designing the ideal ATX power supply replacement for vintage Macs

[bigmessowires]

@jmacz and I have been discussing various ATX power supply options for Mac II series computers, and the challenges of 5V and 12V regulation. We're hoping to fit the ATX guts inside the original Mac PSU's enclosure so that it looks stock. For most Macs (we're looking at the IIci, IIcx, and IIfx now) this requires using a small form factor SFX PSU or an ATX Flex PSU, because a standard size ATX is physically too big. There are lots of posts here and elsewhere about how to do this, and building a small inverter circuit to make a PS_ON control signal from the Mac's PFW signal. It works, except...

If you're using an ATX PSU with your vintage Mac, have you actually measured the 5V and 12V supply voltages while the Mac is running? If your Mac is heavily loaded with lots of cards and accelerators and other goodies, you may be unpleasantly surprised. Naturally we want to use an inexpensive ATX PSU for a project like this, and since it doesn't require high wattage by modern PC standards, that seems like it should be possible. But cheaper ATX PSUs are typically group regulated, which means that the 5V and 12V supplies are regulated using a single shared feedback circuit instead of being independently regulated. This can create problems in a cross-loading situation where the load on one supply is much greater than the other, like in a vintage Mac with a high 5V load but minimal 12V load. In contrast, the original Mac PSUs are (I think) independently regulated.

@zigzagjoe pointed out this problem in another thread recently, but initially I wasn't convinced. Then @jmacz and I tried an ATX-conversion based on this Logisys PSU, and the results were... disappointing. Testing in a Mac IIci and a Mac IIcx, the more cards and peripherals we added, the lower the 5V supply sank and the higher the 12V supply rose. With a fully-stuffed IIcx my 5V supply sank as low as 4.73V and @jmacz's IIci sank as low as 4.69V. That's low enough that it will very likely cause stability problems.

I you have an ATX PSU and your Mac is heavily loaded, I would love to see data about your Mac's configuration, what ATX PSU you're using, and what your 5V supply voltage is when the system is running. I'm guessing that most people who've done an ATX mod either have lightly-loaded Macs where this isn't much of a problem, or they just never measured the voltage and didn't realize there was a problem.

So how do you fix this problem? The simple way out is to buy a more expensive PSU that's not group regulated, but in practice this is challenging and expensive. The candidate PSU must have independent regulation, and be a small form factor, and supply sufficient amps on the 5V rail (most don't). There aren't many options, and most of the options that do exist are $100 are more.

A better idea (from a technical standpoint) would be a community-designed PSU adapter that's specifically designed around the needs of vintage Macs and similar computers that demand high amps on 5V with tight regulation. I think the best way to do this would be starting with a commercially-available regulated 12V power supply (not necessarily an ATX supply) and then add DC-to-DC regulators and converters to make -12V, +5V, and +5V trickle from 12V. The end result would be similar to an independently-regulated ATX supply except without the unneeded baggage of 3.3V or ATX connectors/cables, without an unnecessary overabundance of 12V capacity, and the 5V capacity and the physical size could be specced to the needs of Macs.

You can already buy all the necessary 12V to XXX regulators and converters as separate modules to create a Rube Goldeberg solution. I think what we need is somebody who understands PSU design and can incorporate all of those modules onto a single space-efficient PCB that could fit inside the PSU enclosure along with the 12V supply. I don't have the knowledge to do this myself, but maybe others do, or a group of people working together. Once done, it could be a "standard" design that anybody could get assembled at their favorite Chinese vendor and combine with an off-the-shelf 12V supply for a robust final solution.

Thoughts on this idea? Criticism? Is this crazy? Is there a better way?

 

[zigzagjoe]

To help group regulated PSUs, you can put a small load on the 12v rail to pull it down a little - especially without a HDD present. The 5V rail should rise a little in return. A dummy load may also be useful for the original PSU (when no HDD) as the 12v rail tends to float high without a drive.

