Building A Silent Linux Desktop For 2022 With The Streacom DB4
On the article
Even though I left out many topics that seemed relevant or important earlier on as I worked on the article, it does still touch on multiple topics and we will be going off on a tangent here and there, just because it is noteworthy or I found it too peculiar to withhold it from you. The overarching question this article aims to answer is if building a passively cooled desktop computer is still something worth pursuing and the second question is how does the DB4 fare as the platform for a modern day passively cooled computer.
I mainly wrote the article because I consider myself a computer technology enthusiast and I like to tinker with both text and technology. However..
Beware of the biased author
In April 2020 I found that someone offered a second hand ZF240 power supply and bought it. Provided that it was in working order, a lucky find.
It was not until October 2021 until it occurred to me that I could write an article for the Phoronix readership again, this time on building and testing a passively cooled computer, starring the Streacom DB4. I decided to write Streacom. To my surprise and delight they welcomed the idea and sent me a ‘titanium’ DB4 - the very one you see in the photos - and an LH6 heatpipe set to help me along with my build and article.
So here you have it: Streacom provided review samples of the DB4 case, the LH6 kit, some extra TX-13 thermal paste and some good advice.
DB4 with wooden background.
The innards
Building a passively cooled system, implies being constrained on what components you use for the build in terms of heat generation. You have a strict ‘TDP budget’ to consider, more so than with an actively cooled case where you could swap in a better cooler if necessary, you cannot change the properties of the case. The power supply, the ZF-240, can supply 240 Watt, so whichever components I would choose, combined these should never exceed 240 W, preferably a lot less. I also decided on a - ‘no fans at all’ - purist approach. That reduced the number of viable motherboards that have an AM4 socket, are of mini-ITX size and lacked a fan to only two, if I counted correctly.
Red or blue? Pick a brand. What to tap into?
CPU
Choosing the CPU was not that hard. Actually I feel it should have been harder.
Even though Apple has some really appealing, efficient and powerful CPUs these days, one cannot buy these without buying an Apple computer, can you? Which, to me seems a bit wasteful. Even though converting an Apple computer into a DB4pple might be an interesting idea for someone to try one day, this is not that day and I am not that one. That being said, there are interesting things happening outside of the x86-space. I will keep a close eye on future CPUs based on RISC-V designs, but for now these do not offer the kind of performance I would like my day-to-day desktop computer to have.
Realistically, it was either going to be Intels 12th generation, Alder Lake, or an AMD ‘Zen 3’ APU. The latter would have faster iGPU graphics and the former promised more cores and - maybe - better performance. The problem was that, halfway november 2021, most of Intel Alder Lake’s line-up was somewhere over the horizon. There was little known on when the non-K parts would be availabe to the general public and frankly I also found the choice between DDR5 and DDR4 hard if I were to go with Alder Lake. Even though I did not pick Intel this time, I do love the direction they are taking as of recently.
I chose the AMD Ryzen 7 5700G. From Michael’s earlier Phoronix review I knew what to expect in terms of power consumption and performance. I had good hopes for it to be a practical, reasonably efficient, fast and versatile choice.
Memory
Some sources point out that Ryzen 5000 series benefit from memory with higher bandwidth. The second takeaway from reading up on ‘Zen 3’ RAM considerations is that having dually ranked memory is preferable over singly ranked memory modules. As being concerned with memory rankedness was novel to me, I had to look up what it was, and why it was that it would matter. According to the Wikipedia entry on ‘memory rank’](https://en.wikipedia.org/wiki/Memory_rank) : >A memory rank is a set of DRAM chips connected to the same chip select, which are therefore accessed simultaneously.
And you may find one or more of such ranks on a single module!
According to the same source, an efficiency benefit of multi-rank modules over their lesser ranked siblings may be explained by: >Multi-rank modules allow several open DRAM pages (row) in each rank (typically eight pages per rank). This increases the possibility of getting a hit on an already open row address. The performance gain that can be achieved is highly dependent on the application and the memory controller’s ability to take advantage of open pages.
I imagine how a process with many heap allocations (with a predominantly random access pattern) may benefit from increased odds on getting a page hit on an already open page.
I fell for it. This sounds convincing to me. I just could not bear the thought of holding back my 5700G. So I perhaps spent a little bit more than added real-world performance can justify. I took a kit that needs a relatively low voltage for its XMP profile at 1.4 V. at 4000MT.
I bought a 32 Gb kit of GSkill Trident Z Neo F4 4000 C18.
