Building a personal workstation: Putting together the pieces

Article By : Brian Dipert

Two months ago, I told you about (almost) all the components I planned to use in building a personal workstation, with the notable exception of what CPU I would include.

Last month’s follow-up piece resolved that all-important omission:

And now it’s time to tackle the build! Before continuing, I feel compelled to make a few community-thank-you mentions. Whenever I tackle a teardown, for example, I always first hit up Google to see if anyone else has already done the same (or a similar) project…not because I want to steal from their work (you may have already noticed, in fact, that I regularly include reference links to others’ writeups and videos) but because I want to learn and benefit from following in their footsteps. If someone destroyed a device in part or in full due to unwise-in-retrospect disassembly techniques, for example, I’ll make sure to try a different approach instead. Conversely, if someone’s gotten a device apart fully intact, I’m likely to emulate them.

The same goes for a project like this, which I guess you could consider a reverse teardown. To wit, here are some links to content I found especially helpful:

• Enclosure review
• CPU installation instructions (video)
• AIO installation instructions (video)

The last (but definitely not least) recommendation is a series of videos from a guy name Gill Boyd, who manages the BuildOrBuy YouTube channel. A couple of years ago, he built largely the same system that I’m targeting; the identical GIGABYTE TRX40 Designare motherboard model (albeit one revision older, the original v1.0 vs my v1.1), for example, and the same CPU family (albeit using the 32-core AMD Ryzen Threadripper 3970x versus my 24-core Threadripper 3960x). He did a series of videos on the assembly, debugging, results and subsequent upgrades; search his channel using keywords such as “Threadripper” to find them. They’re long and detailed, but also both informative and entertaining, and therefore IMHO well worth your viewing time. By the way, he also did a video on the ASUS ProArt B550-Creator motherboard, which long-time readers may remember is also on my to-tackle pile.

Speaking of motherboards, here it is in real life, versus the previous writeups’ “stock” photos:

Underneath it is the Thunderbolt 3 add-in card, along with documentation:

And underneath that is the M.2 SSD RAID add-in card, along with an abundance of cabling:

Here’s the NZXT AIO (all-in-one liquid cooler) complete with three 140 mm fans, straight out of the box, with the Intel LGA1200 mounting bracket (which isn’t going to do us much good in this particular project!) pre-installed:

AIO and fans installed (but hold that thought):

And here’s one of the three 140 mm fans intended for the bottom of the case (I didn’t use the included Molex adapter cable, relying on the integrated connector to the motherboard for fan speed control; note that I did not select PWM-supportive fans since less-expensive standard ones can still be speed-modulated by the system in response to temperature sensor readings):

As I was starting to install these latter fans, I remembered that I’d need to comprehend intake-vs-exhaust direction. The general plan for this system is to have airflow entering the case from the front/side (the AIO) and the bottom (the aforementioned fans) and exiting out the rear (the combination of another 140 mm fan and passive vents). There are two reliable ways of determining the intake-versus-exhaust orientation of a fan. Sometimes airflow direction is marked on the side(s), as with these Rosewill fans:

Otherwise, you can identify the exhaust “end” by noting where the motor is located, along with the additional frame structures that hold it in place:

Note in this picture that by the time I took it, I’d already also figured out that my initial installation of the NZXT fans for the AIO was backwards, and had reversed them as well:

And last, but not least, here’s the final (and correctly oriented from the get-go) 140 mm fan to send hot air out the back of the case:

My normal practice is to populate motherboards as much as possible (with the CPU, DRAM modules and SSDs, for example) while it’s evenly supported underneath, prior to installing it in the case (where its support points are restricted to a handful of standoffs, a situation begging for cracks in response to any notable downward pressure). The foam assemblage surrounding the TRX40 Designare out-of-box worked perfectly:

DRAM first:

Next, the CPU (again, note the presence of incriminating thermal paste remnants on this supposed “open box” chip package’s integrated heat spreader):

I’m not going to step-by-step describe the socketing process shown in the images that follow; instead, if you’re interested, here (this time embedded) is the tutorial video I linked to earlier:

That said, one point I would like to emphasize is that, like modern Intel CPUs but unlike AMD’s mainstream Ryzen CPUs at least through the AM4 socket generation, the Threadripper package is leadless (LGA). The “pins” are instead on the socket itself. Here’s another: to tighten the socket lid over the CPU, you could use a standard hex driver. The one that came with the bracket (which I’ll mention next) is preferable, however, because it’s “ratcheted”; once proper tightness is achieved, further torque is prevented…that’s pretty slick!

