The CUDA Toolkit 7.0 Release Candidate (RC) is now available to members of NVIDIA’s free registered developer program. Especially interesting is the claim of 3-6x faster LAPACK dense solvers over MKL (The Intel Math Kernel Library).
C++11 support makes it easier for C++ developers to accelerate their applications
- Write less code with ‘auto’ and ‘lambda’, especially when using the Thrust template library.
New cuSOLVER library of dense and sparse direct solvers
- Significant acceleration for Computer Vision, CFD, Computational Chemistry, and Linear Optimization applications.
- Key LAPACK dense solvers 3-6x faster than MKL.
- Dense solvers include Cholesky, LU, SVD and QR
- Sparse direct solvers 2-14x faster than CPU-only equivalents.
- Sparse solvers include direct solvers and eigensolvers
Runtime Compilation enables highly optimized kernels to be generated at runtime.
- Improve performance by removing conditional logic and only evaluating special cases when necessary.