Getting PCIe power from a proprietary PSU without enough connectors (without causing a fire)

Valantar

Shrink Way Wielder
Original poster
Jan 20, 2018
1,743
1,626
Tl;dr: I'm not an electrical engineer, and I'm planning on doing stuff I probably shouldn't. Can you help?


So I'm working on a project that's growing increasingly complex, "fun", and risky at the same time, modding an old Optiplex 990 into a light gaming PC. After discovering that the PCIe slot is limited to 35W and thus can't run anything beyond a 1030 without auxiliary power, I decided to go balls-to-the-wall, got a (really cheap!) used RX 570 ITX, and I'm now working on two issues: how to fit it in the case, and how to power it.

The PC has a (quite shitty, as far as I can tell) proprietary 240W PSU, with a maximum 12V power output of 17A (204W) on its single 12V rail. According to Dell's manual there are two PSU options for this PC: one 80+ rated, and one with a "mean efficiency" of a downright terrible 65%. I'm pretty sure I have the latter, and can't find the former for sale anywhere reasonable, so for now I'm stuck with this (I have options for a PSU upgrade down the line, but that's not in the cards for now - this is a near-zero budget build).

Paired with an i5-2400, 4 sticks of RAM, an SSD and three case fans, some underclocking of the GPU is definitely in order if I'm to avoid blowing the PSU up (or at least tripping OCP (if the PSU has it)). That should be doable, though, as I'm not looking to squeeze maximum performance out of this, but make something very cheap and reasonably portable for light gaming. People on mining forums report getting RX 580s under 100W, so I'm thinking this is doable - running Prime95, the CPU never reports package power above 71W, despite its 95W TDP.


So, to the point of this thread: I need a 6-pin PCIe power source for the GPU. The PSU, reasonably, doesn't have one. It has three accessible 12V leads: two 18AWG for the 4-pin EPS, and one 20AWG for the SATA connectors. The 24-pin connector is there too, but too short to be tapped into. My thought is to cut these three wires (along with matching ground wires), terminate them with a 6-pin mini-fit connector, and do the same on the other side, crimping an extra wire for PCIe power into each pin alongside the EPS/SATA power cables. Given that they're all coming from the same 12V rail, I don't see how this would be a problem.

From looking at wire gauge calculators, this should be okay. The three wires combined have an effective gauge of around 14AWG (three 20AWG =~15AWG, three 18AWG =~13AWG), which should be perfectly capable of handling far more than the PSU's maximum 17A over 20cm or so before the first terminal (again, according to a calculator, 14AWG can handle 17A over 5.6' with below 2% loss, and more than 2' with <1% loss). The EPS wire will be left alone, while I'll be running a custom cable from the SATA power cable to power the PCIe riser and a USB 3.0 AIC.

I made a very rough wiring diagram (EEs: please don't hate on my lack of diagram skills!)



Am I going to burn my house down?
 

Valantar

Shrink Way Wielder
Original poster
Jan 20, 2018
1,743
1,626
So, two likes and no comments. Am I to interpret a like as a vote towards this not being likely to burn my house down? Or more along the lines of "do it, we want to see if you survive"? :p

At any rate, I've gone ahead and ordered some mini-fit connectors. As I'm trying to spend as little as possible on this, I ordered from Aliexpress, which means it'll likely be a month until this is put into practice. Plenty of time to figure out how to get the GPU into the case, at least, and to figure out how to fill the front with fans without compromising structural integrity too much.
 
  • Like
Reactions: caniplaymayo

Choidebu

"Banned"
Aug 16, 2017
1,172
1,160
Splitting just the eps into eps and pcie is fine. However, you don't need to cut that sata cable. At all. Remember that current flows through path of least resistance, so if you mix up multiple cables of differing awg the thinnest one, hence the least resistant one will carry more current than the thicker one. Didn't do the math but it's possibly gonna melt.
 

Thehack

Spatial Philosopher
Creator
Bronze Supporter
Mar 6, 2016
2,735
3,494
J-hackcompany.com
Am I going to burn my house down?

