BLAS
Basic Linear Algebra Subprograms, a standard that defines a set of low-level routines for performing basic vector and matrix operations, forming a foundation of high-performance computing.
Timeline (last 90 days)
BLAS is a standard interface for a set of basic linear algebra routines, and matrix multiplication is the computational core of LLM inference; llama.cpp can be compiled and linked against different acceleration implementations: NVIDIA GPUs use CUDA, while AMD, Intel, or machines without a discrete GPU use Vulkan or CPU-side OpenBLAS, etc.
The U.S. Weather Bureau, the European Centre for Medium-Range Weather Forecasts, NASA's simulation systems, and the reference implementations of BLAS and LAPACK all have deep ties to Fortran.
BLAS defines low-level operation interfaces such as vector addition and matrix multiplication, and LAPACK implements higher-level algorithms such as linear equation solving, eigenvalue decomposition, and singular value decomposition on top of it, both of which were originally written in Fortran 77 with publicly available reference implementations.
The underlying linear algebra of almost all scientific computing languages (NumPy, MATLAB, R, Julia) ultimately calls libraries such as BLAS/LAPACK.
BLAS (Basic Linear Algebra Subprograms) is the foundational building block of deep learning and scientific computing, with neural network training and inference relying on large-scale matrix multiplication (GEMM).
Contiguous memory layout maximizes CPU cache hit rates and facilitates interfacing with underlying math libraries such as BLAS and LAPACK
All Facts (6)
BLAS is a standard interface for a set of basic linear algebra routines, and matrix multiplication is the computational core of LLM inference; llama.cpp can be compiled and linked against different acceleration implementations: NVIDIA GPUs use CUDA, while AMD, Intel, or machines without a discrete GPU use Vulkan or CPU-side OpenBLAS, etc.
50%UnverifiedThe U.S. Weather Bureau, the European Centre for Medium-Range Weather Forecasts, NASA's simulation systems, and the reference implementations of BLAS and LAPACK all have deep ties to Fortran.
50%UnverifiedBLAS defines low-level operation interfaces such as vector addition and matrix multiplication, and LAPACK implements higher-level algorithms such as linear equation solving, eigenvalue decomposition, and singular value decomposition on top of it, both of which were originally written in Fortran 77 with publicly available reference implementations.
50%UnverifiedThe underlying linear algebra of almost all scientific computing languages (NumPy, MATLAB, R, Julia) ultimately calls libraries such as BLAS/LAPACK.
50%UnverifiedBLAS (Basic Linear Algebra Subprograms) is the foundational building block of deep learning and scientific computing, with neural network training and inference relying on large-scale matrix multiplication (GEMM).
50%UnverifiedContiguous memory layout maximizes CPU cache hit rates and facilitates interfacing with underlying math libraries such as BLAS and LAPACK
50%