Intel Compute Stick Performance Surprises Under Ubuntu Linux

Written by Michael Larabel in Computers on 1 June 2015 at 03:10 PM EDT. Page 1 of 4. 8 Comments.

Since last week I've been testing the Intel Compute Stick, the quad-core Atom Z3735F Atom powered PC that's a little bigger than the size of an HDMI connector. In this article are some benchmarks of this $150 quad-core + 32GB eMMC + 2GB RAM tiny computer in a variety of benchmarks comparing it to other low-power x86 and ARMv7 hardware.

In making for an interesting comparison, the systems I'm comparing the Bay-Trail-T-based Compute Stick to for judging its performance are:

- The Intel DN2820FYK Bay Trail NUC with Celeron N2820 dual-core 2.4GHz SoC.

- A Haswell-based Apple Mac Mini with Intel Core i5 4278U quad-core 3.1GHz processor.

- The Broadwell-U-based Intel NUC5i3RYB with Core i3 5010U Broadwell processor.

- The CompuLab Utilite with i.MX6 quad-core Cortex-A9 1.0GHz SoC.

- NVIDIA's Jetson TK1 quad-core Cortex-A15 development board.

All of the Intel x86 systems were running Ubuntu 15.04 with the Linux 4.1 kernel and the rest of the same software make-up. With the Utilite, Ubuntu 12.04 with the Linux 3.0 kernel was used due to newer releases not being supported by CompuLab. With the Jetson TK1 was Ubuntu 14.04 with the Linux 3.10 kernel, likewise due to NVIDIA not providing any newer official images. Due to running OpenGL (non-GLES) tests, only for the x86 systems are the graphics test results while for all of the processor-bound tests are results for all six systems in total.

It's important to note the testing was done on the x86 systems with the Linux 4.1 development kernel, which I've found to offer significant performance advantages over Linux 3.19 and older.

Being added to these results in the days ahead will be the new CompuLab Fitlet AMD SoC system, MIPS Creator development board, and perhaps some other low-power device tests too that I didn't have done in time for this initial Compute Stick comparison with Ubuntu Linux.

All of the benchmarks across the x86/ARM systems were configured the same way and run in a fully-automated and reproducible manner using the open-source Phoronix Test Suite benchmarking software. Due to varying differences beyond just the CPU/SoC, the power consumption of the different systems weren't compared for today's article.


Related Articles