50/50 chance it seems like. I don't think it's a matter if can you run 240W through it, I think it's a matter of can you wire it cleanly? That and your PSU is likely bottom barrel so it'll fry if you try to use it with a GPU.
 

Valantar

Shrink Way Wielder
Original poster
Jan 20, 2018
1,743
1,626
Splitting just the eps into eps and pcie is fine. However, you don't need to cut that sata cable. At all. Remember that current flows through path of least resistance, so if you mix up multiple cables of differing awg the thinnest one, hence the least resistant one will carry more current than the thicker one. Didn't do the math but it's possibly gonna melt.
Won't the thinnest cable be the most resistant one? My understanding of electrical resistance is that it increases the longer and thinner a cable is - have i misunderstood this completely? Also, if it's the least resistant as you say, why would it heat up, when resistance is what causes heat losses?

As for current flowing through the path of least resistance, all three wires come from the same 12V source, and will be going into the same sink (the GPU), so as far as I understand the three wires would then act as a combined, thicker wire, with the current dividing itself across them to match their resistance. Using more wires thus simply serves to lower overall resistance, and thus minimize heat losses in the cables. Please correct me if I'm wrong, of course, but per my current understanding adding a third, thinner wire will still lower the stress on the two others (and save me the hassle of splitting one of the two EPS 12V wires into the third 12V required by the PCIe cable.

50/50 chance it seems like. I don't think it's a matter if can you run 240W through it, I think it's a matter of can you wire it cleanly? That and your PSU is likely bottom barrel so it'll fry if you try to use it with a GPU.
I have a decent quality crimper and have modified PSU cables previously (which are handling far more power than this in my main desktop, among other things powering my Fury X), so I'll probably be fine in terms of wiring. I don't have much experience with crimping two wires into one pin, so that will be a minor challenge, but with some spare pins to practice on I'm sure I'll be fine. It's not like 18AWG is very thick, after all.

I agree that the PSU seems like utter garbage, I'm just hoping that it can sustain close to its maximum 12V output rating with sporadic use until I can replace it with something better. I'll be putting some effort into minimizing power draw here, of course. The 3.5" drive that the PC came with is being replaced with a spare mSATA SSD I have lying around, and I'll be underclocking and undervolting the GPU until I reach acceptable power levels. Ideally, I'd want to get to around 220-240W wall power under load, which would mean a 12V DC draw well below 200W with the terrible efficiency of this PSU. For reference, when I said the system reported 71W package power under Prime95 above, that measured as ~135W at the wall, though that was including the 3.5" drive (need an adapter for the SSD that hasn't arrived yet) and a Radeon HD 5450, so if I add something like 20W to the CPU package power for the rest of the system, that translates to 67% efficiency - right where it's rated. 200W would then turn into 270W at the wall, which would also mean that the poor 60mm PSU fan would be screaming its head off trying to dissipate 70W, but I'm helping it along by directing the stock high speed 80mm fan into its intake. Still, minimizing power draw is definitely a job to be done here. I'll be running the PC off a spare ATX power supply while tuning it to avoid any unintended meltdowns and check if this is at all viable.

When it comes to PSU replacement options, ideally I'd want this stock one to last survive until I can afford a motherboard+cpu+RAM upgrade here, as I can fit a TFX PSU in here if the motherboard is replaced with an ITX board - which would make all of this quite simple.If I have to replace it before then, the best option seems to be using an external brick - I have a spare 250W pico-style DC-DC PSU lying around, so I might just look into getting a powerful enough 12V brick to run that off, though with the minimal spare cash I have available at the moment that'd likely have to be a janky AliExpress brick, which isn't likely to be any better than the current PSU - at least that's made by a reputable company and certified for use in the EU. I would love to find some sort of internal AC-DC unit that could fit in the space left by the internal PSU, but sadly its long and thin shape makes that a challenge. The only one I've found that might fit is the MeanWell UHP-350-12 (350W output with convection cooling, 91% efficiency, so near ideal) but it has a max voltage ripple spec of 200mv, which is well beyond the ATX spec. It's likely that actual ripple is far lower than this, but it's still concerning, so I'm not willing to jump on anything like that yet. At the very least I'd need to look into adding capacitors to the output to lower that ripple.
 

