Date: March 2017
The STREAM benchmark is a simple synthetic benchmark program that measures sustainable memory bandwidth (in MB/s) and the corresponding computation rate for simple vector kernels. The STREAM benchmark was developed by John McCalpin at the University of Delaware. (Link)
root@susie:~/sbc-bench > bin/stream > results/stream.txt
| System | CPU | Arch | MHz | OS | Copy | Scale | Add | Triad |
|---|---|---|---|---|---|---|---|---|
| Raspberry Pi 1B | Broadcom BCM2835 | ARMv6 | 700 | Raspbian 8 | 770.6 | 212.3 | 311.4 | 296.6 |
| Raspberry Pi 0W | Broadcom BCM2835 | ARMv6 | 1000 | Raspbian 8 | 842.5 | 274.4 | 393.6 | 384.6 |
| Beaglebone Black | AM3359 Cortex-A8 | ARMv7 | 1000 | Ubuntu 14.04 | 562.7 | 429.3 | 516.2 | 482.7 |
| Raspberry Pi 2B | Broadcom BCM2836 | ARMv7 | 900 | Raspbian 8 | 1359.5 | 1411.6 | 1602.7 | 1577.1 |
| Altera Atlas SoC | 5CSEMA4U23C6N A9 | ARMv7 | 925 | Ubuntu 16.04 | 935.5 | 1279.6 | 939.0 | 922.4 |
| Raspberry Pi 3 | Broadcom BCM2837 | ARMv8 | 1200 | Raspbian 8 | 2184.9 | 2177.3 | 2099.0 | 2092.3 |
| NanoPi NEO 2 | Allwinner H5 | ARMv8 | 1500 | Ubuntu 16.04 | 1848.4 | 1875.8 | 1730.5 | 1731.9 |
| Raspberry Pi 4 | Broadcom BCM2711 | ARMv8 | 1500 | Raspbian 10 | 3403.4 | 3265.5 | 2918.9 | 2882.6 |
The same benchmark was run on the consumer hardware below to provide a frame of reference:
| System | CPU | Arch | MHz | OS | Copy | Scale | Add | Triad |
|---|---|---|---|---|---|---|---|---|
| Apple MacMini 2007 | Core 2 Duo T7200 | Intel | 2000 | Linux | 2427.1 | 2404.0 | 2662.6 | 2645.0 |
| Xperia Z4 Tablet | Snapdragon 810 | ARMv8 | 2000 | Android 6.0 | 8726.0 | 10904.4 | 9009.0 | 9353.4 |
| MS Surface Go 8GB | Pentium 4415Y | Intel | 1600 | Ubuntu 16.04 | 22060.0 | 14749.2 | 16074.0 | 16064.2 |
| Linode 4GB VM | Xeon E5-2680 VM | Intel | 2800 | Ubuntu 14.04 | 9691.7 | 10193.7 | 13994.9 | 14472.3 |
Stream can detect multi-core systems and start the appropriate number of threads. Miss-detection can be corrected by setting the OpenMP environment variable "OMP_NUM_THREADS".