Without further ado:

With the CPU in place, it’s time to mount the AIO block on top of it. As previously mentioned, the NZXT AIO comes out-of-box with an Intel LGA1200 mounting bracket attached to the block (which, note, comes with thermal paste pre-applied):

The bracket’s bayonet mount makes it easy to replace it with the sTRX4 one I got standalone on eBay (as I mentioned last month, my CPU wasn’t “retail”, so the bracket wasn’t included):

Again, here’s an embed of the previously mentioned tutorial video for all the details:

Next, how about those M.2 SSDs? The TRX40 Designare supports installing up to four of them, all supporting up-to-PCIe 4.0 speeds. Two connect directly to the CPU:

• Socket 3, M key, type 2260/2280/22110 SATA and PCIe 4.0 x4/x2 SSD support
• Socket 3, M key, type 2280 SATA and PCIe 4.0 x4/x2 SSD support

The other two leverage the chipset as the CPU intermediary:

• Socket 3, M key, type 2280/22110 SATA and PCIe 4.0 x4/x2 SSD support
• Socket 3, M key, type 2280 PCIe 4.0 x4/x2 SSD support

For performance-optimization reasons, I went with the first two (the ones on the left in the following image):

each of which included a dedicated motherboard-mounted heat sink:

For the shorter M.2 2280-max slot (right), the heatsink mounting screw and SSD mounting screw are one and the same. For the longer M.2 22110 slot (left), I leveraged separate heatsink and SSD mounting screws for the shorter M.2 2280 SSD I installed there:

Finally, it was time to drop the motherboard into the enclosure. Before doing so, I snapped a photo of the motherboard backside:

And here we go:

Even though this was my first time installing an E-ATX form factor motherboard, I still knew how many mounting holes (corresponding to case-installed standoffs underneath the motherboard) I should have, but I kept coming up two holes short (or, said another way, with two screws left over). After a few minutes of head-scratching, I realized that the remaining two holes were underneath the shared heatsink for the (unused) two remaining M.2 SSD slots:

Next step: the GPU:

Unsurprisingly, the NVIDIA GeForce RTX 3080 Ti’s power demands far exceed the 75W supply capabilities of the PCIe slot standalone. Supplemental power, straight from the supply, is fed to the graphics board via a proprietary 12-pin connector on top. Along with the GPU, NVIDIA supplies a cable adapter that converts it to dual 8-pin standard PCIe power connectors:

Finally, speaking of power, it was time to hook up the wiring, sourced both from the power supply and from the case itself (power and reset switches, USB ports, audio connections, etc.). The AIO block was particularly complicated, due to the multiple connections required to power and control both the pump and other circuitry:

I also ended up swapping out one of the 140 mm fans at the bottom for a 120 mm alternative (after snapping the photo, I even decided to swap that for an all-black 120 mm one so the colors would match), so that I could strain-free plug the case’s USB-C cable into the motherboard:

Here’s what the process and end result look like from the backside (for now, I’m only using one of the multiple PCIe power output connections supported by the power supply, for the GPU, since I haven’t yet added either the Thunderbolt 3 card or SSD RAID card):

And the front:

The result is (for now) intentionally un-tidy; this initial step is intended solely as a “smoke test” to make sure everything works, before I bother taking time to clean up the wiring routing (that said, I did make sure that no wires would get stuck in fan blades). I flipped the switch on the power supply, pressed the power button on the case, and…nothing. Well, not nothing. The fans built into the power supply and graphics card, along with the standalone ones at bottom, back and attached to the AIO radiator, all spun. And the motherboard status LEDs, standalone and multi-LED code cluster, also lit up. But what they told me was disconcerting, the CPU was DOA:

Trust me. I tried pretty much everything (save for attempting a CPU-not-required Q-Flash Plus BIOS update, which shouldn’t have been necessary to at least get to the BIOS setup screen):

• Double-checking every electrical connection
• Using a power supply tester to confirm valid voltages on all outputs (with the qualifier that since these widgets don’t test under load, they can’t confirm within-spec current)
• Re-socketing the CPU and re-installing the AIO block
• Etc.

I even first manually reset the CMOS RAM, then completely replaced the CMOS battery, as online research had suggested that this (a failing battery) sometimes was the culprit. Nothing did the trick. Thankfully (and intentionally), I’d tackled the build shortly before the 30-day eBay Buyer Protection period expired, so although the seller had claimed “no returns” on his post, he ended up sending us a postage-paid international shipping label. It took more than two months for the CPU to clear customs in China, and even then, the seller sat on the return until eBay stepped in and interceded on our behalf…but we finally got all of our money back.

This situation, along with other PC builds I’ve subsequently done (which I’ll tell you more about in future posts) is an effective reminder of just how complicated PCs and other electronics systems are; how even the most minute flaw in just one link in the operating chain is enough to turn the whole assemblage into something no more useful than a doorstop. And speaking of further builds, what are my plans for the refunded funds? Nearing 2,000 words, I’ll save further discussion here for the future. For now, I welcome your insights in the comments!

 

This article was originally published on EDN.

Brian Dipert is the Editor-in-Chief of the Edge AI and Vision Alliance, and a Senior Analyst at BDTI and Editor-in-Chief of InsideDSP, the company’s online newsletter.