Dear memory module manufacturers, would you please be so kind to mention the rankedness of the modules in their specification sheets from now on forward? Thank you in advance.
The Motherboard
I bought the Gigabyte B550i AORUS PRO AX rev 1.0 mostly because it is fanless and because I found that its most reported problem, ‘disappearing Bluetooth’, is currently considered solved on Linux with late 2021 firmware - or by being lucky and receiving a v1.1 revision of the board.
The difference between those two revisions is that the v1.0 revision has an ’ Intel® Wi-Fi 6 AX200’ mini pci-e module on the board, while the v1.1 has a ’AMD Wi-Fi 6E RZ608 (MT7921K) module. The Intel one has been notorious for Bluetooth drop outs, where the OS would not see the device anymore at all.
I contacted two retailers to order a v1.1 revision. One said they had not seen a v1.1 revision come by and the other said they had no say in what they received from the European distributor. I received a v1.0 board.
SSD
These are fast and efficient with an average power consumption of 4.6 - and a maximum of 5.5W.
Having one of these strip-of-chewing gum sized, ultra thin and feather-light SSDs in the palm of my hand, I could not help but be in awe with just how far we have come since punch cards and all the spinning media that preceded this.
Building the DB4
I once knew the owner of a particularly rough bar. The owner’s voice was coarse and grainy. So much so that, when he spoke, I feared for the paint on the walls to come off spontaneously. He once said the best advertising he did, was to tell people the truth. ‘My bar is great, but it’s not for you. It’s too rough for your liking.’ Most people would then feel challenged and he would always see them drop by later, he said. Because the ‘advertisement’ had in no respect been overstated, it is to be suspected most left even before ordering their first drink.
The boxed cooling kits
With the DB4 case comes a full-colour user manual. It is obvious it received a lot of love. It shows in clear, detailed steps how to build your DB4. On page two it says: >It’s not rocket science, but our cases can be a little challenging to assemble at first because of the non traditional design and the materials used.
It is the truth. If you do not like the prospect of being challenged during the build, or are looking for a quick fix, you might want to ask someone more experienced to build it for you or look for some other case altogether. This is not for you.
If you, however, are somewhat meticulous and patient and preferably have built computers before, you will be just fine or rather, you will love this case.
I will be honest. I did find it challenging at times. I made a few mistakes along the way that, looking back, were unnecessary had I read the user guide better or had I not deviated from the instructions. Building the computer took me far longer than I had hoped. The mistakes made me have to backtrack and do things over later on to set things straight.
It appears the wires are in.
Even though I do not like to make excuses. However, there are factors that explain why it is safe to assume that most others will find building the DB4 - or building anything for that matter - a lot less challenging than I did. I am visually impaired. I see little detail in the centre of my field of view. To a lesser extent, but a factor nonetheless, my right hand and arm are deformed in interesting but no helpful ways.
Building, readjusting and correcting errors became easier over time, as I came to understand the the hardware and the case better. Let me share a few things that I found difficult or where I made a mistake.
Both sides of the CPU backplate
I found it can be a challenge to get the tiny double sided adhesive pads on the M3 nuts without ruining the tiny pads. I failed on my first two attempts and that made me apprehensive. I had eight pads and I needed to be successful on at least four. I decided to be smart instead and repurpose the CPU-backplate with builtin threaded holes found on the back of the Gigabyte motherboard. It had occupied the same holes in the motherboard as where the M3 nuts were supposed to go. It had had the standoffs for the CPU fan screwed on prior to removal. (see the motherboard photo above - it still has the standoffs in place.) The backplate seemed ideal as it protects the motherboard by more evenly distributing tensile forces on the motherboard when either a fan or - in our case - a heatpipe assembly is screwed into place.
A full heatpipe assembly, with LH6 kit
I would not understand this was a mistake until the motherboard had been screwed to the tray, the tray screwed into the case, the shim and the CPU were pasted and the heatpipe assembly test fitted, ‘final fitted’ but failed to be screwed into place. The distance between thread and nut was too large I think. After giving it several tries, it dawned upon me. I got uneasy. I knew failing here meant having to go back to the task I knew I found hard but had to.
Tip!: There is a technique to get the the pads from the sheet they are punched into, onto the M3 nuts: First use a prying tool to carefully pry the centre out of the disc. Then remove the foil from one side of the disc. Note that the disc remains in the sheet. Then move the nut through the hole, have the adhesive end stick to the base of the nut’s head. Now the nut reinforces the fragile disc, you can now more easily get the disc from the sheet.
The DB4 without its side panels.