When Ethereum exploded in popularity (and value) in Summer of 2017, I
decided to educate myself about cryptocurrencies and blockchain technology. As
part of that process, I built an Ethereum (Classic) mining rig.
Here I’ll discuss how I designed and built my miner, focusing primarily on the
construction of the chassis using OpenBeam, Fusion 360, and
a Shapeoko 3. (Later, I’ll discuss BIOS configuration in Part 2,
mining strategy optimization in Part 3, and compute performance
optimization in Part 4.)
My strategy for the build was first to choose the mining hardware (the “guts”
of the machine), and then design the chassis around it. Having never built a
miner before, I began by researching what hardware was typically used for
The lay-flat solution seemed practical and inexpensive, but I had a few issues
Spatial constraints in my apartment forced me to store my miner on my desk,
and I was unwilling to sacrifice the amount of desk-space that the lay-flat
design would require.
The lay-flat design places all GPUs side-by-side, such that only the
outermost GPU has access to cold air - the others’ intake fans draw directly
from their neighbor’s heat-sink. I feared that this arrangement could
contribute to overheating.
Subjectively, I thought many of the lay-flat builds were ugly.
Given the above, I decided to eschew convention and design a vertical miner
that resembled a common PC tower.
With a strategy in mind, I began to think through the design details. I
eventually decided that:
The power supply should be mounted at the bottom of the chassis to provide a
low center-of-gravity and minimize the risk of tipping.
Six total GPUs should be positioned above the power supply, in pairs, in
three 150mm bays. This arrangement would give 3 of 6 GPUs access to cold air,
and would sandwich no GPUs.
The motherboard should be mounted off the side of the chassis, on the side
opposite the GPU intake fans (such as not to obstruct cold air intake).
Satisfied with the plan, I began to research construction materials, hoping to
find an attractive material that facilitated rapid prototyping. I eventually
discovered OpenBeam, which touted itself as “a low-cost, open-source
extruded aluminum construction system”:
I ordered an OpenBeam set, the “mundane” computer components (motherboard, CPU,
RAM, etc.), and a single GPU, and then began prototyping the chassis design.
OpenBeam was ideal for this application. While I have some experience with
Sketchup and Fusion 360, I personally find it easier to reason through spatial
problems in “meatspace” when possible. OpenBeam spared me from having to model
GPUs and other components in CAD, and enabled me to iterate using real,
After a few days’ worth of experimentation, I settled upon the following
(You’ll see a 500W power supply in the photo. I was using it as a placeholder
while I waited for a 1200W unit to arrive in the mail.)
With the prototype assembled, I decided to sanity-check the miner’s hashrate
before purchasing the remaining GPUs. I installed Ubuntu 17.04 Server Edition
and configured Claymore to dual-mine ETC and DCR, and ran the miner
overnight. The hashrates were consistent with what I expected, so I continued
With the hardware proven, I purchased and installed the remaining five GPUs:
(These pictures were taken before the final two GPUs arrived in the mail.)
Initially, nvidia-smi could not detect two of the new GPUs, and I suspected
that the PCI-E risers were defective. After experimentation, however, I
determined that the risers themselves were fine, but that the USB cables they
shipped with were faulty.
I resolved the problem by replacing all of the PCI-E riser cables with new ones
from Amazon, having lost confidence in the factory cables.
I also had to update my motherboard’s BIOS in order to detect more than four
GPUs. I’ll discuss that process in detail in Part 2.
While I mulled over how to permanently mount the fans, I decided to first solve
an easier problem: mounting the power switch and hard-drives.
Wanting to make the job easier, I started by purchasing a double mounting
bracket for the hard-drives:
I then designed a panel for the top of the chassis, and cut it from 1/4” Lexan:
I attached the power switch and hard-drive mounting bracket to the panel, and
then attached the panel to the chassis. It held everything neatly in place:
The mar in the photo occurred because my mill failed to withdraw to its
“retract height” when travelling to make its first cut. I’m not sure why it did
that, but I suspect that I overlooked some parameter in Fusion 360 that I
should have set. (I’m really enjoying Fusion 360 so far, but I’ve found that it
has a steep learning-curve.)
Milling blemish aside, I thought the panel turned out really well.
With the drives mounted, I turned my attention back to the fans.
I wanted to position a fan in front of each GPU bay. Each bay was 150mm wide,
and each fan was 140mm wide (with mounting holes slightly closer together than
Because I was happy with the Lexan mounting panel for the hard-drives and power
switch, my first instinct was to take a similar approach with the fans. I
designed and cut a prototype mounting panel:
The panel milled perfectly. After experimenting with positioning it, though, I
observed some problems:
The holes that mounted the fan to the panel and the holes that mounted the
panel to the chassis were so close together (out of necessity) that their
screws competed for space.
Each GPU bay had somewhat different geometry due to the hardware used to
mount the motherboard to the side of the chassis. Accommodating this would
require me to either mill a distinct panel for each bay, or to “jump” some
of the mounting hardware with stand-offs.
After staring at the chassis for a bit, I decided to ditch the Lexan panel and
opt for a simpler solution - mount the fans directly to a chassis rail:
(The fans are white, though they look somewhat blue in the photos.)
(The blue device sitting on top of the chassis is a fan controller. It hadn’t
been mounted at this point.)
This approach positioned the fans slightly (5mm) off-center, which is why I
avoided it initially. Functionally, though, that was inconsequential, and I
found that the aesthetics didn’t bother me. So, I declared that solution “good
enough” and moved on.
The final piece of hardware to mount was the fan controller:
I had initially hoped to mount it above the topmost GPU fan, but the enclosure was
too deep, and collided with the power switch.
I then changed plans and decided to mount it off the side of the chassis, above
the motherboard. There existed no convenient way to mount the fan controller
enclosure to the chassis, though, so I disassembled the fan controller to see
if I could drill two holes in its enclosure to use as mount points:
The enclosure turned out to be mostly empty. The useful part of the fan
controller was a circuit board attached to the front panel, which lifted
effortlessly out of the enclosure:
Having discovered this, I discarded the enclosure and returned to my original
plan of mounting the fan controller above the topmost GPU fan. I only had to
make one modification to the chassis: the fan controller collided with a
horizontal strut that connected the tops of the front chassis rails.
The offending strut was structurally unimportant, so I simply removed it.
(Thanks to OpenBeam, it only took a few seconds to make this change.)
Next, I once again used Fusion 360 and my Shapeoko to design and cut a Lexan
mounting panel for the fan controller:
The fan controller snapped cleanly into the mounting panel, and I reinforced
its fit with Superglue. (In hindsight, I should have been tidier with the
glue.) I then mounted the panel to the chassis:
With the fan controller installed, I powered on the machine and tested the fan
acoustics. Dependent on the fan controller settings, the fan noise ranged from
“barely audible” to “oscillating desk fan”. Even at the highest setting, the
fan noise never became unpleasant.