A power resistor works, or an automotive light bulb is the cheaper/easier option. A 10 watt bulb should be fine, make sure you locate it away from plastics. Of course this is only a band-aid, and you're throwing away the efficiency improvements of using a modern PSU, but it does work. I'd be curious if you wanted to measure how much of a change you see.

WRT a long term soluation: It may make sense to reverse the DC-DC for a MacII PSU design. I suggest to find an off the shelf PSU module (from Mean Well, CUI, Lite-on, etc) that can deliver oodles of reasonable-quality 5V, then have secondary DC-DCs to generate the ancillary rails. The 5v rail is the most fiddly, after all, where the secondary rails are much less sensitive. Reference designs would likely be sufficient for the extra rails without much design heartburn required. Often you can find designs to generate +- 12v in one shot, too. You give up some efficiency since you'd be boosting to +-12v rather than bucking down, but given the low amp requirements on these rails it'd save having to validate a DC-DC design that can deliver 20+amps@5v.

My data:

SE/30, Carrera 040, Interware GrandVimage, Maccon, 8x SIMMs, HDD
Seasonic SSP-300SUB - Group regulated
55 watts at the wall (kill-a-watt), 4.8v, 12.3v

 

[cheesestraws]

given the low amp requirements on these rails it'd save having to validate a DC-DC design that can deliver 20+amps@5v.

Absolutely agreed. The things along this line I have done have always stepped up 5V rather than down 12V precisely so the prosthetic regulators can be smaller

 

[jmacz]

A power resistor works, or an automotive light bulb is the cheaper/easier option. A 10 watt bulb should be fine, make sure you locate it away from plastics. Of course this is only a band-aid, and you're throwing away the efficiency improvements of using a modern PSU, but it does work. I'd be curious if you wanted to measure how much of a change you see.

@bigmessowires tried using lights and it did help but then the point of this exercise was to keep things within the stock enclosure otherwise using one of the various 10pin-ATX adapters with a standard ATX-size PSU sitting outside provides many more options that might be cost effective.

I played with some power resistors, some 25W 25ohm and 25W 6ohm variants in aluminum finned housings mounted to the stock PSU enclosure. I was able to get the 5V rail up from around 4.69V to 4.9V but they were really really hot. Not sure if ceramic cement ones would have done better but they were too hot to touch. Temperature gun for example on the 25W 6ohm ones showed > 300F.

In my loaded IIci (accelerated video card, 50Mhz accelerator, etc) at 4.69V, I was seeing video card issues. At 4.76V I was still seeing issues. At 4.85V it was ok but there were still a couple glitches from time to time. At 4.9V the card was solid.

Another thought was maybe I should design a different PSU enclosure as well as drive carrier in the IIci that would provide more room to include a full ATX size PSU.

 

[jmacz]

Also, I think @bigmessowires looked at the schematic for the IIci/IIcx and determined that for -12V, it doesn't look like we need anything close to the 1A/0.8A on the stock PSU labels. Looks like we only need < 100mA, as long as you're not using some exotic nubus card that needs -12V.

 

[zigzagjoe]

@bigmessowires tried using lights and it did help but then the point of this exercise was to keep things within the stock enclosure otherwise using one of the various 10pin-ATX adapters with a standard ATX-size PSU sitting outside provides many more options that might be cost effective.

I played with some power resistors, some 25W 25ohm and 25W 6ohm variants in aluminum finned housings mounted to the stock PSU enclosure. I was able to get the 5V rail up from around 4.69V to 4.9V but they were really really hot. Not sure if ceramic cement ones would have done better but they were too hot to touch. Temperature gun for example on the 25W 6ohm ones showed > 300F.

In my loaded IIci (accelerated video card, 50Mhz accelerator, etc) at 4.69V, I was seeing video card issues. At 4.76V I was still seeing issues. At 4.85V it was ok but there were still a couple glitches from time to time. At 4.9V the card was solid.

Another thought was maybe I should design a different PSU enclosure as well as drive carrier in the IIci that would provide more room to include a full ATX size PSU.

