Notes for a complete list of new features and platforms supported in this release. Highlights include support for:
• New C++11 features such as auto, lamba and many more
• New cuSOLVER library with:
◦ Key LAPACK dense solvers (Cholesky, LU, SVD, QA) that are 3–6x faster than MKL
◦ Sparse direct solvers that are 2–14x faster than CPU–only equivalents
• Runtime Compilation, enabling highly optimized kernels to be generated at runtime
• Calling Thrust inside compute kernels
• Performance improvements in key Thrust and cuFFT routines
• GPU core dump analysis in Nsight Eclipse Edition
• New initcheck and synccheck features in CUDA-MEMCHECK
So, will that double the usual speeds? (fingers & toes crossed)
Hmm, I'm getting +10kh with just their latest included driver ^^