Choidebu

"Banned"
Aug 16, 2017
1,172
1,160
My mistake - you're right it'll split the current between them according to resistance, basically thickness and length.
And thinner means more resistance not less.

But I stand by my thinking that it's not a good idea. More resistance means more heat (I squared * R), and while it's fine for sata ports that drives only takes tens of watts, when you're talking hundreds split into 2 beefier cable and 1 thinner cable, we can do the math and see how much wattage the thinner one will waste as heat.

While I'd wager (can't be arsed to do the math lol) it's fine within its rating, we also have to consider the fact that multiple cables running the same rail through a port is done so not just to split the load, but to provide minimum redundancy so that even if one doesn't mate properly for some reason the other can pickup the slack.

What this means is you don't play just right below the rating, best practice is 50% that. So maybe figure out what the waste would be for 1 beefier cable and 1 thinner cable, see if it's still within the thinner one's rating.

But yeah after all that it'd probably be fine.... your solution above just looks overengineered and convoluted for me.
 

Valantar

Shrink Way Wielder
Original poster
Jan 20, 2018
1,743
1,626
My mistake - you're right it'll split the current between them according to resistance, basically thickness and length.
And thinner means more resistance not less.

But I stand by my thinking that it's not a good idea. More resistance means more heat (I squared * R), and while it's fine for sata ports that drives only takes tens of watts, when you're talking hundreds split into 2 beefier cable and 1 thinner cable, we can do the math and see how much wattage the thinner one will waste as heat.

While I'd wager (can't be arsed to do the math lol) it's fine within its rating, we also have to consider the fact that multiple cables running the same rail through a port is done so not just to split the load, but to provide minimum redundancy so that even if one doesn't mate properly for some reason the other can pickup the slack.

What this means is you don't play just right below the rating, best practice is 50% that. So maybe figure out what the waste would be for 1 beefier cable and 1 thinner cable, see if it's still within the thinner one's rating.

But yeah after all that it'd probably be fine.... your solution above just looks overengineered and convoluted for me.
I think we're still either misunderstanding or talking past each other. Adding a third wire, regardless of thickness, will lower overall resistance, which is all that matters when the source and sink are the same - in effect, the three wires act as one. I might be wrong here, but my understanding is that the current will divide itself across the wires according to resistance, in other words more current will flow through the thicker wires.

As for any overhead, here's the calculations when put through two calculators (combined wire gauge calculator and resistance/loss calculator):
(All calculations are based on 12V, 17A - the max rating of the PSU, which I'm not planning to hit, and setting 1% loss as acceptable - as low as the calculator goes)

2x 18AWG + 1x 20AWG = effective gauge 14, max length 2.8'

1x 18AWG + 1x 20AWG = effective gauge 16, max length 1,76'

2x 18AWG = effective gauge 15, max length somewhere in between (calculator doesn't show 15), likely over 2'

1x 20AWG = max length 0.7'

So: running this off a single 20AWG wire is theoretically possible, as the length from the PSU's PCB to the splice will be no longer than 0.7' - though this is of course riding the edge very close - but would still just mean a minor voltage drop and around 2W of heat to dissipate from the wire, which ought to be possible. A 20AWG and 18AWG combined can do this with more than 50% margin, and all three put together is more than 3x what's needed - which is good for a DIY project like this.

As for my solution being overengineered - why would I go any other way? I'm modifying a PSU, and a bad one at that - why not be extra careful? And given that I have to add a splice anyhow (unless I want to solder new leads onto the existing ones, which I really don't want to do), and connectors are always the limiting factor on current transmission, a bigger connector (more pins) is always better. Pulling in a third wire to share the load off the two others that will also be powering the CPU just seems like common sense to me. Better safe than sorry, especially when the effort required is as minimal as this. Besides, not pulling in a third wire from the PSU would require me to bridge one or both of the EPS wires at one of the connectors or split one to two to get the three required for the PCIe connector, which seems like a much worse idea to me - more difficult crimping, more complex wiring, nothing to be gained except a less robust solution.