The end of our tests, and the power supply is (mostly) performing at an excellent level! Our main concerns were voltage output regulation which it passed with no problems. I’m confident the supply is safe enough to power the motherboard and all of the peripherals. There are a few tests where we uncovered some issues, so let’s look at those in depth.
Capacitive load (+5V and +12V)
+5V voltage hold-up time
+5Vsb current delivery
+5Vsb hold-up time during power down
+5Vsb short circuit test
Both the +5V and +12V outputs were out of regulation during the capacitive load test. The +5V output had a 10,000 µF capacitor as a load, and the +12V output had a 1,000 µF capacitor. This test is really about seeing if the power supply can turn on with essentially short circuits at the outputs. Capacitors act as short circuits initially until charge is acquired, and their DC resistance increases until it is essentially an open circuit.
So the regulation issue could really just be that when the capacitors are fully charged, there is basically zero load on the power supply (other than the load I needed for the supply to start). This very light load might explain the poor regulation. I am a little confused as to why the supply wouldn’t start with just the capacitors as the load, though. It may be that the rate of current increase triggers overcurrent protection. There may be a couple of tests we can do to narrow down the cause.
+5V hold-up time
With a purely resistive load, the hold-up time looked OK. With the optical drive as a load, there was a significant drop in voltage, followed by a recovery, then followed by the falling slope.
We need to experiment some to see where this is coming from. I tried the test again with an IDE HDD as the only load, and I got a similar hold-up time as the power resistor load. It may be something peculiar with the optical drive as the only load on the supply, so I’ll disregard this for now. I’ll take another look at this once the system is fully assembled again.
+5V turn-off slope
Unlike the other voltage rails, the turn-off slope on the +5V output had two distinct portions. It dropped from the nominal +5V output to +4.4V in around 75ms, and then dropped to 0V in around 200ms. We’ll have to see how the outputs differ in the actual circuit to see why this behavior might be different.
+5Vsb current delivery
The ATX specification calls for a minimum of 720mA current capacity for the +5V standby output. If you recall, this voltage source is used for powering portions of the motherboard to allow it to respond to the soft power switch, wake-on events, and so on.
We were only able to get 250mA of current with a short-circuit. We’ll look at the circuit to see what’s going on. My guess is that it is deliberately being limited to 250mA. Dell knows what standby current is needed, so this could be where this supply differs from the ATX specification on purpose.
We know from the earlier research into how this model changed that at some point Dell included USB ports. The USB 1.0 standard allowed a maximum current draw of 500mA per port, but I’m not sure if older systems provided the +5V standby power to USB while the system was “off.” What I’m getting at is– maybe there was no need for anything over 250mA standby current in the Pro200n. I would be interested to see what power supply was used for the later model with USB, as a different supply with higher standby current might confirm this theory.
+5Vsb hold-up time
We saw an anomaly where the +5V standby output spiked to +5.8V when the AC power was removed. This shouldn’t happen, and that is outside of its regulation threshold which is a no-no. We’ll take a look at the circuit to see if something is at fault, or perhaps it is just a poorly implemented design.
Circuit design and components
We’ll briefly take a quick look at the overall layout of the PS-5201-1D and then look in detail at a few sections.
Components and layout
Let’s take a look at the PS-5201-1D and label some of the component locations:
The +5V standby supply uses a tiny transformer which we’ll look into later.
We’re mainly interested in the secondary side of the supply, so let’s focus there.
Transformers and rectifiers
Since we had a little bit of an odd result with the +5V output, we can take a look at how that particular output is generated and regulated. In order to do that, though, we’ll need to do some surgery.
The output rectifiers mounted on the heatsink shown above couldn’t be seen in the overhead photo that is labeled.
There are two ST STPS3045CP rectifiers: one for +5V and the other for +3.3V. These can each handle 15A per diode, or 30A per device. Remember the requirement of the +5V and +3.3V outputs not exceeding 140W total power draw? One of the reasons might be a temperature limitation since they’re both on the same heatsink.
+12V, -12V, and -5V
Also mounted on the heatsink, pictured above, is a Toshiba 10DL2CZ rectifier for the +12V output. This device can handle an average forward current of 10A, so plenty for our specified 6A limit.
The -12V and -5V are regulated using a pair of SEC KA7912 and KA7905 regulators respectively.
From the rectifier, the +5V output goes through the large output filter and a capacitor, and after that, another inductor and capacitor.
