Skip to content
SPECIFICTION
Get listed

STREAM

cs.virginia.edu

WindowsmacOSLinux
Visit cs.virginia.edu
Spec score 7.2 #59 of 73 in PC Benchmark Software

Spec sheet

Purpose
STREAM is a synthetic benchmark that measures sustainable memory bandwidth and computation rates for simple vector kernels.Read 5 Oct 2026 · cs.virginia.edu
Free plan
YesRead 22 Sep 2026 · cs.virginia.edu
Cache-sized data
STREAM is designed to use datasets much larger than available cache so results better reflect performance for large vector-style applications.Read 5 Oct 2026 · cs.virginia.edu
CLI or automation
YesRead 22 Sep 2026 · cs.virginia.edu
Comparison database
YesRead 22 Sep 2026 · cs.virginia.edu
Custom test profiles
YesRead 22 Sep 2026 · cs.virginia.edu
Intended use
It is designed to use datasets much larger than available cache, to make results more indicative of large vector style applications.Read 5 Oct 2026 · cs.virginia.edu
Kernels
The benchmark repeats four kernels: Copy, Scale, Add, and Triad.Read 5 Oct 2026 · cs.virginia.edu
Maintainer
The FAQ identifies John McCalpin as STREAM's creator and maintainer.Read 5 Oct 2026 · cs.virginia.edu
Operating systems
Linux; Windows; DOS; Power MacRead 22 Sep 2026 · cs.virginia.edu
Parallel runs
The standard C and Fortran code includes OpenMP directives, and an MPI version is available for cluster runs.Read 5 Oct 2026 · cs.virginia.edu
Project hosting
The site says STREAM is hosted at the University of Virginia in Charlottesville, Virginia.Read 5 Oct 2026 · cs.virginia.edu
Project status
The site describes STREAM as an independent academic project hosted at the University of Virginia.Read 5 Oct 2026 · cs.virginia.edu
Published results
Results labeled as standard STREAM benchmark results must conform to the STREAM Run Rules; modified or nonconforming runs must be clearly labeled.Read 5 Oct 2026 · cs.virginia.edu
Publishing results
Published results called “STREAM benchmark results” must conform to the run rules, and modified-code results must be clearly labelled.Read 5 Oct 2026 · cs.virginia.edu
Result export formats
comma-delimited raw data filesRead 22 Sep 2026 · cs.virginia.edu
Results database
The site publishes benchmark results in tables covering bandwidth, machine balance, and equivalent MFLOPS.Read 5 Oct 2026 · cs.virginia.edu
Source code
The site recommends compiling the benchmark from its C or Fortran source code.Read 5 Oct 2026 · cs.virginia.edu
Source languages
The maker recommends compiling the source code, which is available in C or Fortran.Read 5 Oct 2026 · cs.virginia.edu
Standards
The standard results set uses C or Fortran versions with 64-bit data types on production hardware and excludes simulated results and significantly modified code.Read 5 Oct 2026 · cs.virginia.edu
Tuned results
The tuned results set allows code modifications, including assembly language, while requiring 64-bit data types on production hardware.Read 5 Oct 2026 · cs.virginia.edu
Use and redistribution
The run rules permit use and redistribution, and permit modification for personal use including commercial use, subject to publication restrictions.Read 5 Oct 2026 · cs.virginia.edu
Warranty
The run rules state that no warranty is expressed or implied.Read 5 Oct 2026 · cs.virginia.edu
Windows and Linux binaries
The FAQ lists precompiled binaries for Windows and Linux but describes the available binaries as obsolete and recommends compiling the source.Read 5 Oct 2026 · cs.virginia.edu
Workload areas
memory bandwidth; vector-kernel computation rate; COPY; SCALE; ADD; TRIADRead 22 Sep 2026 · cs.virginia.edu
Workloads
The benchmark runs four kernels: Copy, Scale, Add, and Triad.Read 5 Oct 2026 · cs.virginia.edu

Where Specifiction places it

Alternatives to STREAMAll alternatives →