Building A Silent Linux Desktop For 2022 With The Streacom DB4

A photo of the DB4.

Conclusions

It is hard to overstate how much of an improvement a good fit of the heatpipe mounts alongside these pastes makes on thermal performance.

In the first PTS run, with pads, the LH6 concession and perhaps suboptimal fit, I would see temperatures on the die reach 95 °C within an hour and stay there for prolonged periods of time to only drop below 95 °C every so often, when a less strenuous test was running.

With the latest improvements, this behavior seems to be completely inverted. The CPU die temperature hardly ever hits the temperature ceiling during the ~8 hour run. And when it does, it is mostly briefly.

The DB4 is a wonderful case and eventhough I felt a bit silly because of the mistakes I made, in the end, my feelings of accomplishment far outweigh any other. I am very happy with the outcome.

I think building an efficient, fanless, present-day desktop computer is viable and absolutely worth pursuing if you like the idea of eliminating all moving parts from your computer and would like to enjoy an inherently silent, more efficient computer that does not actively collect dust, the DB4 makes for a great platform for building such a machine.

Passively cooled computers work best whenever their use is more typical in nature rather than a full continuous load. If you are looking for a case for mining, you will probably want to look elsewhere. However a developer or a creative user will find a appealing, powerful, distraction free tool in a computer like this.

Streacom, can however, stil improve upon this case design by resolving the LH6 fit-issue by making it easy to achieve a guaranteed flush and tight fit. Perhaps the engineers at Streacom can find a solution that suits this case for achieving such a perfect fit.

The case comes with almost everything you need. Almost, because I think you do need isopropyl alcohol and some more thermal paste. Having five or ten grams of thermal paste in reserve is not a luxury.

But wait.. there’s more. I have performed some wall power measurements, have a few words of gratitude and some blatant self promotion before we continue on to the icing on the cake: he benchmarks.

A transformer box with a DB4 on top.

Wall power consumption

The Alecto EM-16 v.3 power meter used in the 2017 article was fairly uncertain. Its margin for error was rated at +/- 3% or +/- 2W, which makes it unsuitable to measure small power consumption, such as with devices on ‘stand-by’. My father sent me a ‘Brennenstuhl 231 E’, which is a big improvement in this regard, its uncertainty is rated at +/- 1% or +/- 0.2W.

For reference, I redid the measurements on the Intel I7-7700T.

[2]: Firefox 95.0.1 playing full screen the ‘Cross Bones Style’ performed by ‘Cat Power’ video at 1080p. [3]: ‘smallpt 20000’, once read at start and once at 67% completed. ‘smallpt 5000’ on the i7. Values are approximate due to variation. [4]: Building the open-source Rust rendition of a Spotify GUI client. cargo clean && cargo b --release, note that this varies wildly. May be read as ‘mostly between M and upperbound N’. [5]: stress-ng --cpu 8 --io 4 --vm 2 --vm-bytes 128M --timeout 60s

The most important conclusion here is that the I7’s power supply lacks power. It is too often too close to its rated maximum of 84 Watt to be comfortable. Being an external power supply makes that a bit easier to resolve. Let’s not panic either, it compiled many projects over the past four and a half years without a hitch. Let’s hope I can find a suitable replacement around 100-120 Watt.

I suspect that most computers are idling most of the time. I would applaud it if CPUs would scale down clocks even further during idle time. I feel there is room for further improvement when idling at login costs 24 W. I may try to undervolt the 5700G slightly. If you would like to see how that works out, with respect to power consumption and performance, just let me know in the article’ s comments.

Thank you

Coen Meijer for joining me for a photoshoot. Siem Schipper for putting up with me bemoaning thermal paste on my hands, your words of encouragement and lending me your eyes on occasion. Harm and Mat van der Duim for sending the ‘Brennenstuhl’ power meter. Also, thank you Michael Larabel for helping me out getting the Phoronix Test Suite to finally spit out ‘performance per Watt’ metrics.

Last but not least thanks goes out to Streacom, to Ahmet and his colleagues, who understood there was a need for an article on building a present-day passively cooled desktop computer using the DB4, Thank you for answering my questions, for the DB4 and LH6-kit and thermal paste later on.

And thank you Phoronix reader!

Shameless plug

I am looking for a job! - remote would be preferable. If your organization is on the lookout for someone who can write (advanced beginner / competent) Rust and is a reasonably skilled writer, then please contact me - or hand your colleague at human resources a link to this article and hint them that they and I should get acquainted. It would mean the world to me.

Thanks in advance!

occupy dot luuk AT freedom dot nl

Benchmarks

DB4 benched.

In good Phoronix tradition, I kept the best for last.

I chose to repeat the test suite run Michael used in his August 2021 review of the 5700G. This allows us to see how and if having a passive solution can hold up against an actively cooled 5700G.

Michael's testing was with a Noctua NH-U9S heatsink and with a Roseweill 4U enclosure. There's different hardware at play, but ere is a look:

Broadly speaking, this DB4 configuration performs a little better, likely due to its wider memory bandwidth / updated system firmware / newer Ubuntu release / similar factors, but differences are mostly small. See all the raw benchmarks and individual power numbers in full via this OpenBenchmarking.org result page.

Enjoy.

Thank you for reading - and see you in the comments..

Thanks again to Luuk van der Duim for this guest contribution article to Phoronix. Those interested can find the DB4 at Amazon (affliate link) among other Internet retailers.