The two output capacitors are United Chemi-Con LXF series 3300µF 16V. They both tested with zero ESR and ~3200µF, perfectly healthy. This output is also connected to a LM339N comparator along with a 10µF capacitor which is also in great shape.
So how is the +5V output regulated? It is connected to an onsemi KA3501 supervisor IC (along with the +12V which it also monitors). One of the optoisolators mentioned earlier has its anode connected directly to the +5V output, and the cathode is connected to the feedback pin of the KA3501. The other side of the opto is connected to the primary-side PWM controller: an onsemi KA3843A. If the voltage drifts from +5V (if the load increases for instance), this will change the output of the optoisolator providing feedback to the PWM controller. It will then adjust the switching frequency to account for the change.
As for why the turn-off behavior is different compared to the 3.3V? It likely has something to do with how the +5V is used as feedback for regulation. I don’t think I’ll investigate any further though as the turn-off/hold-up behavior won’t have any impact to the performance of the supply.
I also desoldered the standby transformer used for the +5V standby voltage.
The small rectangular outline is where the standby transformer was. You can also see the three optoisolators across the slot separating the primary and secondary sides of the supply. I think the TO-220 device just to the left of the transformer outline is a switching MOSFET for the standby.
It’s not easy to see, but there is a single diode next to the electrolytic cap inside the heatsink area, so I think this may also be a simple flyback converter topology. This diode is in series with a 31µH inductor, and there are two electrolytic capacitors in parallel, one on either side of the inductor. The +5V standby voltage is tapped off of the latter side of the inductor. There is no current limiting that I can determine on the secondary side of the standby circuit, so it is being done by the PWM controller.
The voltage spike we’re seeing on shut off is most likely due to the controller losing its reference. We might be able to counteract this by increasing the capacitance after the 31µH inductor. The capacitor there is 47µF, but increasing this to say, 100µF, might do the trick.
Now that we’ve gotten a decent handle on how the power supply is designed and how it works, we can formulate a plan of action.
Despite the caps measuring OK, at least the ones I’ve desoldered so far, I think it’s a no-brainer to go ahead and replace them while the supply is disassembled. They won’t last forever, and it’s an easy and cheap way to ensure we won’t have any nasty failures in the coming years.
I’d like to run the board through an ultrasonic cleaner to get it looking brand new. There’s a bunch of old, dried up staking compound all over the board that needs to be removed as well.
The next update will include both of the above items and a quick look at the cooling fan as well.
During testing, I’ll be powering the supply through an isolation transformer for safety, and will also use a variac to test the power supply with lower and higher AC voltages. For measurement, I’ll be using a Rigol DS1102E oscilloscope as well as a BK Precision 2709B digital multimeter. This is an entry-level scope with limited capabilities, but hopefully we’ll still be able to see some interesting results and learn a little.
Computer power supplies are switch mode power supplies and that means they have an inherent danger to them when testing. I always take extra precaution when working on them as mistakes can destroy your equipment, or even severely injure or kill you.
ATX power supply design guide
Even though this is not technically an ATX power supply due to the motherboard connector differences described above, we can safely assume that Lite-On designed it to ATX specifications since we know this model was used as a basis for ATX-compatible designs.
The earliest ATX power supply specification I’ve been able to find is from 1998 and is labeled as version 0.9. These were developed by Intel as guides for manufacturers to use when designing ATX-compatible power supplies. These guides are distinct from the actual specification versions. The ATX specification itself dates back to July 1995, with version 2.01 being released in February 1997. This power supply guide is based on version 2.02. Therefore, we know version 2.02 came out between 02/97 and 09/98. That means this guide is slightly newer than the Pro200n, but it’s not clear if that has any impact on specifications that are relevant to this power supply.
Regardless of that uncertainty, we’ll use this as a basis for testing as it’s the closest we’ve got. I’d be curious to see what power supplies other Dell models from this era used, or maybe slightly earlier than the Pro200n.
I thought it’d be neat to go through and test the power supply to see if it still meets the ATX specification. I won’t be able to test everything, but the goal is to ensure the supply will safely power the Pro200n for the next decade or more.
What tests can we run?
We want to test the supply to make sure it operates (mostly) in accordance with the ATX specification, but I have also have limitations around what I can safely test here in my lab. Here’s a rundown of what I think can be tested from the ATX spec:
AC input voltage
DC voltage regulation
+5V/+3.3V power sequencing
Voltage hold-up time
+5Vsb at AC power down
Overshoot at turn-on/turn-off
Short circuit protection
All of these tests will tell us a great deal about the condition of the supply, and should any of these tests fall out of spec, the specific test that failed will give us a clue of where to look for a problem.
