BLAS libraries benchmarks

Andrzej Wójtowicz

Document generation date: 2016-12-01 12:24:11

This document presents timing results for BLAS (Basic Linear Algebra Subprograms) libraries in R on diverse CPUs and GPUs.

Changelog

• 2016-12-01: results: updated timing for Intel Xeon E3-1275 v5; code: added possible compilation fix for invalid operands error in GotoBLAS2.
• 2016-11-30: results: added Intel Xeon E5-1620 v4.
• 2016-11-29: results: added Intel Xeon E3-1275 v5.
• 2016-11-25: results: added Intel Atom C2758.
• 2016-07-14: results: added Intel Core i5-6500; changed results view of gcbd benchmark to relative performance gain; changed reference CPU (Intel Pentium Dual-Core E5300) and GPU (NVIDIA GeForce GT 630M); code: fixed target architecture detection for Intel Core i5-6500-like CPUs in multi-threaded Atlas library; added info how to force target architecture in GotoBLAS2 and BLIS libraries.

Table of Contents

1. Configuration
2. Results per host
   • Intel Xeon E3-1275 v5
   • Intel Xeon E5-1620 v4
   • Intel Core i7-4790K + MSI GeForce GTX 980 Ti Lightning
   • Intel Core i5-4590 + NVIDIA GeForce GT 430
   • Intel Core i5-4590 + NVIDIA GeForce GTX 750 Ti
   • Intel Core i5-6500
   • Intel Core i5-3570
   • Intel Core i3-2120
   • Intel Core i3-3120M
   • Intel Core i5-3317U + NVIDIA GeForce GT 630M
   • Intel Atom C2758
   • Intel Pentium Dual-Core E5300
3. Results per library
   • Netlib
   • Atlas (st)
   • OpenBLAS
   • Atlas (mt)
   • GotoBLAS2
   • MKL
   • BLIS
   • cuBLAS

---

Configuration

OS: Debian Jessie, kernel 4.4

R software: Microsoft R Open (3.2.4)

Libraries:

CPU (single-threaded)	CPU (multi-threaded)	GPU
Netlib (debian package, blas 1.2.20110419, lapack 3.5.0)	OpenBLAS (debian package, 0.2.12)	NVIDIA cuBLAS (NVBLAS 6.5 + Intel MKL)
ATLAS (debian package, 3.10.2)	ATLAS (dev branch, 3.11.38)
	GotoBLAS2 (Survive fork, 3.141)
	Intel MKL (part of RevoMath package, 3.2.4)
	BLIS (dev branch, 0.2.0+/17.05.2016)

Hosts:

No.	CPU	GPU
1.	Intel Xeon E3-1275 v5	-
2.	Intel Xeon E5-1620 v4	-
3.	Intel Core i7-4790K (OC 4.5 GHz)	MSI GeForce GTX 980 Ti Lightning
4.	Intel Core i5-4590	NVIDIA GeForce GT 430
5.	Intel Core i5-4590	NVIDIA GeForce GTX 750 Ti
6.	Intel Core i5-6500	-
7.	Intel Core i5-3570	-
8.	Intel Core i3-2120	-
9.	Intel Core i3-3120M	-
10.	Intel Core i5-3317U	NVIDIA GeForce GT 630M
11.	Intel Atom C2758	-
12.	Intel Pentium Dual-Core E5300	-

Benchmarks: R-benchmark-25, Revolution, Gcbd.

Results per host

Intel Xeon E3-1275 v5

R-benchmark-25

2800x2800 cross-product matrix

Time in seconds - 10 runs - lower is better

Linear regr. over a 2000x2000 matrix

Time in seconds - 10 runs - lower is better

Eigenvalues of a 600x600 random matrix

Time in seconds - 10 runs - lower is better

Determinant of a 2500x2500 random matrix

Time in seconds - 10 runs - lower is better

Cholesky decomposition of a 3000x3000 matrix

Time in seconds - 10 runs - lower is better

Inverse of a 1600x1600 random matrix

Time in seconds - 10 runs - lower is better

Escoufier's method on a 45x45 matrix

Time in seconds - 10 runs - lower is better

Revolution benchmark

Matrix Multiply

Time in seconds - 10 runs - lower is better

Cholesky Factorization

Time in seconds - 10 runs - lower is better

Singular Value Deomposition

Time in seconds - 10 runs - lower is better

Principal Components Analysis

Time in seconds - 10 runs - lower is better

Linear Discriminant Analysis

Time in seconds - 10 runs - lower is better

Gcbd benchmark

Matrix Multiply

Performance gain regarding matrix size - reference: Netlib - from 50 to 5 runs - higher is better

