Cincoze GM-1000 Embedded GPU Computer Review

Cincoze GM-1000 is an industrial embedded GPU computer. It is highly expandable including supporting an embedded MXM GPU. As a rugged computer, it can either be passively cooled or actively cooled with add-on fans. It can support an operating temperature ranging from -40°C – 70°C and is also able to survive being dropped or hit with a vibration/shock tolerance of 5G/50G. It is marketed as having all the features required for a compact, reliable, and high-performance computing system for field applications in machine vision, image processing, and artificial intelligence.

In this review, I’ll cover some performance metrics from both Windows and Ubuntu and also discuss the thermals.

Hardware Overview

The model loaned by Cincoze for review came with an Intel Core i7-9700TE which is an eight-core 8-thread 1.80 GHz Coffee Lake-R processor boosting to 3.80 GHz with Intel’s UHD Graphics 630. It also came with an NVIDIA Quadro Embedded P2000 MXM module installed, 32GB of RAM, and a 512GB NVMe drive together with an expansion of 2x 10G Intel X550 RJ45 LAN ports and an ‘External Fan Kit’ which allows four 80mm fans to be installed on top of the device:

The unit cosmetically looks like a giant heat sink and measures 260 mm x 200 mm x 85 mm (10.24 x 7.87 x 3.35 inches) with a weight of 4.6 kg (10 pounds 2.26 ounces). It also requires a large power brick (GST220A24-CIN) rated at 221W.

The full specification including all configurable/expansion options for the GM-1000 include:

Review Methodology

Being an industrial unit I decided to review using a dual-boot of Windows 10 IOT Enterprise LTSC version 1809 and Ubuntu 18.04 LTS point release 5 and test with a selection of commonly used Windows benchmarks and/or equivalents for Linux together with Thomas Kaiser’s ‘sbc-bench’ which is a small set of different CPU performance tests focusing on server performance when run on Ubuntu. Additionally, I used ‘Phoronix Test Suite’ to benchmark the same set of tests on both Windows and Ubuntu for comparison purposes. On Ubuntu, I also compiled the v5.4 Linux kernel using the default config as a test of performance.

Prior to benchmarking, I perform all the necessary installations and updates. I also capture some basic details of the device for each OS.

Installation Issues

The embedded GPU computer came with Ubuntu already installed with a preconfigured user (‘cincoze’) and an unknown password. To boot Ubuntu with a single monitor there were effectively six connection options of either the HDMI or DisplayPort outputs on the rear of the device or any of the four DisplayPort outputs on the front of the device which are provided by the MXM-P2000 module. However when trying to boot into the BIOS only the four DisplayPort outputs on the front of the device display anything. This was also the case when trying to boot from an installation USB.

So Windows was installed using a front DisplayPort connection and after installation Ethernet was not detected. Fortunately, all the required Windows drivers are available from the Cincoze ‘SUPPORTS’ webpage with the exception of the NVIDIA driver required for the P2000. After installing the NVIDIA driver when booting with a monitor connected to a rear output port the display used Intel UHD Graphics 630 and if booted when connected to a front DisplayPort the NVIDIA Quadro P2000 was used.

Installing Ubuntu was similarly straightforward by performing an installation without selecting the ‘Install third-party software for graphics’ option. Although this resulted with no display when booting with the monitor connected to a rear HDMI output (although I didn’t try ‘nomodeset’) the display worked with Ubuntu when connected to a front DisplayPort output as it used the ‘llvmpipe’ graphics driver allowing me to use ‘Additional Drivers’ in ‘Software & Updates’ to select and use the NVIDIA driver metapackage. Then regardless of whether connected using a rear or front output port the graphics display driver always showed as Quadro P2000. Importantly though when connected by a rear output port and then booting the GRUB menu is not visible as this can only be seen when connected by one of the front DisplayPort outputs.

One other point to note was that the ‘Selenium’ test from the ‘Phoronix Test Suite’ benchmarks refused to run the ‘Chrome’ option so the Octane tests had to be run manually and edited into the final results.

