This will not come as a surprise to anyone who is deeply familiar with PC hardware; the other 99% please bear with me.
A good quality and compatible power supply is crucial to the healthy operation of a PC. The catch is that whether a PSU (Power Supply Unit) is actually compatible and truly good quality may not be very apparent. Even worse, when there are problems, the symptoms may be extremely non-obvious and tend toward “analog” failures—sometimes things work, sometimes they don’t, seemingly with no rhyme or reason.
Probably my favorite retro mainboard is an Alaris Cougar, a VL-bus OEM board manufactured by IBM, also known as IBM Cobalt-AT. The board has an onboard 100 MHz IBM BL3 processor (a triple-clocked Blue Lightning 486DLC, i.e. IBM’s rocket-boosted 386), separate Socket 2 for a 5 Volt 486 SX/DX/DX2 or Pentium Overdrive, a blazingly fast Adaptec VLB IDE controller, and MR BIOS which POSTs about hundred times faster than conventional BIOS implementations (well, not really but it often feels that way, because it is easily 10 times faster).
The board is from 1993 or 1994 and of course it uses the Baby AT form factor, with a classic AT P8/P9 power connector (ever plugged those in backwards? I have…). AT power supplies are getting harder to find and their on/off switches are not suitable for bench operation. An ATX PSU with a switch on the back, combined with a simple ATX to AT adapter, does a better and usually quieter job. Except not always.
For a while I used an old ATX PSU with a 20-pin connector to feed the Cougar board and that worked pretty well. But that PSU eventually died, and I replaced it with a much newer ATX model. Some readers probably already know that that wasn’t a great idea.
The Alaris board worked with the new PSU… but not always. Only it was extremely difficult to pin down the symptoms. Sometimes the board worked just fine. Sometimes it hung during POST. Sometimes it complained that the CMOS checksum was bad. I replaced the RTC, but the errors didn’t really go away. Sometimes there were weird disk read errors when booting, in rare cases actually corrupting the disk (CF card).
I eventually determined that the behavior to some extent depended on the cards plugged in. Some VGA cards worked most of the time, others not at all. After a while I realized that plugging in a sound card with a WaveBlaster style add-on is more or less guaranteed to cause a hang during POST, even though there was nothing wrong with the card, and even though the sound card certainly wasn’t being used at all!
After mentally going through a list of things that the sound card could potentially be doing during POST, I concluded that it could not possibly be doing anything… besides needing extra power. WaveBlaster boards almost always have their own CPU and especially the older ones probably draw a non-negligible amount of power. And they might even do funny stuff like draw power on the -5V and/or -12V lines.
That’s where the evolution of ATX comes into play. The initial ATX systems in the mid to late 1990s had ISA slots and the PSU needed to supply the voltages that such machines required: +12V (hard disks), +5V (most of the system logic), -12V (serial ports, some ISA cards), and -5V (some ISA cards).
In the early 2000s, ISA slots were effectively gone. The -5V rail became entirely redundant because there was simply nothing left to use it. The ATX12V 1.2 specification from 2002 made the -5V pin optional and later it was removed from the ATX specification entirely. The upshot is that a basic ATX to AT adapter is inadequate with newer PSUs if ISA cards need -5V power.
There has been another, fairly gradual change in the ATX specification: Initially a lot of power was supplied on the 5V and 3.3V rails, but over time more and more power shifted to 12V because it’s more efficient. Modern (ATX12V 2.x) PSUs supply most of their power on the 12V rail and not a lot on the 5V rail, which means a modern PSU may not be able to supply an old board, unless it’s a really beefy PSU—because providing 500W on the 12V rail is of very little use to an AT or early ATX system.
There are several possible solutions. One is using an AT PSU, which avoids possible compatibility issues, but has other problems—an AT PSU is going to be 20+ years old, probably noisy, certainly inefficient compared to modern ones, and as mentioned above, the power switch does get in the way.
The other solution is carefully picking an ATX PSU. Finding a suitable older PSUs is not too hard, but one has to look closely. If the PSU has a -5V line, it ought to handle vintage machines using an ATX to AT adapter, and naturally it will also work with vintage ATX boards. Of course if the PSU is old, the problem is the same as above—the PSU will probably be noisy, inefficient, and may fail sooner rather than later.
The last option is buying a new “specialty” PSU such as the Advantech PS8-250ATX-ZE. That PSU provides a -5V supply and delivers most of its power on the 5V and 3.3V rails, just like old ATX PSUs did.
In the end I went for both the first and last option. I acquired an old 150W AT power supply (small form factor); 150W is more than enough for bench testing of AT boards with CF cards instead of hard disks.
The PSU works. It’s a bit noisy, and probably not terribly efficient, and the switch is a pain, but it is 100% compatible. I will probably only use it if I suspect some PSU incompatibility that a true AT PSU might prove or disprove.
I also bought the above mentioned Advantech PS8-250ATX-ZE PSU, brand new. The biggest negative: It’s not cheap (about 60 Euro all in all). But it’s new, it’s very quiet, it has a -5V line, it’s perfectly suitable for older ATX boards, and yes it does work with a simple ATX to AT adapter. It’s also quite efficient—after half an hour of feeding an ATX board with 733 MHz Pentium III, the PSU case was still cold to the touch, as if it hadn’t been on at all.
The PSU also did well on a test with the vintage Alaris board. I loaded it with a sound card with the original Wave Blaster (assuming that it has to be more of a power guzzler than newer boards) and an old Ethernet card (3Com EtherLink 3C501, again with all the NMOS logic etc. it can’t be very energy efficient). The board had no trouble booting and functioned without a glitch, going through benchmarks at the expected pace.
There may be a slight drop on the +5V line when the board is busy, especially when a socketed CPU is installed (I don’t think the onboard CPU is fully turned off). This was reported by a Kickstart IRQ diagnostic board which has LEDs for the +5V/+12/-5V/-12V lines. But this is where I’d need to pull out the noisy AT PSU again—it’s possible that the slight drop is normal on an ISA slot, and it did not cause observable problems.
At any rate, using an ATX PSU for AT boards can work, but not every ATX PSU will do.