Building an Unregulated Linear Power Supply

[oneguy]

As soon as a fatten up my wallet from my server build this is next on my list. Thanks @elan120!

[oneguy]

12 hours ago, dminches said:

@elan120, why did you opt for copper bars instead of wire?  Also, where did you get the copper?

I haven’t built this power supply yet but you can also get copper bar here:

 

https://www.onlinemetals.com/en/buy/copper-rectangle-bar

 

I used bad stock from them in a kit car fuse box and it worked well. 

[oneguy]

Here’s a scale drawing I did for my planned ULPS build in a H5 case. I don’t have the funds yet for all the components but I figured someone else may want the inspiration. It’s to scale except for the hole (wide enough to drop 6-pin connectors through). 
 

Notes:

-Transformer is sized for Noratel transformer

-VCAPs not depicted but they will go on top of Mundorfs

-Heavy black lines will be copper plates to partition the different stages (power conversion/rectification, capacitor, and DC-ATX)

-Five 6-pin (four for the ATX and one for the CPU) connectors will come out the top of the lower H5 case (server)and plug directly into the Taiko. Short routing and no connectors to worry about. 

[oneguy]

For those in North America, it’s cheaper to order the Mundorf capacitors, Hypex soft start, and Furutech IEC from https://www.audiophonics.fr/en/ than using parts http://partsconnexion.com

 

$700 with shipping from Audiophonics vs $930 from partsconnexion and that doesn’t include shipping or the soft start.

 

In fact, you can even add in the slimline case like modushop sells and the perforated case bottom and it’s still only $875 with shipping from Audiophonics. 
 

Pretty big savings.

 

On the subject of savings, it’s $43 for shipping for the Saligny rectifier which cost $44. The shipping is flat no matter how many you order. I can place a large order so we save on shipping then mail them out to people individually. If anyone wants to jump in just shoot me a PM by Feb 25th

 

Costs below include USPS priority mail to from me to you. 

[oneguy]

10 minutes ago, bit01 said:

@oneguyDo you plan to fasten or brace the cases together in some way or is one simply sitting (with its feet) on top of the other?

When I was planning on putting the Taiko in the server chassis I was going to use and XLR connection to link the two (one on the top of the server and one on the bottom of the ULPS) which would have required precise mating and a guide. The method I described in my previous post wouldn’t require precise mating and unless there was an earthquake they would slide in relation to each other so no bracing is required.

 

It can certainly be added though

-Spiked feet isolators

-A cup and ball type system

-guiding rod and cylinder

[oneguy]

Closing the Saligny Rectifier group buy at midnight PST. PM me if interested. 

 

[oneguy]

I finally got the bulk of my components in so I can set the layout. The plan is to stack this on top of my server and drop the DC wire to the Taiko DC ATX where the red circle is here:

in the top of my server case here:

the ultimately to the Taiko DC in here (not my photo):

 

No connectors to worry about except the one into the Taiko and very direct. 

[oneguy]

14 minutes ago, MarcelNL said:

I'd rather stack the server on top, did you consider that? You want the heat of the passive cooling to dissipate and convection drives it upwards.

My take was convection heat travels upwards. I am less concerned with the server preheating the air going to the the ULPS than the ULPS preheating the air going to the server since the server will already be running hot with upsampling. 

 

Also a nice secondary is the massive copper heatsink may provide an additional obstacle to EMI from the transformer in the case directly above. 

[oneguy]

@MarcelNLeither way I guess it’s good the top holes is one case don’t really line up with the bottom holes of the other case they won’t direct port hot air from one to the other. 
 

https://www.ebay.com/itm/4Pcs-Aluminum-HiFi-Speaker-Isolation-Stand-Feet-Base-Pad-Ceramic-Bead-Anti-shock-/373760117444?mkcid=16&mkevt=1&_trksid=p2349624.m46890.l49286&mkrid=711-127632-2357-0

 

I have taller feet to separate the cases since the stock feet aren’t tall enough. They split into an upper and lower half so the top half will stay mounted to the ULPS and lower half mounted to the server. Dampening and perfect alignment each time there are stacked. 

 

The server is also getting those feet to elevate it so it traps less hot air below it. 
 

Other random touches: copper washers on all the screw terminals and copper screws. I just ordered the copper screws today now that I know the dimensions. M5x8mm for the 22,000 capacitor and M6x10mm for 47,000 capacitor. 
 

https://www.ebay.com/itm/99-9-Pure-Copper-ISO7380-Hex-Socket-Bolts-Button-Head-Screws-M3-M4-M5-M6-/185146765196?mkcid=16&mkevt=1&_trksid=p2349624.m46890.l49286&mkrid=711-127632-2357-0

 

https://www.amazon.com/dp/B08SHYWDFK/ref=cm_sw_r_cp_api_i_C5DE7T0N3AKNGV74S3V4?_encoding=UTF8&psc=1

[oneguy]

1 hour ago, MarcelNL said:

It's not about the holes, but the heatsinks. I can easily see a temperature difference when airflow is unhindered underneath and above my 300W heatsink for the CPU versus when something is blocking it.

