Beelink T45 Review with Windows and Linux, and Tweaking BIOS Power Limits

[Update December 23, 2019: We’ve been informed by Beelink that the T45 has been updated to the 6W Celeron N4200 processor instead, and the system is now cooled with a fan. The model reviewed in this post is the fanless version with a 10W Intel J4250 processor, which was never sold]

[Update January 30, 2020: We’ve now posted a review of the new model at Beelink Gemini T45 Pentium N4200 Mini PC Review]

Beelink have further extended their ‘Gemini’ range of mini PCs by adding the T45. This is a passively cooled mini PC that is effectively a companion to the J45 as it again uses the slightly older Apollo Lake Intel Pentium J4205 CPU which is a quad-core 4-thread 1.50 GHz processor boosting to 2.60 GHz with Intel’s HD Graphics 505.

Although the T45 is a ‘NUC’ style mini PC physically consisting of a 119 x 119 x 17.7 mm (4.69 x 4.69 x 0.70 inches) all-metal (and surprisingly quite heavy) rectangular case, it is just under half the thickness of the J45 and is very similar in size to the earlier AP34 and AP42 models from Beelink. The front panel has only a blue ‘power’ LED as it is the rear panel that includes the power button, power jack, dual HDMI ports, a gigabit Ethernet port, and a headphone jack. On the right side there are a couple of USB 3.0 ports and on the left side are a couple more USB 3.0 ports and a full-sized SD slot.

The full specifications include:

The T45 comes with a 2242 M.2 SSD (512GB in this reviewed device) with pre-installed Windows 10 Home (version 1903 OS build 18362.356) together with 8GB of soldered DDR3 RAM.

In the box you get a mounting bracket with screws for attaching the device to behind a monitor together with a couple of HDMI cables, a power adapter (not in shown the photo below), a manual and a service card:

Starting with a quick look at the hardware information shows it is aligned to the specification:

After fully updating Windows to version 1909 OS build 18363.535 I ran my standard set of benchmarking tools to look at performance under Windows:

and to compare with other Intel mini PCs:

The results need interpreting carefully otherwise they could be misleading because mini PC benchmarks are heavily influenced by the quantity and type of memory and storage installed.

However, when comparing the T45 with the similarly spec’d J45 the results are significantly lower and are more inline with a Celeron N3350/N3450 processor.

Additionally, I also tested with Geekbench 4 and got a single-core score of 1538, a multi-core score of 4011 and an OpenGL score of 5436:

Next, I shrunk the Windows partition and created a new 100GB partition so I could install and dual boot Ubuntu using an Ubuntu 18.04.3 ISO:

After installation and updates, a brief check showed everything working including Wi-Fi, Bluetooth, audio, and Ethernet. I then ran some basic commands to look at the hardware in more detail:

Interestingly the Ethernet driver is similar to the J45 in that the T45 uses the default r8169 module. However, it uses a different ‘mac version’ with the ‘rtl8168g-2_0.0.1’ firmware and therefore does not have the same problems as the J45 and works as expected.

I next ran my standard Phoronix Test Suite benchmarks and the results can be compared with those from the previous testing on other mini PCs:

As the full results might be a little confusing because for some tests, higher is better, whereas for others, lower is better, the following bar chart may be easier to understand:

Next I’ve run ‘sbc-bench’ which is a small set of different CPU performance tests focusing on server performance, ‘glmark2’ from the standard repositories which is a benchmark for OpenGL (ES) 2.0, some real-world timing tests for the compilation, zipping and unzipping of the Linux mainline v5.2 kernel, ‘iozone’ also from the standard repositories which is a filesystem benchmark tool and finally ‘Octane 2’ which is a JavaScript benchmark and was run in Chrome.

A summary of the results from each of the above benchmark tests was compared with previously tested mini PCs as follows:

The poor performance for the T45 that was seen in Windows when compared with the J45 can also be seen in the Linux benchmarks.

For a direct comparison with Windows I also tested Geekbench 4:

with the Ubuntu results being slightly better that those for Windows (1579 vs 1538 and 4282 vs 4011).

For real-world testing, I played videos in Edge, Chrome, and Kodi in Windows and in Firefox, Chrome, and Kodi in Ubuntu. The results were slightly worse than with other mini PCs and the following tables summarise the tests and results for each of web browsing, Kodi in general and Kodi playing specific videos:

I also noticed that both Edge and Chrome were slow to launch when using YouTube and felt laggy.

Finally, on both OS I also installed and ran the UNIGINE Heaven benchmark:

and the results show that the T45 will offer very limited gaming performance.