I won’t be testing the absolute maximum rated outputs on the supply for a few reasons. I don’t want to stress something that is vintage and original to this system. We can load it with enough resistors and/or hardware to more or less simulate what a realistic load would be. This is where an electronic load would come in handy, especially one designed for multiple voltage rails, but those are definitely outside of the budget.
A small caveat is that I will be running the tests with the cover off because it’s simply going to be more convenient for me, but it shouldn’t make a large difference in the outcome.
With all that being said, let’s jump right into it! If interested, you may want to have the power supply design guide open so you can follow along with the requirements for each test.
AC input tests
AC input voltage
The power supply has a manual switch to designate either a 115VAC or 230VAC input. I don’t have a step-up transformer, so we’ll just be focusing on the 115VAC testing.
Section 3.1 states that the minimum acceptable input voltage is 90VAC, and it should be able to start up with this voltage under peak loading. The maximum voltage (115VAC nominal) is 135VAC. I don’t have the equipment to be able to alter the AC frequency, but those minimum and maximum values are specified as well.
To test the minimum and maximum input voltage, I’ll be using a variac. This will allow me to alter the AC voltage going to the power supply. First, I’ll load the power supply with some power resistors, and maybe a fan or two on the +12V rail for an inductive load. All we’re looking for here is that the power supply starts under the minimum and maximum voltage.
I’ll be monitoring the input voltage using a multimeter at the power inlet on the supply.
Albeit not with peak load, but with a 1.65A load on the +3.3V and 1.25A on the +5V, the power supply started no problem at 90VAC input.
My variac has a slightly stepped-up winding to enable 140VAC output, so I could test the maximum required at 135VAC. It started up no problem:
I kept trying lower and lower AC voltage until I finally found that the supply wouldn’t start with anything under 56VAC. It did actually start with 56VAC though:
That being said, there was no obvious damage so I assume it passed this test, and by a large margin.
DC output tests
DC voltage regulation
The ATX specification provides a tolerance for each of the DC output voltages. Because of the nature of switch mode power supplies, the regulation should be worst at either near full load or without any load (or nearly any). The specification states the outputs should be within the tolerances specified under *all* line, load, and environmental conditions, so it doesn’t really matter either way as the supply should behave regardless.
Since I don’t have what’s needed to fully load the power supply, we’ll try using a small load on the +5V rail. I connected a 4 Ω resistor which will draw 1.25A, or 6.25W. I’ll use the oscilloscope to measure each voltage output, and will record the values below.
My oscilloscope had bandwidth limiting on, with a 1X probe setting, DC-coupled, and I was using a small ground loop rather than the alligator clip lead. This reduces the loop area and hopefully reduces the amount of EMI being picked up. I was measuring at a 4-pin Molex or ATX connector for each reading about 8″ away from the power supply itself. The average voltage of each rail was well within the specification. The min and max did fall outside of the tolerance for several rails, however, this was due to noise which was easy to see on the scope. We’ll take a closer look at that in the next test.
Output ripple / noise
For this test, the ATX specification has a few requirements:
Ripple and noise are defined as periodic or random signals over a frequency of 10 Hz to 20 MHz.
Measurements should be made with an oscilloscope with 20 MHz bandwidth.
Outputs should be bypassed at the connector with a 0.1µF ceramic disk capacitor and a 10µF electrolytic capacitor to simulate system loading.
We’ll set the oscilloscope to AC-coupled as we only want to see the ripple and noise, since we already looked at the DC voltage regulation in the previous test. This means that the oscilloscope will block DC voltages and will only show us voltage changes, which is exactly what we want to see.
To make testing easier, I’m going to use a breadboard with the specified capacitors in parallel:
I’ll record the highest peak-to-peak voltage measured by the oscilloscope for each voltage rail:
Remember to reverse the electrolytic capacitor leads in the circuit when testing the negative voltages!
The bypass capacitors make a huge difference in the output. Compare the above values to the Vmin/Vmax values in the regulation tests. This makes sense, and this is realistic to the type of loads that this power supply would see.
This is a test that measures how the supply will handle a purely capacitive load at startup. Capacitors act like short circuits when first energized, and this places a large current load on the output of the supply. Here’s what the ATX specification defines for the test:
Capacitive load (µF)
Why the heck were some of these values chosen? They are odd ball, but we can make it work. Capacitance, unlike resistance, is additive in parallel. So in other words, if we connect two 3000µF capacitors in parallel, this is equivalent to 6000µF.