Design is about trade offs and I’m prioritizing shorter cable runs over the server case being completely unrestricted above it. There will be an inch of clearance between the two cases and heatsinks. Also the heatsinks in the case above do not completely block convection airflow from the heatsinks below. If the cases line up perfectly, which is my intention, then the free space between the heatsinks will also line up perfectly so the convection air currents from the heatsink will have a direct path upwards. 

[oneguy]

Finally got to drilling and tapping holes today. The final capacitor will go in once I cut the screw for the foot that’s poking up to be flush. 

The blue tape represents where the shielding plate will be. 

[oneguy]

I read that comment that adding a screw through center of a toroidal transformer reduces it efficiency. Obviously not everything you read is true but has anyone come across that? My plan to avoid that was to use 3M Dual Lock tape to secure it. Thoughts?

I was uncomfortable with the distance of the the capacitor terminals to the lid so I added 1/8 adhesive backed neoprene to the lid. 

[oneguy]

@Rovogood to know. Sounds like a trip to the hardware store is in order.
 

I wish I could use the hex head cutout on the opposite side of the transformer but the orientation doesn’t work for my layout. 

[oneguy]

The internal case size of the H5 is 88mm and the transformer is 67mm tall so I’ll be safe there. Hex bolt head and washer is probably 10mm max. 

[oneguy]

Wiring inside the server done for the most part. 

Im using the stock HDPLEX ATX power cable to connect the ULPS to the ATX until I can validate having the server on bottom due to heat concerns. Once that’s done I’ll make the cable out of 14 awg and drill the appropriate holes to pass the cable through. 

I also need to solder in the fuse holder that I drilled a hole for and remove the inline fuses that I am currently using as well as put in a thicker ground wire. 

I highly recommend this little guy for stripping the Teflon off the Neotech wires. It made it a breeze. 

 

Anyone else notice their CPU becoming audibly more noisy when using the ULPS? Mine is currently emoting a noise like its trying to read something. It’s the BIOS noise but with pulses. 

[oneguy]

3 hours ago, Nenon said:

 

Someone is really taking short wires to extreme levels. Nice. 

 

Thanks. My original plan was to put the Taiko ATX copper side up directly above the 24-Pin but with the RAM there it would limit how close I can get the Taiko so I decided it probably wouldn’t be any shorter than this for way more hassle. 
 

Those wires are each 4.5 cm long and I used a marker as a die to uniformly form them all into a U shape before inserting them in into the connector. 
 

For one of the EPS cables routings I could have shaved a few cm with this pathway but the thought didn’t occur to me until just now. 

[oneguy]

The victory was short lived. Apparently the rectifier get really hot when it’s doing it’s thing and I had mine resting on the jacket of the capacitor. Well then it stated melting the jacket of the capacitor and when I went to lift the rectifier off the capacitor jacket, one of the chips stayed stuck to the jacket. 
 

It’s kind of hard to photograph but it’s the shiny item in the green circle. 

 

There was room to gap the two pieces I just didn’t think about doing so because I didn’t anticipate it getting so hot. Hopefully this knowledge helps someone else in the future. 
 

I’m going to have to dissemble and see if the chip can be resoldered or order a new rectifier. 

[oneguy]

Well the only other thing that comes to mind are the jacketed wires feeding the ATX were close to the traces but I don’t think that could be it. The jackets are intact and no sign of shorting through the jacket. 

 

Maybe the neoprene wasn’t enough of an insulator from the case? EDIT: there was a 1/2 to 3/4” gap between between the rectifier and the neoprene. The neoprene is 1/8” thick. 30v won’t jump that large of a gap both ways and go through the neoprene.

 

[oneguy]

I am also pulling more current than those in the US since I’m operating off of 31V rectifier output due to 103-105vac from the wall.

 

Any theories?

[oneguy]

13 hours ago, bit01 said:

There are conditions where you will be drawing more than 5A with the 19-12900K, even during boot. If you happen to saturate the chokes long enough their temperature will rise. The choke lead may be getting warm and conducting heat back to the cap terminal and rectifier lead. With the components face down  as you have them especially if touching the cap, their air cooling is impaired.

I would use an 8A or 10A choke. Saturation drops the inductance and distorts the ripple as well, l introducing additional unwanted harmonics. 

Just a thought.

ATB.

b.