QR Decomposition

Performance gain regarding matrix size - reference: Netlib - from 50 to 5 runs - higher is better

Singular Value Deomposition

Performance gain regarding matrix size - reference: Netlib - from 50 to 5 runs - higher is better

Triangular Decomposition

Performance gain regarding matrix size - reference: Netlib - from 50 to 5 runs - higher is better

Intel Xeon E5-1620 v4

R-benchmark-25

2800x2800 cross-product matrix

Time in seconds - 10 runs - lower is better

Linear regr. over a 2000x2000 matrix

Time in seconds - 10 runs - lower is better

Eigenvalues of a 600x600 random matrix

Time in seconds - 10 runs - lower is better

Determinant of a 2500x2500 random matrix

Time in seconds - 10 runs - lower is better

Cholesky decomposition of a 3000x3000 matrix

Time in seconds - 10 runs - lower is better

Inverse of a 1600x1600 random matrix

Time in seconds - 10 runs - lower is better

Escoufier's method on a 45x45 matrix

Time in seconds - 10 runs - lower is better

Revolution benchmark

Matrix Multiply

Time in seconds - 10 runs - lower is better

Cholesky Factorization

Time in seconds - 10 runs - lower is better

Singular Value Deomposition

Time in seconds - 10 runs - lower is better

Principal Components Analysis

Time in seconds - 10 runs - lower is better

Linear Discriminant Analysis

Time in seconds - 10 runs - lower is better

Gcbd benchmark

Matrix Multiply

Performance gain regarding matrix size - reference: Netlib - from 50 to 5 runs - higher is better

QR Decomposition

Performance gain regarding matrix size - reference: Netlib - from 50 to 5 runs - higher is better

Singular Value Deomposition

Performance gain regarding matrix size - reference: Netlib - from 50 to 5 runs - higher is better

Triangular Decomposition

Performance gain regarding matrix size - reference: Netlib - from 50 to 5 runs - higher is better

Intel Core i7-4790K + MSI GeForce GTX 980 Ti Lightning

R-benchmark-25

2800x2800 cross-product matrix

Time in seconds - 10 runs - lower is better

Linear regr. over a 2000x2000 matrix

Time in seconds - 10 runs - lower is better

Eigenvalues of a 600x600 random matrix

Time in seconds - 10 runs - lower is better

Determinant of a 2500x2500 random matrix

Time in seconds - 10 runs - lower is better

Cholesky decomposition of a 3000x3000 matrix

Time in seconds - 10 runs - lower is better

Inverse of a 1600x1600 random matrix

Time in seconds - 10 runs - lower is better

Escoufier's method on a 45x45 matrix

Time in seconds - 10 runs - lower is better

Revolution benchmark

Matrix Multiply

Time in seconds - 10 runs - lower is better

Cholesky Factorization

Time in seconds - 10 runs - lower is better

Singular Value Deomposition

Time in seconds - 10 runs - lower is better

Principal Components Analysis

Time in seconds - 10 runs - lower is better

Linear Discriminant Analysis

Time in seconds - 10 runs - lower is better

Gcbd benchmark

Matrix Multiply

Performance gain regarding matrix size - reference: Netlib - from 50 to 5 runs - higher is better

QR Decomposition

Performance gain regarding matrix size - reference: Netlib - from 50 to 5 runs - higher is better

Singular Value Deomposition

Performance gain regarding matrix size - reference: Netlib - from 50 to 5 runs - higher is better

Triangular Decomposition

Performance gain regarding matrix size - reference: Netlib - from 50 to 5 runs - higher is better

Intel Core i5-4590 + NVIDIA GeForce GT 430

R-benchmark-25

2800x2800 cross-product matrix

Time in seconds - 10 runs - lower is better

Linear regr. over a 2000x2000 matrix

Time in seconds - 10 runs - lower is better

Eigenvalues of a 600x600 random matrix

Time in seconds - 10 runs - lower is better

Determinant of a 2500x2500 random matrix

Time in seconds - 10 runs - lower is better

Cholesky decomposition of a 3000x3000 matrix

Time in seconds - 10 runs - lower is better

Inverse of a 1600x1600 random matrix

Time in seconds - 10 runs - lower is better

Escoufier's method on a 45x45 matrix

Time in seconds - 10 runs - lower is better

Revolution benchmark

Matrix Multiply

Time in seconds - 10 runs - lower is better

Cholesky Factorization

Time in seconds - 10 runs - lower is better

Singular Value Deomposition

Time in seconds - 10 runs - lower is better

Principal Components Analysis

Time in seconds - 10 runs - lower is better

Linear Discriminant Analysis

Time in seconds - 10 runs - lower is better

Gcbd benchmark

Matrix Multiply

Performance gain regarding matrix size - reference: Netlib - from 50 to 5 runs - higher is better