Agreed. Like I said, it's a band-aid for current users. No matter how you fry it the resistor/bulb approach is sub-optimal as you then have that much more heat in your case to get rid of. And in the case of power resistors, you must cool them to keep them within specs, which means dedicated airflow across the resistors. This is at least (mostly) a nonissue with lightbulbs. Either way: evolutionary dead end.

ATX-centric design is going to end up more expensive and more complicated since it incorporates much that is unnecessary. Ironically, it is also a very legacy design and form-factor in its own way. For a greenfield project, I'd encourage the use of a design that goes 120v->5v with an off the shelf OEM part (not a consumer PSU), FETs/switch logic for softpower, and ancillary DC-DC boost 5v -> +-12v.

Also, I think @bigmessowires looked at the schematic for the IIci/IIcx and determined that for -12V, it doesn't look like we need anything close to the 1A/0.8A on the stock PSU labels. Looks like we only need < 100mA, as long as you're not using some exotic nubus card that needs -12V.

Yup, that's why I was suggesting measuring it to confirm
Only serial, sound, and some modems may use -12v. Very low load.

 

[jmacz]

ATX-centric design is going to end up more expensive and more complicated since it incorporates much that is unnecessary. Ironically, it is also a very legacy design and form-factor in its own way. For a greenfield project, I'd encourage the use of a design that goes 120v->5v with an off the shelf OEM part (not a consumer PSU), FETs/switch logic for softpower, and ancillary DC-DC boost 5v -> +-12v.

For sure. Clearly doesn't make sense to use ATX as a base for a new design.

I was just looking at whether there are possibilities for reusing these other ATX PSUs I have lying around from my kid's PCs without modification (well outside of the PFW signal) and still have it fit within the IIci case. The Logisys one just won't work due to the group regulation (@zigzagjoe thanks again for pointing this out in my IIfx thread ). But for example I have a full size ATX I harvested that has all the right specs and has a DC-DC design for the other rails and works great albeit it has to sit outside which is what I don't want.

Would be great if someone with expertise had the time to design a new one. Until then, there are a couple SFX ones that meet the 5V requirements and also utilize a DC-DC design in the $60-$90 price range (the rest are $130+).

I'm fine with using the Logisys in my unloaded IIci. And I have a stock PSU for either my loaded IIci or Quadra 700. But I need one more that's reasonable and can take some load.

 

[bigmessowires]

I suggest to find an off the shelf PSU module (from Mean Well, CUI, Lite-on, etc) that can deliver oodles of reasonable-quality 5V, then have secondary DC-DCs to generate the ancillary rails.

This makes perfect sense, and for the Mac it's a better approach than the one I originally suggested.

SE/30, Carrera 040, Interware GrandVimage, Maccon, 8x SIMMs, HDD
Seasonic SSP-300SUB - Group regulated
55 watts at the wall (kill-a-watt), 4.8v, 12.3v

I'm a little surprised it's stable at 4.8V. Maybe that's barely enough?

I did some tests with LED light strips as a dummy load on 12V. It helped some, but 12 watts of lighting was still a drop in the bucket. I believe @jmacz tested almost 50 watts of dummy load on 12V (two 6 ohm 25W resistors in parallel) and it still couldn't get the 5V supply up to 5.0V.

Here's a bit more data from my Mac IIci, filled with a cache card, ethernet card, two video cards, spinning hard disk, floppy drive, keyboard and trackball.

Stock GE PSU: 5.062V, 12.157V
Logisys ATX: 4.837V, 12.325V
EVGA 100-W1-0500 ATX: 4.824, 12.334

For a new replacement PSU concept, we'd need to identify a decent 5V regulated supply that's physically small and not too expensive. Then we could begin by stringing together some existing modules to create +12V and -12V from that, and a DIY circuit for soft power and 5V trickle with a FET or relay perhaps. Test everything outside of the Mac's case to confirm it behaves as expected. Then try to reimplement everything on a single new PCB, with a focus on keeping it small. Bonus points if it can support a few different physical connectors for different Macintosh models.

