Building A Silent Linux Desktop For 2022 With The Streacom DB4

It is quite an accessible case. You can and should remove all four side panels while you are working on the case. This allows you to orient it in any position you need to easily get you where you want to be. That is a feature I came to value a lot and used extensively.

A view on the main heatpipe mount without the panel on.

In the end, by far the most challenging part of building the DB4 with LH6 kit is getting the LH6 mount plates to tightly fit and sit flush with the outer panel. Whereas the main heatpipe mount plate and pipes can swivel itself tight against the outer panel, the LH6 plates lacks the provisioning for a perfect fit. The user guide allows you to choose between using thermal pads and thermal paste. Pads are less thermally conductive when compared to thermal pastes, however pads do a better job of bridging the gap between the mount and the outside panel. Therefore, on my first tries, I used pads on the LH6 mount plate(s). This is relevant later, in the paragraph on the ‘blue elephant’-plots. From the same plots later on we will also learn on the importance of a good, tight fit.

Inside the case, heatpipe mount with pad hidden from sight.

Making sense of ‘Blue elephant’-plots and one ‘Benetton Herd’

The first build used only a single LH6 mount plate. I had tried to employ both but getting the plates and pipes to fit between the posts of the DB4 frame is difficult if the LH6 L-shaped pipes are fitted the wrong way around, with the long end on the mount’s side. - Who would do such a thing? - There was a pad the LH6 panel mount and paste on the main mount plate.

When I ran an ~8 hour ‘Phoronix Test Suite’-run, at four hours in, it halted during the ‘DeepSpeech’ test. The diagnostic messagess (dmesg) reported a ‘page table corruption’ in the ‘DeepSpeech’ process’ address space. Oddly I had never encountered a page table corruption before and this being a new build, I grew apprehensive again.

I contacted Streacom and Michael to inform them of what had transpired. Michael suspected the XMP profile might be the culprit and Ahmet of Streacom advised me to apply paste instead of using the pads.

Page table corruptions may be caused by faulty memory. It is hard to exclude faulty memory as a cause, but proving it unlikely is easier. I ran ‘memtest86’ for four passes and it passed with without errors.

Memtest passed fine.

At this point in time I wanted to come to better understand what happened during load so I wrote sense.

Sense is a small and fairly simple tool to help get a better understanding of how the case behaves as it heats up and cools down during and over time. I wrote it in Rust. It samples CPU-die temperature, power- and average frequency values and on exit, it will save plots and raw CSV-data. As of writing, it depends on the zenpower3 module, because the 5.13 kernel shipped with Ubuntu 21.10 has incomplete ‘Zen 3’ sensor support.

A hot run as seen by sense.

Sense generates dual y-axes plots to allow both power and temperature data to be shown in a single plot. The blue temperature plot line under these load conditions tends to resemble the outline of an elephant. With the cooling interval as its ‘identifying mark’: its trunk.

Meanwhile Ahmet of had sent some of Streacom’s new TX13 thermal paste!

TX13

Streacom has a new paste in a new package. They named it TX13 TX13, The paste without the waste, comes in small, single application sachets. The idea is to save on plastic waste by cutting the syringe out of the equation. Streacom pledges to offset negative environmental impact of using thermal pastes by planting trees through the Eden reforestation project.

After “interrogating” the people at Streacom on the ingredients and performance of the paste their product manager could tell me no more than that TX13 was made with a conventional, non-corrosive recipe but he would say their internal testing suggested that TX13’s performance was on-par with the ‘Kryonaut’ paste I had on reserve.

Before I could look into what these performance claims meant in real-world, I had to set my build straight.

improving the build

I rearranged the heatpipes to match the manual’s instructions. I applied TX13 on the CPU and on the copper shim. I learned that for the CPU’s IHS you really need only a single sachet. I applied a few sachets in the aluminium mount that encloses the copper shim and connects the shim to the heatpipes. Here I found the practicality of the sachets a little lacking because of the relatively larger and uneven surface area that needs to be covered. The remaining TX13, I applied on the outside of the mount plates. I also applied the Kryonaut I had on reserve to cover the remainder of these plates. With TX13, the pea sized portions are ideal to cover a typical CPU or GPU IHS. However, larger areas will require you to empty multiple sachets worth of TX13 and there I find the syringe more practical to determine a suitable dose for application.

In the image below, the left column of plots are of the first build, with a pad on a single LH6 mount and paste on the main heatpipe mount. On the right you will find the ‘penultimate build’, which has excellent pastes applied all around.

Pads vs paste plots - when it does not matter.

The temperature in both conditions, will eventually hit the CPU’s temperature ceiling of 95 °C. On each subsequent run below it wil hit the ceiling sooner and run at that threshold temperature for longer and the CPU will throttle to counter and prevent overheating. Notice that cooling down takes longer on each subsequent run. Longer trunks are thus worse trunks.

This poor CPU took a number of beatings. I would like to applaud AMD for making the 5700G as resilient as it turns out to be.

The Benetton Herd Plot

Needless to say, I was rather disappointed to hardly find any difference between the both columns. But I had a sneaking suspicion that this was a case of ‘bad fit’. The case was reopened, again. To help improve backpressure on the heatpipe mounts, I used a washer and the thermal pads I had left. The difference was markably noticeable as on closing the panel I could feel the mounts’ firm pressure. The main mount got a similar treatment and I pressed it firmly into its final position.

It was time for another smallpt run.

DB4 performs better.

Now we are talking. The CPU is thermally more unified with the outer panels, which results in much more resistance to heating up. This is illustrated by the markably less steep temperature curve. This run never even came close to 95 °C! Note how sharp the temperature drop is. A cool trunk!

I gathered a few more cold runs I had, and together wih this last run I created the following plot:

Multiple cold runs on the DB4.

The blue run has TX13 and ‘Arctic MX5’ as pastes, MX13 on the CPU and shim, Arctic MX5 on the heatpipe-shim terminal and on the main heatpipe terminal plate. It has thermal pads on the LH6 terminals. The red, yellow and green conditions all are of the latest build effort: TX13 on all surfaces with additional Kryonaut applied to make up for a shortage on the outer heatpipe mounts.

The green line is the much improved fit over the red and yellow runs.