QR Decomposition

Performance gain regarding matrix size - reference: Netlib - from 50 to 5 runs - higher is better

Singular Value Deomposition

Performance gain regarding matrix size - reference: Netlib - from 50 to 5 runs - higher is better

Triangular Decomposition

Performance gain regarding matrix size - reference: Netlib - from 50 to 5 runs - higher is better

Intel Core i5-4590 + NVIDIA GeForce GTX 750 Ti

R-benchmark-25

2800x2800 cross-product matrix

Time in seconds - 10 runs - lower is better

Linear regr. over a 2000x2000 matrix

Time in seconds - 10 runs - lower is better

Eigenvalues of a 600x600 random matrix

Time in seconds - 10 runs - lower is better

Determinant of a 2500x2500 random matrix

Time in seconds - 10 runs - lower is better

Cholesky decomposition of a 3000x3000 matrix

Time in seconds - 10 runs - lower is better

Inverse of a 1600x1600 random matrix

Time in seconds - 10 runs - lower is better

Escoufier's method on a 45x45 matrix

Time in seconds - 10 runs - lower is better

Revolution benchmark

Matrix Multiply

Time in seconds - 10 runs - lower is better

Cholesky Factorization

Time in seconds - 10 runs - lower is better

Singular Value Deomposition

Time in seconds - 10 runs - lower is better

Principal Components Analysis

Time in seconds - 10 runs - lower is better

Linear Discriminant Analysis

Time in seconds - 10 runs - lower is better

Gcbd benchmark

Matrix Multiply

Performance gain regarding matrix size - reference: Netlib - from 50 to 5 runs - higher is better

QR Decomposition

Performance gain regarding matrix size - reference: Netlib - from 50 to 5 runs - higher is better

Singular Value Deomposition

Performance gain regarding matrix size - reference: Netlib - from 50 to 5 runs - higher is better

Triangular Decomposition

Performance gain regarding matrix size - reference: Netlib - from 50 to 5 runs - higher is better

Intel Core i5-6500

R-benchmark-25

2800x2800 cross-product matrix

Time in seconds - 10 runs - lower is better

Linear regr. over a 2000x2000 matrix

Time in seconds - 10 runs - lower is better

Eigenvalues of a 600x600 random matrix

Time in seconds - 10 runs - lower is better

Determinant of a 2500x2500 random matrix

Time in seconds - 10 runs - lower is better

Cholesky decomposition of a 3000x3000 matrix

Time in seconds - 10 runs - lower is better

Inverse of a 1600x1600 random matrix

Time in seconds - 10 runs - lower is better

Escoufier's method on a 45x45 matrix

Time in seconds - 10 runs - lower is better

Revolution benchmark

Matrix Multiply

Time in seconds - 10 runs - lower is better

Cholesky Factorization

Time in seconds - 10 runs - lower is better

Singular Value Deomposition

Time in seconds - 10 runs - lower is better

Principal Components Analysis

Time in seconds - 10 runs - lower is better

Linear Discriminant Analysis

Time in seconds - 10 runs - lower is better

Gcbd benchmark

Matrix Multiply

Performance gain regarding matrix size - reference: Netlib - from 50 to 5 runs - higher is better

QR Decomposition

Performance gain regarding matrix size - reference: Netlib - from 50 to 5 runs - higher is better

Singular Value Deomposition

Performance gain regarding matrix size - reference: Netlib - from 50 to 5 runs - higher is better

Triangular Decomposition

Performance gain regarding matrix size - reference: Netlib - from 50 to 5 runs - higher is better

Intel Core i5-3570

R-benchmark-25

2800x2800 cross-product matrix

Time in seconds - 10 runs - lower is better

Linear regr. over a 2000x2000 matrix

Time in seconds - 10 runs - lower is better

Eigenvalues of a 600x600 random matrix

Time in seconds - 10 runs - lower is better

Determinant of a 2500x2500 random matrix

Time in seconds - 10 runs - lower is better

Cholesky decomposition of a 3000x3000 matrix

Time in seconds - 10 runs - lower is better

Inverse of a 1600x1600 random matrix

Time in seconds - 10 runs - lower is better

Escoufier's method on a 45x45 matrix

Time in seconds - 10 runs - lower is better

Revolution benchmark

Matrix Multiply

Time in seconds - 10 runs - lower is better

Cholesky Factorization

Time in seconds - 10 runs - lower is better

Singular Value Deomposition

Time in seconds - 10 runs - lower is better

Principal Components Analysis

Time in seconds - 10 runs - lower is better

Linear Discriminant Analysis

Time in seconds - 10 runs - lower is better

Gcbd benchmark

Matrix Multiply

Performance gain regarding matrix size - reference: Netlib - from 50 to 5 runs - higher is better