 

10 hours ago, MarcelNL said:

although the choke may well be saturated at times, yet for the heat to travel up the lead to the cap and then heat the rectifies seems pretty impossible to me or you'd see the cable and or cap meltdown first.

Using a larger choke is a good suggestion anyhow

 

I agree with @MarcelNL. The lead wire would show damage as well. Plus the sticker on the top of the choke would show some type of effects of if that kind of heat was generated. I dot believe I was getting into saturation atleast not at idle. 
 

I managed to get my power consumption low enough to allow the HDPLEX 300W to power my server with DSD512 upsampling (60% CPU load at 4.3Ghz on th P-cores). The HDPLEX only has a 190W capacity on the 19V output so if I was using all 190W with my 30.8V ULPS output then that would be 6.17A. So is it possible the choke was saturated? Yes it is possible, but I don’t think idling at base CPU frequency would have saturated the chokes if a CPU load of 60% at 4.3Ghz is drawing less than 6.17A. In my server only the Optane is using motherboard power. All RGB  lights removed, LAN and SATA controllers disabled and WIFI/Bluetooth transmitter removed.

 

6 hours ago, Exocer said:

What larger choke are people suggesting these days?

I have Hammond 195J10’s onhand which can handle 10A. If the 159ZJ10 proved unworthy I was going to swap them in. Keep in mind you will need 220,000uF of capacitance after the chokes per @Nenon’s post on Sep 14, 2021 in the DIY Music Server thread if you go with a 10A choke. 
 

 

[oneguy]

Yep, mainly just bringing that info back up on everyone’s radar so they know additional requirements come with a larger choke. Not specifically directed at you.

 

Some may not have the space. In my case I’ll have have to rethink my layout because larger chokes and 2 more caps will dominate the space I have left. 

[oneguy]

3 minutes ago, MarcelNL said:

Which post exactly? I just looked back and did not find it on Sep 14?

Sorry I couldn’t remember how to just just grab a partial quote from another thread. 
 

Since the timestamps displays are slaved to your user settings it may appear as Sep 13 depending on the time of day he posted it. 

 

[oneguy]

2 hours ago, elan120 said:

As a reference point, I just did a test to take some readings, and the initial bootup on my i9-12900K maximum power draw is 124.7W / 30.04V / 4.14A.  After bootup, in Bios menu, power draw is 69.4W / 31.84V / 2.18A.  This test ran with CPU at 4.9GHz, and using a 4A fuse on the power supply.  Note, the voltage output with no load is 35.4V.

Makes sense. My voltage output is 87% of yours so extrapolating that information possibly the 3 songs at upsample may have saturated the chokes but the 8-10 hours at idle prior to that wouldn’t. 
 

I’ll be using 100v/50hz wall power up until Nov then I’ll permanently be back on US spec power so if that 13% ends up being the margin that I need to prevent saturation I’ll finish the build here and table the ULPS until I move. 

[oneguy]

Aftermath: 2 chips removed and 1 displaced. 
As you can see from the underside of the chip (they are all the same part number) there is one large pad connecting 4 of the pins and and the other 4 pins connect to their own individual pads. 
 

There was enough solder present to reflow one of the them back but I need to get some solder paste for the other 2. 

 

 

[oneguy]

2 hours ago, MarcelNL said:

 

did you check the rest of your LPSU for shorts ? measure the first choke?

I can only imagine there is a short or similar making the current draw very high, but that active rectifier should be able to withstand currents that would leave your transformer real hot....weird? Did you contact the maker of the Saligny rectifier? What voltage do you measure after your transformer?

 

I went through and checked and nothing is bridging the positive to negative terminals after the rectifier except the VCaps and the intended wiring. Nothing is bridged to ground except the IEC and transformer static shield. Besides above where the rectifier was, I don’t see any signed of excess heating of the neoprene insulator on the case lid. No part touches the lid and to arc through the neoprene would require a burns mark I would imagine.

 

Two of the VCaps are close to terminals but they are surrounded by a plastic insulator and there are no burn marks from arcing. All VCaps measure as an open load so no short there. 

 

I actually popped the cover off to specifically check the transformer temp. It wasn’t that warm. That’s when I discovered a plastic melting smell and let me know something wasn’t right. The server was operating properly. 
 

I had a 4A fuse installed in the output from the IEC and it didn’t blow so no more than 420w could have been flowing. The rectifier is rated for “continuous load current up to 20A or more depending on mosfet configuration.” Assuming there wasn’t a short the Taiko input voltage would have had to sag from 31V to 21V and the server would need to be demanding 420w. Not likely based on what I had powering it before. 
 

Did I miss anything? I’m still stumped.

 

Underneath the thin heat shrink boot around the capacitor there is a metal rim. Maybe the heat shrink got hot and thinned out? But what would have caused the heat shrink to get that hot in the first place?