 

[cheesestraws]

I'm a little surprised it's stable at 4.8V. Maybe that's barely enough?

FWIW, my SE/40 (similar spec, but with a greycake card instead of the Interware) is also stable at 4.8V, also using the SSP-300SUB.

 

[jmacz]

Bonus points if it can support a few different physical connectors for different Macintosh models.

What are the viable models to support?

• Group 1: II, IIx, IIfx ... all have the same size PSU enclosure and same connectors. Can fit something the size of an SFX PSU and then some more room beyond that.
• Group 2: IIcx, IIci, Quadra 700, these use the same PSU and the enclosure fits an SFX size PSU but not much more room after that.

I think the Q800 and IIsi use the same connector as group 2 (10 pin) but from what I remember, the IIsi PSU enclosure is small and can't fit something of the size of an SFX PSU so the solution has to be a lot smaller? I think the Centris/Quadra 610 PSU size was somewhere near the IIsi.

LC/LCII/LCIII etc, there's not much room, so would have to be a really small solution.

 

[cheesestraws]

LC/LCII/LCIII etc, there's not much room, so would have to be a really small solution.

There are other decent solutions for these machines, not sure this needs to be a priority here, fwiw.

 

[zigzagjoe]

This makes perfect sense, and for the Mac it's a better approach than the one I originally suggested.


I'm a little surprised it's stable at 4.8V. Maybe that's barely enough?

I did some tests with LED light strips as a dummy load on 12V. It helped some, but 12 watts of lighting was still a drop in the bucket. I believe @jmacz tested almost 50 watts of dummy load on 12V (two 6 ohm 25W resistors in parallel) and it still couldn't get the 5V supply up to 5.0V.

Here's a bit more data from my Mac IIci, filled with a cache card, ethernet card, two video cards, spinning hard disk, floppy drive, keyboard and trackball.

Stock GE PSU: 5.062V, 12.157V
Logisys ATX: 4.837V, 12.325V
EVGA 100-W1-0500 ATX: 4.824, 12.334

For a new replacement PSU concept, we'd need to identify a decent 5V regulated supply that's physically small and not too expensive. Then we could begin by stringing together some existing modules to create +12V and -12V from that, and a DIY circuit for soft power and 5V trickle with a FET or relay perhaps. Test everything outside of the Mac's case to confirm it behaves as expected. Then try to reimplement everything on a single new PCB, with a focus on keeping it small. Bonus points if it can support a few different physical connectors for different Macintosh models.

Confoundingly, that machine is stable down to about 4.6v. Only below that did things start getting real odd. Extensions and dodgy connections will do that for you. I'm not entirely happy with that, but it's clearly fine.

When measuring voltages, do keep in mind you've got voltage drop over wires and connectors. Because of this you will never get a perfect 5v measured at LB/floppy port out of a modern PSU. The stock PSUs (at least in the SE30) have the ability to be adjusted to compensate, where ATX do not. I'd hazard a guess by default more often than not stock PSU were adjusted high out the door for that reason - based on my experience, anyways.

My suggestion would be something like this. Consider it a block diagram, with some more bits needed, but you get the idea.



I would suggest it is only worth targeting Quadras with card slots and MacIIs. LCs, compacts, and other pizza box machines don't have enough power demands for this to be worthwhile. Aim for a small common design that can be mounted in any housing & an appropriate cable attached.

 

[NJRoadfan]

Perhaps you could use the guts from the Universal PSU Kit? https://wiki.reactivemicro.com/Universal_PSU_Kit

It supplies all the voltages needed, but likely needs an adapter for soft power machines. Basically 90% of the way there. Also need an adapter plate for various Macintosh power supply cases. What I don't know is if the power supply is narrow enough to fit the LC series machines.