QR Decomposition

Performance gain regarding matrix size - reference: Netlib - from 50 to 5 runs - higher is better

Singular Value Deomposition

Performance gain regarding matrix size - reference: Netlib - from 50 to 5 runs - higher is better

Triangular Decomposition

Performance gain regarding matrix size - reference: Netlib - from 50 to 5 runs - higher is better

Intel Core i3-2120

R-benchmark-25

2800x2800 cross-product matrix

Time in seconds - 10 runs - lower is better

Linear regr. over a 2000x2000 matrix

Time in seconds - 10 runs - lower is better

Eigenvalues of a 600x600 random matrix

Time in seconds - 10 runs - lower is better

Determinant of a 2500x2500 random matrix

Time in seconds - 10 runs - lower is better

Cholesky decomposition of a 3000x3000 matrix

Time in seconds - 10 runs - lower is better

Inverse of a 1600x1600 random matrix

Time in seconds - 10 runs - lower is better

Escoufier's method on a 45x45 matrix

Time in seconds - 10 runs - lower is better

Revolution benchmark

Matrix Multiply

Time in seconds - 10 runs - lower is better

Cholesky Factorization

Time in seconds - 10 runs - lower is better

Singular Value Deomposition

Time in seconds - 10 runs - lower is better

Principal Components Analysis

Time in seconds - 10 runs - lower is better

Linear Discriminant Analysis

Time in seconds - 10 runs - lower is better

Gcbd benchmark

Matrix Multiply

Performance gain regarding matrix size - reference: Netlib - from 50 to 5 runs - higher is better

QR Decomposition

Performance gain regarding matrix size - reference: Netlib - from 50 to 5 runs - higher is better

Singular Value Deomposition

Performance gain regarding matrix size - reference: Netlib - from 50 to 5 runs - higher is better

Triangular Decomposition

Performance gain regarding matrix size - reference: Netlib - from 50 to 5 runs - higher is better

Intel Core i3-3120M

R-benchmark-25

2800x2800 cross-product matrix

Time in seconds - 10 runs - lower is better

Linear regr. over a 2000x2000 matrix

Time in seconds - 10 runs - lower is better

Eigenvalues of a 600x600 random matrix

Time in seconds - 10 runs - lower is better

Determinant of a 2500x2500 random matrix

Time in seconds - 10 runs - lower is better

Cholesky decomposition of a 3000x3000 matrix

Time in seconds - 10 runs - lower is better

Inverse of a 1600x1600 random matrix

Time in seconds - 10 runs - lower is better

Escoufier's method on a 45x45 matrix

Time in seconds - 10 runs - lower is better

Revolution benchmark

Matrix Multiply

Time in seconds - 10 runs - lower is better

Cholesky Factorization

Time in seconds - 10 runs - lower is better

Singular Value Deomposition

Time in seconds - 10 runs - lower is better

Principal Components Analysis

Time in seconds - 10 runs - lower is better

Linear Discriminant Analysis

Time in seconds - 10 runs - lower is better

Gcbd benchmark

Matrix Multiply

Performance gain regarding matrix size - reference: Netlib - from 50 to 5 runs - higher is better

QR Decomposition

Performance gain regarding matrix size - reference: Netlib - from 50 to 5 runs - higher is better

Singular Value Deomposition

Performance gain regarding matrix size - reference: Netlib - from 50 to 5 runs - higher is better

Triangular Decomposition

Performance gain regarding matrix size - reference: Netlib - from 50 to 5 runs - higher is better