I don’t have any 330 or 350µF caps, but I do have many 470µF caps. We’ll use these in lieu of the 350µF as it will be a tougher test for the supply anyway.
The setup was as follows:
Capacitive load (µF)
Definitely a harder test on some rails. Curiously, the supply still needed a “normal” load to start, so I connected an optical drive. It started successfully, but several voltages were nearly out of spec:
MEASURED (with multimeter)
Both the +12V and +5V were struggling to regulate in this scenario. This may not be a surprising result, but we’ll see when we look more into this.
+5V/+3.3V power sequencing
This is an interesting requirement that specifies the +5V output must always be equal to or greater than the +3.3V output at all times whether during startup or normal operation. Additionally, the time between both of the rails reaching their minimum in-regulation level must be less than or equal to 20ms. In other words, let’s say the +3.3V rail reaches its nominal voltage first, in order to comply with this specification, the +5V rail must also reach its nominal voltage within 20ms of the +3.3V doing so, and vice versa.
For this we’ll employ the oscilloscope again, and we’ll use two channels.
In the screenshot above, the cursors are positioned at the minimum acceptable regulated level for both voltage rails (4.75V and 3.14V respectively).
Our first requirement is that the +5V output is equal to or greater than the +3.3V at all times– and it is. The second requirement is that they both must reach their regulation within 20ms of each other– and they do (12.5ms).
Voltage hold-up time
This is a requirement that essentially states the power supply has to continue to provide stable, regulated power for at least 17ms at full load. We can’t fully load the supply, so at first I used an optical drive and confirmed that the supply far exceeds 17ms:
The +3.3V trace looks relatively normal except for the dip prior to the downward slope. I suppose the +5V behavior is due to discharging capacitors on the drive? I’m not entirely sure, so I swapped to a purely resistive load, and:
That looks mostly normal. The trigger point is at AC line cutoff, and the output persists for just over 1s. This behavior is probably different if the supply is fully loaded.
Timing / Housekeeping / Control tests
The timing of both the PWR_OK and PS_ON signals is defined in the ATX spec.
The PWR_OK (also known as Power_Good) signal is a +5V TTL signal that is used as an output signifying that the power supply is outputting regulated +5V and +3.3V. In other words, this signal should not be present when either of those voltages are outside of their regulated thresholds or when the power supply is off. The actual voltage of PWR_OK can be between +2.4V and +5V– any value between those is considered a logic level high (as in “power output is good”).
The PS_ON signal conversely is an active-low signal. Its purpose is to allow the motherboard (or any external signal) to command the power supply to enable the DC voltage outputs. So in other words, PS_ON should be at a high level when AC power is present but the power supply is “off.” The ATX spec states that there should be an internal pull-up resistor which limits current on this pin, and there should be internal debounce circuitry to prevent repeated on/off activation from a mechanical switch (all switches will “bounce” between their states rapidly for a short period of time when actuated).
So in an ATX system, the power switch is bringing this PS_ON signal low (to ground)– it is a “soft” power button meaning you’re only switching this low voltage PS_ON signal and not the full +120VAC power as you would in an older AT power supply. I’ve been shocked by the latter and it is not a pleasant experience!
OK, so looking at the timing chart, we need to define a few timings:
Power-on time — The time between when PS_ON is pulled low to when the +5V and +3.3V outputs are within their regulated thresholds. In the chart, this is between when PS_ON goes low and the end of T2 and it is defined as <500ms.
T2 — This is the risetime of the +5V and +3.3V outputs from <10% of nominal to the regulated threshold. This is defined as: 0.1ms ≤ T2 ≥ 20ms.
T3 — This is a delay for the PWR_OK signal. Once the +5V and +3.3V outputs reach their regulated threshold, there should be 100ms ≤ T3 ≥ 500ms before PWR_OK begins to rise.
T4 — When PWR_OK is brought low, there should be a delay before the +5V and +3.3V outputs drop out of their regulated thresholds. This is defined as T4 > 1ms.
T5 — The PWR_OK signal should have a rise time T5 ≤ 10ms.
With those being defined, let’s get to testing!
Power-on time and T2
I only have a 2-channel scope, so we’ll just be able to monitor one voltage rail. Since we know the +3.3V is the last to rise and achieve regulation, we’ll use that one. We’ll be able to calculate both power-on time and T2 intervals in this test.
