Question about PSU fan and side fans installation.
Question about PSU fan and side fans installation.
Hello everyone, this is my initial contribution. I've been observing for a while now and thought this would be an appropriate spot for asking my question.
I'm setting up a 2U server with a custom chassis—constructed from 3D-printed parts and a 2U rack mount shelf—for my homelab. While researching power supply choices, I decided to opt for a used standard ATX PSU. Unfortunately, the placement would block the fan since it would be against the top of the chassis. So, I considered drilling holes in the sides of the PSU case and installing 80mm fans, which is what I usually do when I want a hands-on approach.
I initially bought a Corsair "Shift" PSU with side-mounted connections, thinking it would allow airflow straight through the case. However, the seller sent me a regular RM850x model instead, which I didn’t have on hand, so I didn’t return it. Now that I have access to the unit, I’m considering how to arrange the fans properly.
From what I see, the PSU has clear plastic insulation along its sides. I’m wondering if this insulation is really necessary. It probably adds safety by making it harder for arcs to reach the case, but I’ve seen PSUs without it. I think I need to cut it, but I’m not sure where exactly.
The main decisions come down to two things: which parts are most likely to spark, and which need the most cooling.
I planned to place two 80mm fans on the left side of the PSU (as shown in the first picture) for intake from outside the server case. However, based on my understanding, the large capacitors are probably the main candidates for arcing. I could also put one fan in the top-left spot and cut back only the insulation around the capacitors—this would help airflow to the MOSFETs with heatsinks near the middle. This seems like a safer bet since the part I’d expose by cutting is the PWM board (which controls switching, not the fan). I might pair this with another fan on the rear exhaust vent (as shown in the second picture), though I’m not entirely sure about that option.
Another possibility is an intake fan on the opposite side of the chassis, drawing from within. I’d likely place it in the bottom-right corner, as it’s mostly free and open to airflow. The top-right spot might be blocked by the board where the original fan connects, so cutting insulation there could be safer. This setup could even create a swirling effect inside the case, helping cool everything down (as seen in the last photo).
I only bought two fans, but I’m open to adding more—maybe two on each side and one on the back exhaust. It might seem excessive, but it could provide enough airflow that I don’t need to cut any insulation at all, just the forced air over it.
There are so many options! What do you think is best based on these factors? I’m leaning toward a swirl configuration with an exhaust fan—combining the ideas from the second and third photos.
This post has turned out to be much longer than I expected. Thank you to everyone who takes the time to read it!
Welcome to the forums, newcomer!
First, is this plastic really needed?
It's there to stop an arc from forming when the metal shroud covers the PSU's internal parts.
Including the design for your case would be beneficial in two ways.
However, I’d ensure airflow remains unobstructed over the heatsink areas.
Currently, the fan placements show that most airflow is blocked by the capacitors.
Using 80mm fans in a modern PSU would mean we’re facing outdated efficiency issues from years ago.
What I meant by inefficient PSUs is looking at the temperature graph—how efficiency drops when temperatures rise.
Because clearly, I prefer tackling problems head-on.
If I were you, I’d choose an SFX PSU.
You’re just adding complications when a simpler solution would have worked better.
My two cents on the matter.
Having reviewed PSU designs in industrial and MILSPEC systems for Safety Critical parts, such as Creepage and Clearance distances, I find myself a bit puzzled by the pronounced high sides of this clear plastic sheet.
I imagine a PSU manufacturer might add an insulating layer beneath the PCB if their Quality Control isn’t stringent enough to guarantee all component leads are properly covered and don’t extend too far from the PCB toward the metal base plate. Maybe they include this insulation just in case wires detach from the board and come into contact with the interior of the enclosure. Unlikely, but who knows what the designer or engineers thought?
As you’ve noticed, not every PSU features these thick sides of insulating material. I haven’t seen this in any of the units I’ve examined before.
Regarding arcing, it generally requires about 3,000 Volts per millimetre to initiate in dry air. With the maximum voltage in an ATX PSU typically being around 340V DC (240V AC x √2), we’re far from reaching the 3kV needed to cause a jump of just 1mm. The gaps between components and the case sides are well over 1mm. That’s surprising.
I think it’s reasonable to say that a standard ATX PSU isn’t suitable for a slim 2RU server case. Have you thought about cutting a hole in the lid to improve airflow?
Personally, I’d be fine with trimming back some of the plastic side panels to allow better ventilation. Side fans might not provide perfect airflow, but if they generate enough circulation, it should work.
Ultimately, it’s your responsibility and that of your family—your safety is at stake.
