OpenBLAS 0.3.33 Released With Automatic "BIGNUMA" For More Than 256 CPU Cores
([Programming] 4 Hours Ago
OpenBLAS 0.3.33)
- Reference: 0001629284
- News link: https://www.phoronix.com/news/OpenBLAS-0.3.33-Released
- Source link:
OpenBLAS 0.3.33 is out today as the latest update to this vendor-neutral, optimized Basic Linear Algebra Subprograms "BLAS" library.
For a while now OpenBLAS has supported a "BIGNUMA" setting if wanting to enable support for more than 256 CPU cores. This experimental option has allowed supporting up to 1024 CPU cores and 128 NUMA nodes in the "BIGNUMA=1" configuration. Now with OpenBLAS 0.3.33, the BIGNUMA setting is automatic when the number of CPU cores on the system exceeds 256. Great to see especially with the next-gen Intel Xeon and AMD EPYC processors pushing system core counts even higher.
OpenBLAS 0.3.33 has also made various adjustments from the reference LAPACK implementation to realign with its upcoming v3.13 release.
OpenBLAS 0.3.33 also brings a number of bug fixes, automatic detection for Arm Neoverse V3, some optimizations for Neoverse N2, x86_64 build fixes, and optimized STRSM and DTRSM kernels for WebAssembly.
Downloads and more details on the OpenBLAS 0.3.33 release can be found via [1]GitHub .
[1] https://github.com/OpenMathLib/OpenBLAS/releases/tag/v0.3.33
For a while now OpenBLAS has supported a "BIGNUMA" setting if wanting to enable support for more than 256 CPU cores. This experimental option has allowed supporting up to 1024 CPU cores and 128 NUMA nodes in the "BIGNUMA=1" configuration. Now with OpenBLAS 0.3.33, the BIGNUMA setting is automatic when the number of CPU cores on the system exceeds 256. Great to see especially with the next-gen Intel Xeon and AMD EPYC processors pushing system core counts even higher.
OpenBLAS 0.3.33 has also made various adjustments from the reference LAPACK implementation to realign with its upcoming v3.13 release.
OpenBLAS 0.3.33 also brings a number of bug fixes, automatic detection for Arm Neoverse V3, some optimizations for Neoverse N2, x86_64 build fixes, and optimized STRSM and DTRSM kernels for WebAssembly.
Downloads and more details on the OpenBLAS 0.3.33 release can be found via [1]GitHub .
[1] https://github.com/OpenMathLib/OpenBLAS/releases/tag/v0.3.33