The above shows the +3.3V output in yellow (CH1) and the PS_ON signal in blue (CH2). We’re wanting to know the period of time between when PS_ON went low (which I triggered using a jumper wire to connect PS_ON to ground) and when the +3.3V output reached regulation (+3.14V). The cursors are at the respective positions, so we need to know the delta between them which is 113ms. It needed to be less than 500ms, so this test passes.
This is only showing the +3.3V output, and we see a rise time of 8.6ms, which is between the specified 0.1ms and 20ms. I checked the +5V output and it was around 3ms, so that was in spec as well.
For this one, we’ll need to monitor the +3.3V output and the PWR_OK (also known as Power_Good) signal on the ATX connector. We’re looking for the delay between voltage regulation and PWR_OK rising.
We should have between 100ms and 500ms, and we’re at 236ms.
This is a similar test to the one above, except we’re on the other end: we want to see how long it takes for the +3.3V to drop out of its regulation threshold after PWR_OK is brought low.
So again, PS_ON is in blue and +3.3V is in yellow. We can see the +3.3V persists for 30ms after PS_ON goes low. This only needed to be above 1ms, so we’re set.
No need to test this one, as we’ve seen in the last two tests that PWR_OK is pretty much instantaneous in its rise and fall time. It just needed to be less than or equal to 10ms, so no problems here either.
Now we’ll take a quick look at the +5V standby output.
We first need to see what kind of current we can draw from this output. The specification states that it should be capable of delivering 720mA ± 5% (684mA to 756mA) at minimum, but they recommend having designs capable of delivering up to 1.5A.
For this test, we’ll use the multimeter in current mode, and we’ll first use a load of 8 Ω. This will draw 625mA (5V / 8 Ω). We won’t bother actually turning on the main voltage outputs as we’re just testing the standby supply here.
The power supply was only able to deliver 240mA. Not only that, but after shutting off the AC power, there was a very high frequency squeal that the supply was emitting as the capacitor, I’m assuming, on that rail discharged. Something doesn’t seem right, and it shouldn’t have any problems delivering higher power. We’ll have to take a look at the circuit itself to draw any more conclusions, though.
Hold-up time at AC power down
This is same test we did earlier (Voltage hold-up time), but for the standby supply. I’ll use the oscilloscope to measure this, and we’re looking for continuous +5V output for at least 17ms. The specification states it should be monotonic in nature where it drops to 0.0V without any perturbations.
The +5Vsb output lasted longer than 17ms alright… more like closer to 30s (forgot to mark the actual AC trigger point). There’s really nothing wrong with a long delay like this, but the spike before the drop is concerning. Let’s look in more detail:
That’s not great. We see the output spiking to +5.8V before dropping in a non-linear fashion. The output spikes, drops back down to +5V, and then drops like we would expect.
This is an important finding because not all devices can withstand a spike that large. Some +5V devices will only accept up to +5.5V for example. On a standby circuit, this may not present a huge issue, but it’s definitely something that needs to be investigated further.
Overshoot at turn-on / turn-off
This sounds like a complicated test in the specification, but it’s really not. They’re simply looking for a smooth ramp-up and ramp-down of voltages during power-on and power-off respectively. The first requirement is that no voltages should overshoot their nominal voltage by 10% or more when AC power is applied or lost or when PS_ON is brought low or high.
Next, there should be a smooth and continuous ramp from 10% to 90% of the final voltage for each output. The slope of the turn-on waveform should have a value between 0 V/ms and [nominal V]/ 0.1ms.
Additionally, for any 5 ms segment of the 10%-90% ramp, a straight line drawn between the end points of the waveform in that segment must have a slope ≥ [nominal V]/20ms.
We’ve actually already seen the +3.3V rise and fall slopes, and they look like they’re correct. Looking back at the +5V, we may want to take another look. We’ll also test the +12V. I won’t bother with the negative voltages.
Here’s turn on:
And turn off:
Everything looks good with the turn on rise. The turn off isn’t perfect. Instead of a smooth, continuous fall, there is an initial fall with a distinct increase in rate at +4.4V. We have very small spikes at the tail end as well, but those are of no concern. This would ideally be a smooth, consistent line without a step like that.
The +12V rise and fall was pretty much perfect.
We may need to look further into the +5V fall behavior, especially because both the +3.3V and +12V rails do not exhibit this issue.
Short circuit protection
An easy test for our last– simply short each of the positive voltage rails to see how the supply reacts. The test is specified to use an output impedance of less than 0.1 Ω. The ammeter I’m using has an internal resistance of 0.08 Ω as measured by an LCR meter. I tried the measurement at different frequencies and got about the same result, so I think it’s OK to use.