Intel Core i5-3317U + NVIDIA GeForce GT 630M

R-benchmark-25

2800x2800 cross-product matrix

Time in seconds - 10 runs - lower is better

Linear regr. over a 2000x2000 matrix

Time in seconds - 10 runs - lower is better

Eigenvalues of a 600x600 random matrix

Time in seconds - 10 runs - lower is better

Determinant of a 2500x2500 random matrix

Time in seconds - 10 runs - lower is better

Cholesky decomposition of a 3000x3000 matrix

Time in seconds - 10 runs - lower is better

Inverse of a 1600x1600 random matrix

Time in seconds - 10 runs - lower is better

Escoufier's method on a 45x45 matrix

Time in seconds - 10 runs - lower is better

Revolution benchmark

Matrix Multiply

Time in seconds - 10 runs - lower is better

Cholesky Factorization

Time in seconds - 10 runs - lower is better

Singular Value Deomposition

Time in seconds - 10 runs - lower is better

Principal Components Analysis

Time in seconds - 10 runs - lower is better

Linear Discriminant Analysis

Time in seconds - 10 runs - lower is better

Gcbd benchmark

Matrix Multiply

Performance gain regarding matrix size - reference: Netlib - from 50 to 5 runs - higher is better

QR Decomposition

Performance gain regarding matrix size - reference: Netlib - from 50 to 5 runs - higher is better

Singular Value Deomposition

Performance gain regarding matrix size - reference: Netlib - from 50 to 5 runs - higher is better

Triangular Decomposition

Performance gain regarding matrix size - reference: Netlib - from 50 to 5 runs - higher is better

Intel Atom C2758

R-benchmark-25

2800x2800 cross-product matrix

Time in seconds - 10 runs - lower is better

Linear regr. over a 2000x2000 matrix

Time in seconds - 10 runs - lower is better

Eigenvalues of a 600x600 random matrix

Time in seconds - 10 runs - lower is better

Determinant of a 2500x2500 random matrix

ATLAS (mt) crashes in this test

Time in seconds - 10 runs - lower is better

Cholesky decomposition of a 3000x3000 matrix

Time in seconds - 10 runs - lower is better

Inverse of a 1600x1600 random matrix

ATLAS (mt) crashes in this test

Time in seconds - 10 runs - lower is better

Escoufier's method on a 45x45 matrix

Time in seconds - 10 runs - lower is better

Revolution benchmark

Matrix Multiply

Time in seconds - 10 runs - lower is better

Cholesky Factorization

Time in seconds - 10 runs - lower is better

Singular Value Deomposition

Time in seconds - 10 runs - lower is better

Principal Components Analysis

Time in seconds - 10 runs - lower is better

Linear Discriminant Analysis

Time in seconds - 10 runs - lower is better

Gcbd benchmark

Matrix Multiply

Performance gain regarding matrix size - reference: Netlib - from 50 to 5 runs - higher is better

QR Decomposition

Performance gain regarding matrix size - reference: Netlib - from 50 to 5 runs - higher is better

Singular Value Deomposition

Performance gain regarding matrix size - reference: Netlib - from 50 to 5 runs - higher is better

Triangular Decomposition

ATLAS (mt) crashes in this test

Performance gain regarding matrix size - reference: Netlib - from 50 to 5 runs - higher is better

Intel Pentium Dual-Core E5300

R-benchmark-25

2800x2800 cross-product matrix

Time in seconds - 10 runs - lower is better

Linear regr. over a 2000x2000 matrix

Time in seconds - 10 runs - lower is better

Eigenvalues of a 600x600 random matrix

BLIS hangs in this test

Time in seconds - 10 runs - lower is better

Determinant of a 2500x2500 random matrix

Time in seconds - 10 runs - lower is better

Cholesky decomposition of a 3000x3000 matrix

Time in seconds - 10 runs - lower is better

Inverse of a 1600x1600 random matrix

Time in seconds - 10 runs - lower is better

Escoufier's method on a 45x45 matrix

Time in seconds - 10 runs - lower is better

Revolution benchmark

Matrix Multiply

Time in seconds - 10 runs - lower is better

Cholesky Factorization

Time in seconds - 10 runs - lower is better

Singular Value Deomposition

Time in seconds - 10 runs - lower is better

Principal Components Analysis

Time in seconds - 10 runs - lower is better

Linear Discriminant Analysis

Time in seconds - 10 runs - lower is better

Gcbd benchmark

Matrix Multiply

Performance gain regarding matrix size - reference: Netlib - from 50 to 5 runs - higher is better

QR Decomposition

Performance gain regarding matrix size - reference: Netlib - from 50 to 5 runs - higher is better

Singular Value Deomposition

Performance gain regarding matrix size - reference: Netlib - from 50 to 5 runs - higher is better

Triangular Decomposition

Performance gain regarding matrix size - reference: Netlib - from 50 to 5 runs - higher is better

