Pico PSU (Power Supplies) the junk end

[beanflying]

Had a good search and apart from some older threads no real info or testing on these I could find.

Just running a some stress testing on a system I am putting together to run my CNC gear. Basics are 5600G, MITX, 16Gb, & a 500Gb NVMe.

To keep the board and enclosure as small as possible I decided to have a look at a PICO ATX style of power supply and with more than some suspicion brought a 'better looking' evilbay 200W example for this project.

Powering it at this stage is a 170W Lenovo brick I picked up as NOS. Search for '200W DC 19V Pico ATX' will find any amount of the little boards but there is plenty of other styles out there some look like total garbage and then you can go up to 'branded' boards for more $.

For no reason other than I have it on my NAS in my benchmarking utilities I ran OOCT to stress the system in a 30 minute run as it allows voltage rail monitoring and board temperatures too. https://www.ocbase.com/download

Not shown was a mains wattmeter but it was holding a very steady 140W for the entire time. In spite of the hot spot temperatures being way over what I would think 'sensible' the voltage rails remains stable.

The 80W power limit images which is closer to a real world maximum workload for this PC would see this nominal 200W board likely be 'ok' without mods.

Not conclusive but for this sample of one there is no sane way this board could be rated at 200W (not really surprising either ) in stock form but I will fit some small heatsinks to the offending MOSFETS https://datasheet.lcsc.com/lcsc/2206231830_U-NIKC-UK650BA_C3034555.pdf Reality is I guess anything toward 50% of that rating you shouldn't catch fire.

Also this is sort of worst case I guess as the memory and open board has no airflow here.

[beanflying]

Not great but 'better' used three baby heatsinks that come with the cheap plug in reprap stepper drives. Tests below were 10 minutes into the same 140W test.

[T3sl4co1l]

Those heatsinks aren't showing very well; maybe try to get a look under the edge, directly at the transistors.

Tim

[Jeroen3]

Considering I had almost the same experience with an Inter-tech ITX-601 where at only 80 Watt the DC jack of the case was very hot.
The embedded mini-psu board wasn't in much better condition, it came with a short! But at least it survives normal load.

I had to reduce the power limit on the Ryzen 3200G, which wasn't a problem in a media pc.

A 5600G is asking for problems.

[beanflying]

Given the 3200g and 5600g have the same TDP unlikely to be more or less of an issue. with 100%CPU load is where I was getting the 80W figures from and it isn't until you also crank the GPU to 100% that you can hit 140W which is a really unlikely real world use case unless you are trying a render or video work in which case it just isn't the right device I guess. In my use case GPU loads will be really light as Fusion doesn't use GPU acceleration and the CAM software can be run on a postage stamp. Also I am running gold bullet R/C connectors and different Silicone covered wire than came in the box.

More of an edge on shot of the board. It is a bit hard to tell but I wouldn't mind betting that it could now be the PCB is the hot spot instead of the Mosfets I will have another look at it tomorrow but I NEED to go watch the Giro and it is 10.30pm local

[Jeroen3]

Apparently the 5600g is more efficient. Not sure if this also applies to synthetic loads though... Fun how 2 years of new tech improves things.

[DiTBho]

ummm, we should bench-test those PSUs with test loads before actually using them 

[nctnico]

You can't measure aluminium heatsinks like that with a thermal camera. Their surface is too shiny. You have to create a mat black spot (for example using masking tape which is made black using an ink pen) to measure the temperature accuratly.

Either way, these units are likely designed to have some airflow going over them as well.

[T3sl4co1l]

No worry, plain masking tape is plenty black at thermal wavelengths. Even better, vinyl electrical tape!

Other tapes tend to do poorly: yellow transformer tape (polyester/acrylic?), or clear gold (polyimide), are typically too thin (fractional wavelength) so don't increase emissivity as much as you'd expect from their materials (which is the emissivity of bulk material).

Same goes for ink -- even a heavy mark (let dry and re-draw multiple times) from a permanent marker, while enough to be visible in thermal IR, still has quite low emissivity.  A nice thick layer from a paint, grease or wax pen(cil) does nicely, however.

You can also look for bright peaks -- occasionally you'll find an inside corner with total internal reflection, which therefore acts as an accidental black-body resonator, thus representing the actual temperature of the metal.  These are quite hard to spot / rare to find with low resolution consumer cameras, though.  (As I recall from a good industrial cam, philips screws and inside corners of heatsinks can occasionally do this.  But the spot size is quite small -- specular reflections basically.  To be more specific: the heatsink temperature is at least max(<pixels within the region of the heatsink>) -- but can still be more than that, because you still don't really know the emissivity of those brightest spots.)

Tim

[beanflying]

You can't
measure aluminium heatsinks like that with a thermal camera. Their surface is too shiny. You have to create a mat black spot (for example using masking tape which is made black using an ink pen) to measure the temperature accuratly.

Either way, these units are likely designed to have some airflow going over them as well.

Providing you correct for the emissivity you can use a thermal camera on Aluminium whichI did for those extra tagged points in the Fluke connect software. Can you get them +-1C doing that then NO but it is a whole lot closer and at this scale likely better than tape.

I will add some air outlets in this corner of the case to push a bit of air past them too based on this testing and tweaking and also reduce the ones at the bottom of the case as the NVMe now has a decent heatsink.



EDIT: Tweaked version attached, reduced VRM and NVMe cooling outlets and added some for the Pico