I’d prefer purchasing a proper server PSU. Of course, new ones will be expensive, but the risk is much lower. Stay safe.
Can I provide an image of the case and the intended positions of its components?
Typically, the graphics card consumes the most power during heavy usage on most PCs.
Where will your power consumption originate? With an 850-watt PSU, much of that may not be utilized. In such cases, a PSU fan might not operate.
Airflow is necessary, but perhaps small openings could suffice even without fan support.
Thanks for your feedback everyone!
These heatsinks are definitely the top priority for me. Even with forced air around the capacitors, I believe they should work well since the air will move toward the back vent. Since this GPU can't monitor temperature changes (like my old AX1200i did), I might need to install a sensor somewhere to track temperatures and provide some reassurance.
That sounds like a solid solution. I probably didn’t do enough research because I wasn’t aware of such a form factor.
It gives me confidence. My logic was the same—only when there’s a serious risk would insulation become necessary.
I understand it isn't perfect, but I noticed others using flow-through style ATX PSUs (where air moves from front to back instead of coming in from the side). This kind of design is not very common for ATX PSUs; I only found one and it wasn't modular. I thought I could adapt mine to a flow-through setup. I considered adding a hole in the lid, but my rack is only 12U, and depending on what I put inside it, it might fill up quickly. So I didn’t want to waste space with a whole unit dedicated to airflow.
I believe the situation isn’t that serious since the PSU has an auto-shutdown if it overheats, which should keep it from reaching dangerous temperatures. Still, your idea inspired me to include a temperature sensor. I already have a Raspberry Pi and some sensors; even some relays are set up for remote GPU mining resets. I’ll look into using that to trigger a shutdown once a certain temperature is reached.
If you think this is just a waste of time compared to buying the right PSU, I’m unemployed right now—so you’re wrong! My time is valuable and I have plenty of it. Plus, it would be interesting to monitor temperatures and see if my airflow solution worked well. I think three fans should suffice. I could run them at full speed, but I’m also thinking about connecting them to the PSU’s fan controller.
I had to stop sending my responses because of spam detection.
The workload shouldn't be too heavy at first until I experiment more. Right now, my main ideas for this server are running VM's for security camera software, a node, a lightning node, a BISQ node (maybe), Plex, and for learning or a homelab—some IT experience but I want to get better. My NAS stays separate. The GPU will mainly handle transcoding. Here’s a photo of a mock-up showing the components. Please be gentle; this is just a mockup. It’s a 2U rackmount shelf. I’m still working on the mounting brackets. The GPU will be mounted in a suspended position as described. There will be a lid, plus front and backplates. And yes, the PSU is going to blow out the front...
Just an i5-8600k with an old 1060 GPU. My friend had it lying around. Honestly, I'm not sure if integrated graphics will work well here—I'm really new to Plex and think a 1060 is probably too much. But since it wasn't being used, I thought giving it a purpose would help. As Kurt Vonnegut said in The Sirens of Titan:
"The worst thing that could possibly happen to anybody would be to not be used for anything by anybody." lol
I considered how I could monitor the PSU temperatures for better assurance, and I recalled having several ATtiny85 units from an earlier project, a thermister from the old hot end of my 3D printer (now upgraded to E3D v6), and an MLX90614 IR temperature sensor from a previous build. I believe the thermister would be ideal for attaching to the switching MOSFETs to directly measure their temperatures, as they produce the most heat. The IR sensor would be placed on the inside top of the PSU, "looking down," to track the overall temperature of the remaining components.
I borrowed code from the internet that uses the thermister to send temperature readings via virtual serial over USB for logging to an application on a server (https://www.hackster.io/najad/phone-ther...y85-991942).
Regarding the MLX90614, I wasn’t able to find exactly what I needed, so I let ChatGPT generate a similar snippet for me. Then I merged it with the code I found, allowing both sensors to run from the same device. I’m planning to create a simple script using ChatGPT to automatically shut down the server if temperatures rise too high—no need for a complex serial logging tool. I won’t be able to test it until the resistors I ordered arrive in a day or two; I’ll share updates once everything is set up.
Below is the IR sensor already soldered to a cord and heatshrunk (left), the thermister (top) and two ATtiny85s.
I understand "vibe coding" often faces criticism, but it’s been incredibly useful for completing this small project while also teaching me a lot. Despite having very little experience with digital electronics and programming, I asked ChatGPT numerous questions, and its responses were both detailed and easy to follow. It’s still early to test the code, but based on my feedback, it should compile without issues and fit well on the ATtiny85. I’m confident it won’t be too difficult to adjust it for a successful run.