Results per library

Netlib

R-benchmark-25

2800x2800 cross-product matrix

Time in seconds - 10 runs - lower is better

Linear regr. over a 2000x2000 matrix

Time in seconds - 10 runs - lower is better

Eigenvalues of a 600x600 random matrix

Time in seconds - 10 runs - lower is better

Determinant of a 2500x2500 random matrix

Time in seconds - 10 runs - lower is better

Cholesky decomposition of a 3000x3000 matrix

Time in seconds - 10 runs - lower is better

Inverse of a 1600x1600 random matrix

Time in seconds - 10 runs - lower is better

Escoufier's method on a 45x45 matrix

Time in seconds - 10 runs - lower is better

Revolution benchmark

Matrix Multiply

Time in seconds - 10 runs - lower is better

Cholesky Factorization

Time in seconds - 10 runs - lower is better

Singular Value Deomposition

Time in seconds - 10 runs - lower is better

Principal Components Analysis

Time in seconds - 10 runs - lower is better

Linear Discriminant Analysis

Time in seconds - 10 runs - lower is better

Gcbd benchmark

Matrix Multiply

Performance gain regarding matrix size - reference: Intel Pentium Dual-Core E5300 - from 50 to 5 runs - higher is better

QR Decomposition

Performance gain regarding matrix size - reference: Intel Pentium Dual-Core E5300 - from 50 to 5 runs - higher is better

Singular Value Deomposition

Performance gain regarding matrix size - reference: Intel Pentium Dual-Core E5300 - from 50 to 5 runs - higher is better

Triangular Decomposition

Performance gain regarding matrix size - reference: Intel Pentium Dual-Core E5300 - from 50 to 5 runs - higher is better

ATLAS (st)

R-benchmark-25

2800x2800 cross-product matrix

Time in seconds - 10 runs - lower is better

Linear regr. over a 2000x2000 matrix

Time in seconds - 10 runs - lower is better

Eigenvalues of a 600x600 random matrix

Time in seconds - 10 runs - lower is better

Determinant of a 2500x2500 random matrix

Time in seconds - 10 runs - lower is better

Cholesky decomposition of a 3000x3000 matrix

Time in seconds - 10 runs - lower is better

Inverse of a 1600x1600 random matrix

Time in seconds - 10 runs - lower is better

Escoufier's method on a 45x45 matrix

Time in seconds - 10 runs - lower is better

Revolution benchmark

Matrix Multiply

Time in seconds - 10 runs - lower is better

Cholesky Factorization

Time in seconds - 10 runs - lower is better

Singular Value Deomposition

Time in seconds - 10 runs - lower is better

Principal Components Analysis

Time in seconds - 10 runs - lower is better

Linear Discriminant Analysis

Time in seconds - 10 runs - lower is better

Gcbd benchmark

Matrix Multiply

Performance gain regarding matrix size - reference: Intel Pentium Dual-Core E5300 - from 50 to 5 runs - higher is better

QR Decomposition

Performance gain regarding matrix size - reference: Intel Pentium Dual-Core E5300 - from 50 to 5 runs - higher is better

Singular Value Deomposition

Performance gain regarding matrix size - reference: Intel Pentium Dual-Core E5300 - from 50 to 5 runs - higher is better

Triangular Decomposition

Performance gain regarding matrix size - reference: Intel Pentium Dual-Core E5300 - from 50 to 5 runs - higher is better

OpenBLAS

R-benchmark-25

2800x2800 cross-product matrix

Time in seconds - 10 runs - lower is better

Linear regr. over a 2000x2000 matrix

Time in seconds - 10 runs - lower is better

Eigenvalues of a 600x600 random matrix

Time in seconds - 10 runs - lower is better

Determinant of a 2500x2500 random matrix

Time in seconds - 10 runs - lower is better

Cholesky decomposition of a 3000x3000 matrix

Time in seconds - 10 runs - lower is better

Inverse of a 1600x1600 random matrix

Time in seconds - 10 runs - lower is better

Escoufier's method on a 45x45 matrix

Time in seconds - 10 runs - lower is better

Revolution benchmark

Matrix Multiply

Time in seconds - 10 runs - lower is better

Cholesky Factorization

Time in seconds - 10 runs - lower is better

Singular Value Deomposition

Time in seconds - 10 runs - lower is better

Principal Components Analysis

Time in seconds - 10 runs - lower is better

Linear Discriminant Analysis

Time in seconds - 10 runs - lower is better

Gcbd benchmark

Matrix Multiply

Performance gain regarding matrix size - reference: Intel Pentium Dual-Core E5300 - from 50 to 5 runs - higher is better