An added requirement is that the +5Vsb output is supposed to be able to be shorted indefinitely. I encountered the same maximum current draw of 240mA as we did in the +5Vsb current delivery test. After cutting the power, the same high frequency squeal was emitted from the supply.
The next post will be the summary of our testing, as well as what failed and investigations into that. Stay tuned!
Before powering up the system or doing anything else, I want to see what condition the power supply is in. The importance of the power supply is obvious: without it functioning properly, the condition of anything else in the system is a moot point. Not only do we need the appropriate voltages, we need them to be stable and clean. “Clean” meaning with as little voltage ripple and noise as possible. A power supply in a computer is responsible for generating a handful of different voltages, mostly positive but also some negative, that need to be as stable as possible to ensure the proper operation of everything else in the system.
Why is this important? When DC voltages fluctuate, this can stress components like capacitors and can introduce errors in digital circuits. Of course, no power supply is perfect and there is always some ripple and always some noise. In fact, the ATX specification actually has an acceptable range of ripple for each voltage rail that an ATX-compatible power supply supplies. We’ll be taking a closer look into this later on.
Made by “Dell”
The power supply in the Pro200n is the original. The model number is PS-5201-1D and the Dell part number is 00006081. Even though it sports a Dell logo, Dell did not (and maybe still does not?) manufacture power supplies. There were a handful of popular power supply manufacturers during this era for OEMs with one being Lite-On.
When I was doing some research on this model, I found several references to Lite-On PS-5201-## models, including a photo of one:
I also found references to Compaq, HP, Lenovo, and so on PS-5201 models. It seems likely that Lite-On licensed this model to PC manufacturers, and could customize it for them as needed.
The proprietary scourge
In Dell’s case, they opted for a proprietary layout for the motherboard power. So while this power supply is physically the same size as any other standard ATX supply, it differs in that the +3.3V outputs are located on a separate AT-style connector instead of on the main 20-pin ATX connector:
Here’s the 20-pin ATX connector (P1 above) pinout from pinouts.ru:
The P7 connector houses the +3.3V outputs along with 3 corresponding grounds.
A standard 20-pin ATX connector has three +3.3V outputs and 7 total grounds, whereas the Dell scheme has 10 total ground connections: 7 on the ATX and 3 on the 6-pin AT connector. We’ll need to delve into this deeper to see if there was any technical reason for this, or if they simply took a page out of Apple’s playbook and were attempting to force users to buy specific power supplies (ideally through Dell, I’m sure).
What’s the impact of this change? You cannot use this Dell supply in a standard ATX system without an adapter, nor can you use a standard ATX supply in any model Dell that uses this scheme. Doing so would lead to damage to the motherboard, the power supply, or maybe both.
Here are the specifications of the power supply:
PS-5201-1D Rev. 08
Combined power on +3.3V and +5V rails not to exceed 140W
Max ambient temperature
The revision of this model is “08” and after doing a little searching, the earliest I could find was listed as “02.” The latest I found was “L10.”
Interestingly, most of the other “PS-5201” models I’ve seen do not allow for as high of combined power on the +5V and +3.3V rails as the Dell supply does. This is likely due to manufacturers specifying certain requirements to Lite-On, and different revisions reflecting those requirements and as component availability changes.
Before doing anything else, I went ahead and disassembled the power supply to make sure there was no obvious physical damage.
Nothing out of the ordinary, and it looks like any other ATX power supply from the era. We’ll go through and understand more about each section of the supply and how it works later on. What’s important now is that everything visually looks intact. I don’t see any obvious capacitor leakage or bulging and no burn marks on components or the PCB.
I did see a revision marking on the PCB:
I suppose that means this is the first iteration of the model 1D PS-5201? The exterior label indicates revision 08, so maybe this means that earlier revisions all used the same internal PCB?
Is it safe to turn on?
Since we verified that everything looks visually intact, we should now see if some key components are electrically intact.
We’ll first see if the input rectifier is intact. This converts the incoming AC voltage to an unregulated DC voltage. Full-wave rectifiers like the one in this power supply are simply four diodes in a bridge configuration, and they are generally in one integrated component rather than four discrete diodes.
Component BD101 is the bridge rectifier. Because these are essentially just diodes, we can test these using a multimeter in diode mode. We should read a normal diode drop from each AC input to the positive output, and no connection from each AC input to the negative output.
That’s exactly what I found: a ~0.5V drop from AC to + and an open line from AC to -.