Windows Performance

I first installed Windows 10 Enterprise LTSC Evaluation version 1809 and updated to OS build 17763.1637. A quick look at the hardware information shows:

I then set the power plan to ‘Ultimate Performance’:

and ran some benchmarking tools to look at performance under Windows:

 

I also tested the Blender ‘BMW’ benchmark for both CPU and GPU (CUDA):

For my specific set of Phoronix Test Suite tests the results were:

The CPU performance is limited due to the low 35W TDP of the processor coupled with PL1 being set to match it at 35W. The result is ‘Power Limit Throttling’ where the processor is down-clocked or throttled to ensure it remains within its thermal envelope:

Ubuntu Performance

After shrinking the Windows partition in half and creating a new partition I installed Ubuntu using an Ubuntu 18.04.5 ISO as dual boot. After installation and updates, the key hardware information is as follows:

I then set the CPU Scaling Governor to ‘performance’ and ran some Linux benchmarks:

and also tested the Blender ‘BMW’ benchmark for both CPU and GPU (CUDA):

For the same set of Phoronix Test Suite tests the results were:

The CPU throttling observed in Windows also occurs in Ubuntu. This can be seen clearly when repeating the Blender ‘BMW’ benchmark for CPU whilst monitoring CPU utilization, maximum frequency, and package temperature:

4K Video Playback

I tested 4K video playback in Edge, Chrome and Kodi on Windows and in Firefox, Chrome and Kodi on Ubuntu. 4K at 30 FPS presented no problem however 4K at 60 FPS when played in browsers on Windows resulted in the occasional dropped frame:

whereas this wasn’t an issue on Ubuntu:

Although hardware acceleration is not supported on Ubuntu for NVIDIA graphics when decoding VP9 and 10-bit HEVC (H.265) videos resulting in software decoding being used the playback was still flawless:

Thermals

During benchmarking the room temperature was around 24°C and the CPU temperatures peaked at 69°C when running Blender on Windows. This resulted in the temperature of the surface of the exterior heat sink measuring 59°C making it too hot to touch, and it’s actually normal for a fanless metal enclosure, as the heat should dissipate through the enclosure instead of being trapped inside. As the device was passively cooled it takes time for both the CPU and the heat sink to cool down.

To illustrate this I ran ‘stress’ in Ubuntu which shows the CPUs immediately reaching 60°C and then climb to a maximum average of 66°C whilst the room temperature was 23°C:

The temperature of the surface of the heat sink taken immediately after the test was 57°C. Twenty minutes later the CPUs had dropped to an average of 52°C (with one CPU at 58°C) and the heat sink had dropped to 50°C:

Installing the optional external fan kit resulted in the fans constantly running and consequently quite noisy at 50 dBA although effective as the new baseline average temperature for the CPUs became 32°C.

Running the ‘stress’ test again saw the CPUs temperature immediately reach 40°C and then climb to an average of 44°C:

After the test finished the CPUs dropped quickly to 37°C before returning back to 32°C:

The temperature of the heat sink was not measured due to be covered by the fan assembly.

The ‘stress’ test was run whilst monitoring CPU utilization, maximum frequency and package temperature first with the external fan:

and then repeated without the fan:

again showing that the CPU throttling is related to power and not temperature.

Windows vs Ubuntu on Cincoze GM-1000

Whilst a detailed comparison between the two operating systems is beyond the scope of this review, it is worth noting some of the key findings I observed. First looking at the performance tools common between the two systems. Overall Ubuntu performs slightly better in the majority of the benchmarks than Windows and this can be visually shown by comparing the same Blender benchmark in each OS:

Interestingly the idle CPU frequency is higher in Windows than in Ubuntu however this is reversed when the CPUs are put under load.

Networking

Network connectivity throughput was measured for the rear 1GB Ethernet port on Ubuntu using ‘iperf’. Upload was measured at 934 Mbits/sec and download at 908 Mbits/sec.

Power Consumption

Power consumption without the eternal fans installed was measured as follows:

• Powered off (shutdown) – 3.4W (Windows) and 3.5W (Ubuntu)
• BIOS*  – 44.8W
• GRUB boot menu – 46.1W
• Idle – 27.9W (Windows) and 27.5W (Ubuntu)
• CPU – 103.2W then 53.4W (Windows ‘cinebench’) and 71.6W then 56.4W (Ubuntu ‘stress’)
• 4K 60 FPS videos** – 58.7W (Windows Edge) and 62.2/74.3W (Ubuntu Firefox/Chrome)

*BIOS (see below)

**The power figures fluctuate so the value is the average of the median high and median low power readings. The 4K 60 FPS video power draw in Ubuntu Chrome was higher than other browsers.

BIOS

The BIOS is quite unrestricted and covered in detail in the user manual.

Final Observations

Whilst the GM-1000 embedded GPU computer’s passive performance is good the key highlights of the device are the configurability, the expansion capabilities, and the high-quality documentation. You can check US pricing and options on OnLogic website.