QR Decomposition

Performance gain regarding matrix size - reference: Intel Pentium Dual-Core E5300 - from 50 to 5 runs - higher is better

Singular Value Deomposition

Performance gain regarding matrix size - reference: Intel Pentium Dual-Core E5300 - from 50 to 5 runs - higher is better

Triangular Decomposition

Performance gain regarding matrix size - reference: Intel Pentium Dual-Core E5300 - from 50 to 5 runs - higher is better

ATLAS (mt)

R-benchmark-25

2800x2800 cross-product matrix

Time in seconds - 10 runs - lower is better

Linear regr. over a 2000x2000 matrix

Time in seconds - 10 runs - lower is better

Eigenvalues of a 600x600 random matrix

Time in seconds - 10 runs - lower is better

Determinant of a 2500x2500 random matrix

Library crashes on Intel Atom C2758 in this test

Time in seconds - 10 runs - lower is better

Cholesky decomposition of a 3000x3000 matrix

Time in seconds - 10 runs - lower is better

Inverse of a 1600x1600 random matrix

Library crashes on Intel Atom C2758 in this test

Time in seconds - 10 runs - lower is better

Escoufier's method on a 45x45 matrix

Time in seconds - 10 runs - lower is better

Revolution benchmark

Matrix Multiply

Time in seconds - 10 runs - lower is better

Cholesky Factorization

Time in seconds - 10 runs - lower is better

Singular Value Deomposition

Time in seconds - 10 runs - lower is better

Principal Components Analysis

Time in seconds - 10 runs - lower is better

Linear Discriminant Analysis

Time in seconds - 10 runs - lower is better

Gcbd benchmark

Matrix Multiply

Performance gain regarding matrix size - reference: Intel Pentium Dual-Core E5300 - from 50 to 5 runs - higher is better

QR Decomposition

Performance gain regarding matrix size - reference: Intel Pentium Dual-Core E5300 - from 50 to 5 runs - higher is better

Singular Value Deomposition

Performance gain regarding matrix size - reference: Intel Pentium Dual-Core E5300 - from 50 to 5 runs - higher is better

Triangular Decomposition

Library crashes on Intel Atom C2758 in this test

Performance gain regarding matrix size - reference: Intel Pentium Dual-Core E5300 - from 50 to 5 runs - higher is better

GotoBLAS2

R-benchmark-25

2800x2800 cross-product matrix

Time in seconds - 10 runs - lower is better

Linear regr. over a 2000x2000 matrix

Time in seconds - 10 runs - lower is better

Eigenvalues of a 600x600 random matrix

Time in seconds - 10 runs - lower is better

Determinant of a 2500x2500 random matrix

Time in seconds - 10 runs - lower is better

Cholesky decomposition of a 3000x3000 matrix

Time in seconds - 10 runs - lower is better

Inverse of a 1600x1600 random matrix

Time in seconds - 10 runs - lower is better

Escoufier's method on a 45x45 matrix

Time in seconds - 10 runs - lower is better

Revolution benchmark

Matrix Multiply

Time in seconds - 10 runs - lower is better

Cholesky Factorization

Time in seconds - 10 runs - lower is better

Singular Value Deomposition

Time in seconds - 10 runs - lower is better

Principal Components Analysis

Time in seconds - 10 runs - lower is better

Linear Discriminant Analysis

Time in seconds - 10 runs - lower is better

Gcbd benchmark

Matrix Multiply

Performance gain regarding matrix size - reference: Intel Pentium Dual-Core E5300 - from 50 to 5 runs - higher is better

QR Decomposition

Performance gain regarding matrix size - reference: Intel Pentium Dual-Core E5300 - from 50 to 5 runs - higher is better

Singular Value Deomposition

Performance gain regarding matrix size - reference: Intel Pentium Dual-Core E5300 - from 50 to 5 runs - higher is better

Triangular Decomposition

Performance gain regarding matrix size - reference: Intel Pentium Dual-Core E5300 - from 50 to 5 runs - higher is better

