I am looking for eigenvalue decomposition of large matrices. As i understand, the old interface, syevd, had limited matrix size because it used 32-bit integers for the work size and so has been limited by ~2 G doubles. The new interface, Xsyevd, uses int64_t integers and is supposed not have that limitation. However, it cannot go beyond ~25000 matrix size and the subroutine, which is supposed to provide the buffer size on the device, fails even so the device memory maximum is not reached.
Related topics
Topic | Replies | Views | Activity | |
---|---|---|---|---|
Limitations of cusolverDn<t>syevd() | 1 | 456 | January 30, 2025 | |
Syevj performance | 3 | 1054 | May 20, 2021 | |
The Origin of the m <= 32 and n <= 32 Limitations in gesvdjBatched? | 1 | 507 | September 14, 2023 | |
How to perform eigenvalue decompositions for large matrices with Python | 9 | 862 | May 17, 2024 | |
Excessive lwork memory request in cusolverDnDsyevd | 3 | 473 | January 23, 2020 | |
cuSolver memory limit? svd solver cannot handle >128 matrices | 3 | 896 | July 17, 2023 | |
cusolver: no batched version of 'cusolverDnSsyevd' library to solve large small symmetric eigenvalue and eigenvector | 0 | 431 | June 19, 2017 | |
Gesvdj vs gesvda vs syevj for finding eigenvalues/singular values | 1 | 917 | March 12, 2023 | |
Truncated Orthogonal Decomposition of very large Matrix using cusolverDnSgesvd | 0 | 470 | April 18, 2019 | |
cusparseScsrsv_analysis: any limitation of the metric size? | 3 | 1704 | February 21, 2013 |