So our bridge is good according to our basic test. Let’s now take a look at our main filter capacitors. These ensure that our rectified DC voltage is smoothed to be as consistent as possible. Because this is a switch-mode supply, we are dealing with very high DC voltage at this stage of the circuit. If we take our standard line voltage here in the USA, we should be seeing (120V * 1.414) = ~170VDC. These capacitors have to be stout and capable of handling this high voltage. In the photo of the power supply with the cover removed, they are the large black cylinders in the bottom right.
Right now, all I want to do is make sure they are actually working as capacitors. We’ll recheck them in depth later on, but let’s simply set the multimeter to resistance mode, and see if we get an increasing resistance as the capacitor charges.
Each capacitor started out at 0 Ω (an uncharged capacitor is basically a dead short) and then quickly increased to over 5 kΩ. This tells me that the capacitors were indeed storing a charge, so at the most basic level, they are working and are not shorted (the important part).
There’s also a fuse near the +5V standby transformer that I tested, which was intact and not blown.
With those results in mind, we can try turning on the supply. There are many, many more components that could cause problems, but we know at least that the initial high voltage components are okay and this is about the extent of what we can do with a multimeter and not removing any components. Additionally, if this were even older (let’s say from the 1980s or earlier), we’d want to do even more to ensure it was safe to power up, but not applicable here.
Turning it on
Unlike older AT supplies, ATX power supplies do not use a power switch connected to the incoming AC. Instead, they use a standby circuit which waits for a signal from the motherboard (which in turn is signaled by the actual power switch) to power up. This standby circuit is constantly energized as long as the power supply is connected to AC power, and it outputs a +5V signal to the PS_ON pin of the main ATX connector (see pinout diagram above). When this signal is brought low to ground, the standby circuit then commands the rest of the power supply to start. You can read more about this on page 17 of the ATX power supply design guide.
Since the supply is out of the PC, we’ll instead use a jumper wire to connect the PS_ON output to a ground. This is a short, but that’s okay because there is an internal pull-up resistor that limits current.
But first, I checked to make sure that we had +5V output on the PS_ON pin:
The last step before power-up is to connect some sort of load. Switch-mode supplies in general do not operate well without a load of some type, though some can if designed that way. This is acceptable according to the ATX spec (section 3.5.3).
For this, I’ll simply use a resistor to provide a load on the +5V rail. I’m using an 8 Ω 50W resistor which draws 0.63A or around 3W.
I’ll monitor the +12V output with the multimeter to make sure we’ve got voltage output.
The power supply started up, and I measured +11.85V on the +12V rail. This is obviously low, but not alarming as we have such a light load on the supply. Once the power supply is driving a normal load, this regulation should improve.
Now that we’ve gotten to this point, we will move onto the testing…
After getting the system out of the closet, I took photos to document the current condition. Overall, it’s in nice shape given its age. I didn’t see anything structural that was damaged, e.g. no missing or broken plastic.
The plastic on the case as well as on the drives has yellowed over time. This is from exposure to UV light, and it can be reversed using a process called Retr0bright. There are some smudges and other marks that will need to be cleaned up as well.
A sticker on the back panel has a barcode with the text “906FX” or “90GFX,” but I think it’s the former. Perhaps this is a precursor to the Dell service tag? Searching for this string, in either form, yields nothing relevant.
Moving on, it looks like both a video card and a NIC are installed. I’m fairly certain I added this NIC at some point after I took de facto ownership of the system. We must have had a modem installed, I think originally, because this was right around the time I was heavily into BBSing, and my dad would have been using Compuserve or Nando.
I do distinctly remember my dad opting for the Iomega Zip drive when we configured and ordered the system. At the time this was seen as a convenient and inexpensive way of storing a large amount of data on removable and rewritable media– all 100MB of it.
Removing the side panel was a cinch. The panel uses a single thumbscrew, and then two clips: one at the top, and one at the bottom that you depress and then slide the panel off. There was a decent amount of dust that I immediately vacuumed out, but pretty much everything is coated in it.
I hadn’t seen SIMM memory modules in quite awhile, and it was cool to see a vertical mounting scheme for the HDD as I didn’t think that was a popular method back then.
