New algorithm (CUSPARSE_SPMM_CSR_ALG3) for Sparse Matrix - Matrix Multiplication (cusparseSpMM) with better performance especially for small matrices. Support for deterministic and non-deterministic computation. Support for mixed regular-complex data type computation. Support for regular/complex bfloat16 data types for both uniform and mixed-precision computation. Extended functionalities for cusparseSpMV:. New algorithms for CSR/COO Sparse Matrix - Vector Multiplication (cusparseSpMV) with better performance. New Tensor Core-accelerated Block Sparse Matrix - Matrix Multiplication (cusparseSpMM) and introduction of the Blocked-Ellpack storage format. The user has to link libcusolver.so with the correct version of libcublas.so. However, it breaks backward compatibility. This reduces the binary size of libcusolver.so. libcusolver.so no longer links libcublas_static.a instead, it depends on libcublas.so. 
GESVDR computes partial spectrum with random sampling, an order of magnitude faster than GESVD.
New singular value decomposition (GESVDR) is added. Previously, when using recent versions of VS 2019 host compiler, a call to pow(double, int) or pow(float, int) in host or device code sometimes caused build failures. This section summarizes the changes in CUDA 11.2.1 (11.2 Update 1) since the 11.2.0 GA release. Parallel Nsight 2.0 now available for Windows developers with new debugging and profiling features. GPU binary disassembler for Fermi architecture (cuobjdump). C++ debugging in CUDA-GDB for Linux and MacOS. Automated Performance Analysis in Visual Profiler. GPUDirect v2.0 support for Peer-to-Peer Communication. Layered Textures for working with same size/format textures at larger sizes and higher performance. Nvidia Performance Primitives (NPP) library for image/video processing. Thrust library of templated performance primitives such as sort, reduce, etc. C++ new/delete and support for virtual functions. No-copy pinning of system memory, a faster alternative to cudaMallocHost(). Use all GPUs in the system concurrently from a single host thread.
0 kommentar(er)