MKL

R-benchmark-25

2800x2800 cross-product matrix

Time in seconds - 10 runs - lower is better

Linear regr. over a 2000x2000 matrix

Time in seconds - 10 runs - lower is better

Eigenvalues of a 600x600 random matrix

Time in seconds - 10 runs - lower is better

Determinant of a 2500x2500 random matrix

Time in seconds - 10 runs - lower is better

Cholesky decomposition of a 3000x3000 matrix

Time in seconds - 10 runs - lower is better

Inverse of a 1600x1600 random matrix

Time in seconds - 10 runs - lower is better

Escoufier's method on a 45x45 matrix

Time in seconds - 10 runs - lower is better

Revolution benchmark

Matrix Multiply

Time in seconds - 10 runs - lower is better

Cholesky Factorization

Time in seconds - 10 runs - lower is better

Singular Value Deomposition

Time in seconds - 10 runs - lower is better

Principal Components Analysis

Time in seconds - 10 runs - lower is better

Linear Discriminant Analysis

Time in seconds - 10 runs - lower is better

Gcbd benchmark

Matrix Multiply

Performance gain regarding matrix size - reference: Intel Pentium Dual-Core E5300 - from 50 to 5 runs - higher is better

QR Decomposition

Performance gain regarding matrix size - reference: Intel Pentium Dual-Core E5300 - from 50 to 5 runs - higher is better

Singular Value Deomposition

Performance gain regarding matrix size - reference: Intel Pentium Dual-Core E5300 - from 50 to 5 runs - higher is better

Triangular Decomposition

Performance gain regarding matrix size - reference: Intel Pentium Dual-Core E5300 - from 50 to 5 runs - higher is better

BLIS

R-benchmark-25

2800x2800 cross-product matrix

Time in seconds - 10 runs - lower is better

Linear regr. over a 2000x2000 matrix

Time in seconds - 10 runs - lower is better

Eigenvalues of a 600x600 random matrix

Library hangs on Intel Pentium Dual-Core E5300 in this test

Time in seconds - 10 runs - lower is better

Determinant of a 2500x2500 random matrix

Time in seconds - 10 runs - lower is better

Cholesky decomposition of a 3000x3000 matrix

Time in seconds - 10 runs - lower is better

Inverse of a 1600x1600 random matrix

Time in seconds - 10 runs - lower is better

Escoufier's method on a 45x45 matrix

Time in seconds - 10 runs - lower is better

Revolution benchmark

Matrix Multiply

Time in seconds - 10 runs - lower is better

Cholesky Factorization

Time in seconds - 10 runs - lower is better

Singular Value Deomposition

Time in seconds - 10 runs - lower is better

Principal Components Analysis

Time in seconds - 10 runs - lower is better

Linear Discriminant Analysis

Time in seconds - 10 runs - lower is better

Gcbd benchmark

Matrix Multiply

Performance gain regarding matrix size - reference: Intel Pentium Dual-Core E5300 - from 50 to 5 runs - higher is better

QR Decomposition

Performance gain regarding matrix size - reference: Intel Pentium Dual-Core E5300 - from 50 to 5 runs - higher is better

Singular Value Deomposition

Performance gain regarding matrix size - reference: Intel Pentium Dual-Core E5300 - from 50 to 5 runs - higher is better

Triangular Decomposition

Performance gain regarding matrix size - reference: Intel Pentium Dual-Core E5300 - from 50 to 5 runs - higher is better

cuBLAS

R-benchmark-25

2800x2800 cross-product matrix

Time in seconds - 10 runs - lower is better

Linear regr. over a 2000x2000 matrix

Time in seconds - 10 runs - lower is better

Eigenvalues of a 600x600 random matrix

Time in seconds - 10 runs - lower is better

Determinant of a 2500x2500 random matrix

Time in seconds - 10 runs - lower is better

Cholesky decomposition of a 3000x3000 matrix

Time in seconds - 10 runs - lower is better

Inverse of a 1600x1600 random matrix

Time in seconds - 10 runs - lower is better

Escoufier's method on a 45x45 matrix

Time in seconds - 10 runs - lower is better

Revolution benchmark

Matrix Multiply

Time in seconds - 10 runs - lower is better

Cholesky Factorization

Time in seconds - 10 runs - lower is better

Singular Value Deomposition

Time in seconds - 10 runs - lower is better

Principal Components Analysis

Time in seconds - 10 runs - lower is better

Linear Discriminant Analysis

Time in seconds - 10 runs - lower is better

Gcbd benchmark

Matrix Multiply

Performance gain regarding matrix size - reference: NVIDIA GeForce GT 630M - from 50 to 5 runs - higher is better

QR Decomposition

Performance gain regarding matrix size - reference: NVIDIA GeForce GT 630M - from 50 to 5 runs - higher is better

Singular Value Deomposition

Performance gain regarding matrix size - reference: NVIDIA GeForce GT 630M - from 50 to 5 runs - higher is better

Triangular Decomposition

Performance gain regarding matrix size - reference: NVIDIA GeForce GT 630M - from 50 to 5 runs - higher is better