Network connectivity throughput was measured on Ubuntu using ‘iperf’:

Power consumption was measured as follows and is slightly higher than comparable mini PCs:

• Powered off – 0.7 Watts
• Shutdown – 0.7 Watts (Windows)
• BIOS*  – 11.8 Watts
• Boot menu – 11.3 Watts
• Idle – 11.5 Watts (Windows) and 9.1 Watts (Ubuntu)
• CPU stressed – 17.3 Watts (Ubuntu)
• Video playback** – 18.6 Watts (4K in Windows) and 17.6 Watts (1080p in Ubuntu)

* BIOS (see below)

** The power figures fluctuate so the value is the average of the median high and median low power readings.

Finally, the BIOS settings are partly visible and unlocked and a brief overview is available in the following video:

As previously mentioned the T45 is a fanless device relying solely on passive heat dissipation. It uses a massive heatsink that covers nearly the entire board:

Note that this heatsink also covers the M.2 SSD drive meaning that it will need to be removed and thermal paste reapplied if accessing the M.2 SSD for any reason.

There is little to no airflow possible over the heatsink so it, in turn, relies on the all-metal case to further dissipate the heat.

Whilst there is a warning on the underside of the case:

it does get hot. During benchmarking the maximum temperature I recorded on the top of the device was 53.5°C:

making the device unpleasant to handle when holding it to insert/remove USBs etc..

However, the cooling does seem to work as the maximum CPU core temperature recorded during testing was 68°C:

despite the fact that the ambient room temperature was 30.0°C as it was a hot day with the outside temperature reaching 42°C (107.6°F).

Under Ubuntu, I ran a VP9 video in Kodi for 20 minutes and the internal temperature whilst it climbed during the playback remained manageable and reached a peak of 65°C given the ambient room temperature had now cooled to 24.8°C and the device’s surface temperature was measured at 49.0°C:

(note the 15 skipped frames occurred when opening and taking the screenshot)

Because of the relatively poor performance when compared to the J45 I looked more closely at the BIOS. Under CPU Power Management Configuration for the T45 the ‘Max Package C State’ is set to ‘C0’ and ‘Power Limit 1 Power’ is set at 6 watts:

which can also be confirmed by HWiNFO:

This can be compared with the settings in the J45’s BIOS:

where the ‘Max Package C State’ is set to ‘PC2’ and ‘Power Limit 1 Power’ is set at 10 watts.

I decided to try just setting the T45’s ‘Power Limit 1 Power’ to 10 watts:

Performance immediately improved in the sample subset of benchmarks I ran:

with Passmark improving from 1000.2 (CPU mark of 1808.7) to 1308.7 (CPU mark of 2319.5), Cinebench R15 from 118 cb to 161 cb and Fire Strike improving from 296 to 498.

I also re-ran Octane which improved from 11951 to 12413 and Geekbench 4 where the single-core score stayed the same at 1538 but the multi-core score went from 4011 to 4648:

These improvements were also seen in real-world testing. A video that played but skipped occasionally in Kodi on Windows:

now played perfectly:

Interestingly there was no observable thermal implication as the device’s external temperature and internal temperatures remained similar to those seen previously. The maximum CPU core temperature for this short duration of testing was measured at 64°C:

whilst the ambient room temperature was 25.4°C and the device’s surface temperature was 49.8°C.

As a final check I also looked at the effect of changing the power limit whilst running Ubuntu.

Octane improved from 11333 to 13397 and the Geekbench 4 single-core score improved slightly from 1579 to 1596 with the multi-core score going from 4282 to 5020:

Compiling the Linux kernel improved by over 20% from

to

A video in Kodi that skipped continuously, dropped out occasionally and whose audio was out of sync with the image:

now played relatively fine with just the occasional frame skip:

And a Youtube video in Firefox that was unwatchable in 1440p@30fps resolution now played fine without stalling or dropping frames:

However whilst changing the power limit has improved the performance and makes the T45 comparable with the J45, the full extent of the thermal implications hasn’t been fully evaluated. Therefore although it is possible to mimic the performance of the J45, and certainly for a limited period, it may not be advisable to make the change permanent without constant monitoring. Either way, the device’s external case becomes hot when under load and whilst that may be a good thing for those enjoying a snowy winter it may be a deciding factor when considering its purchase.

I’d like to thank Beelink for providing the T45 for review. It currently retails at around $240 for the tested configuration from the various mini PC resellers on Amazon, GearBest, Aliexpress, or  GeekBuying among others.

Support CNX Software! Donate via cryptocurrencies, become a Patron on Patreon, or purchase goods on Amazon or Aliexpress