 

[zigzagjoe]

Perhaps you could use the guts from the Universal PSU Kit? https://wiki.reactivemicro.com/Universal_PSU_Kit

It supplies all the voltages needed, but likely needs an adapter for soft power machines. Basically 90% of the way there. Also need an adapter plate for various Macintosh power supply cases. What I don't know is if the power supply is narrow enough to fit the LC series machines.

Unfortunately, that design appears to be good for 6 amps of 5v. 10 amps or more can be required for a Mac with expansion slots. It might be suitable for a LC though.

 

[bigmessowires]

Consider it a block diagram, with some more bits needed, but you get the idea.

I like it. I think the polarity of C3 is backwards though? Maybe add a polyfuse for 12V.

The parts I'm not really familiar with are Q1 and U1. I've never designed anything intended to support amps worth of current. Like can you really pass 20A through a FET? What kind of PCB layout is required for safety and low resistance? Does U1 need cooling? Etc.

EDIT: The bits about not grounding to the casing and importance of common ground are also lost on me.

 

[zigzagjoe]

I like it. I think the polarity of C3 is backwards though? Maybe add a polyfuse for 12V.

The parts I'm not really familiar with are Q1 and U1. I've never designed anything intended to support amps worth of current. Like can you really pass 20A through a FET? What kind of PCB layout is required for safety and low resistance? Does U1 need cooling? Etc.

Remember - that's kind of a block diagram. The implementation would need to vary.

• Trace widths for current capacity are a straightforward calculation with online tools.
• Polyfuse on +12v could be done
  • watch out for drive spin-up current and possibly voltage drop across the fuse however.
• Polarity of the C3 cap is correct, as + needs to go to the PSU common negative rail and - would go to the -12v rail to maintain the appropriate polarity.
  • The first image I uploaded had it the wrong way around, so maybe that showed up instead
• For 5v switching: A relay may make more sense, depends on costs
  • Benefit: you get that vintage click when turning on.
  • In a typical use case MOSFETs (often more than one) would be sized to task and the resistance can be calculated to derive cooling needs.
  • AFAIK they shouldn't need heatsinking in this use when appropriately sized as they aren't switching burst currents of 10s or 100s of amps (as when employed in a switching regulator).

 

[bigmessowires]

A couple of more thoughts on this.

I assume U1 is an integrated module of some sort, not a single IC? It's a dual-output boost regulator, and I believe it needs a switching circuit and an inductor at a minimum.

F1 probably needs to be greater than 20 amps, maybe 30 amps would work. I believe the hungriest computer in the target audience was the Mac IIfx, whose PSU was rated for +5 (18A), +12 (2.5A) and -12 (1A). If anybody knows of a Quadra with slots or other Mac II with a higher power demand, please let me know. Assuming perfect conversion to 12V with no losses, you would need 26.4A from 5V to match the IIfx PSU output on all rails.

 

[bigmessowires]

Polyfuse on +12v could be done

• watch out for drive spin-up current and possibly voltage drop across the fuse however.

Maybe just a plain 2.5A fuse then? Without anything on 12V, it would try to deliver all the power that 5V could provide, which is far more than any correctly-working Mac should be consuming on 12V. I guess it's not necessary, but could be a nice piece of extra safety.

 

[bigmessowires]

Here's an example of a U1 module that does what we need. I would not suggest buying this specific module, since it has practically zero technical information and doesn't even provide an amps rating, but it shows how such a thing could be designed: https://www.amazon.com/Power-Supply-Module-Output-Converter/dp/B09JCQ8L8C

There's a controller IC, an inductor, three diodes (I think that's what those are), and nine other small passive components.

 

[bigmessowires]

I haven't found a +/-12V boost module with the 42+ watt power rating we would need for this (2.5*12 + 1*12). The closest I found was a 30W module, which stated that it needed a heat sink if the output power is over 20W, but where would you attach a heat sink? It's a bunch of inductors and diodes, not a TO-220 with a big metal tab for heat sinking. If you were to design your own circuit, I'm unsure which components determine the max power rating and which would need cooling. I'm guessing it's a question of the maximum current rating of the inductors and the diodes.