The video card is made by Matrox. The large QFP IC reads “Powered by MGA 64-bit graphics” and has a part number of “IS-MGA-2064W-R3.” I did a quick search on this, and found an entry at the VGA Museum. There are many versions available, but I do see what look like 4 DRAM modules on the card. They are SEC KM4232W259A modules at 1 MB each, so this video card has 4 MB of RAM. There’s also a code next to the Matrox silkscreen at the top of the PCB that reads “590-05,” and this corresponds with the “Matrox MGA Millennium 4MB IBM” listed at the VGA Museum. We’ll dive into the video card some more later on, but we now know that it is a Matrox MGA Millennium 4MB video card.
The only other card is a 3COM PCI NIC. I’m fairly certain this did not come with the system, so I’ll need to verify that later on.
The front panel was removed by pressing three levers along the right side of the panel. With that removed, I could take out the floppy drive.
The floppy drive: a Sony MPF920-F 3.5″. It looks pristine in the above photo, but it sustained some accidental damage the next day. More on this later as well.
The hard drive is an IBM DCAA-34330 4.3GB IDE.
The optical drive and the Zip drive were housed in this removable caddy. Removing this made the structure of the case a little flimsy, so I’m not sure why this was meant to be removable. Maybe removing the other side panel is more of a pain than I’m anticipating…
The CD-ROM drive is a NEC CDR-1600A. At the bottom, the manufacture date is listed as April 1997, so we know the system cannot be older than that assuming this drive was never upgraded or changed.
And finally, the Iomega Zip drive that was touched on earlier.
We’ll get to the motherboard and other interesting bits later on.
The original configuration
Before going any further, it might be helpful to have something to compare the current hardware setup to, so we can determine if this hardware is actually original to the system or not.
The problem, of course, is that we’re dealing with hardware from the mid-to-late 90s. There’s no mention of this system on the Dell website in any sort of archive, so instead we’ll have to rely on archived websites (if they exist) and any print media that might give us clues.
I know this system was purchased between 1995 and 1997. How? Because that’s about roughly when the Pentium Pro was manufactured. We can further narrow that down from the CD-ROM drive manufacture date of April 1997, so it’s likely on or after that date. With those dates in mind, we can look at archived versions of Dell.com:
The earliest archived version of the site is from Dec 21, 1996 on the Wayback Machine. Clicking on Dimension Desktops, they have configurations listed for both Home Office and Small Business. Let’s try Home Office:
Dimension XPS Pro200n
200MHz Pentium® Pro Processor
Mini Tower Model
32MB EDO Memory
256KB Internal L2 Cache
3.2GB Hard Drive [9.5ms]
20TD Trinitron Monitor (19.0" v.i.s., .26dp, 1600 X 1200 max. res.)
Matrox Millennium 4MB WRAM Video Card
NEW 12X EIDE CD-ROM Drive
AWE32 Wave Table Sound Card
Altec ACS-490 Full Dolby Surround Sound Speakers w/ Subwoofer
33.6 U.S. Robotics Telephony Modem
MS Office Professional with Bookshelf for Windows 95®
MS Office 97®, Small Business Edition Upgrade Coupon
MS Windows 95/MS Plus!® CD/30 Days Free Support
3 Year Limited Warranty with 1 Year On-Site Service
Price Configure Buy Product Code #501119
What a powerhouse. How about the Small Business configuration?
Dimension XPS Pro200n
200MHz PENTIUM® PRO PROCESSOR
Mini Tower Model
64MB EDO Memory with ECC
256KB Internal L2 Cache
3.2GB Hard Drive [9.5ms]
20TD Trinitron Monitor (19.0" v.i.s., .26dp, 1600 X 1200 max. res.)
Imagine 128 Series 2 Graphics Accelerator with 4MB VRAM
NEW 12X EIDE CD-ROM Drive
MS Office Professional with Bookshelf for Windows 95
Microsoft Windows NT Workstation 4.0/30 Days Free Support/
3 Year Limited Warranty with 1 Year On-Site Service
Price Configure Buy Product Code #501110
$4,299 is equivalent to $8,000 in today’s money. That’s insane. These are the highest configurations I pulled from the archived site. They did actually have a configurator linked, but the server-side dependencies are long gone. On a side note, the other, cheaper configuration listed for the Small Business category used a SCSI HDD and CD-ROM drive. Both were smaller in capacity and slower in speed, respectively, to the higher priced config.
Moving forward in time, let’s see when the Pro200n was no longer listed on their site. Well, the very next archive from June 5, 1997 no longer lists the Pro200n or any Pentium Pro desktops as available configurations. This makes sense, because processor speeds and technology in general were rapidly advancing at the time.
To track this down further, let’s look for advertisements. I assumed they would be advertising the system right up until the point it was no longer available. I hopped over to https://books.google.com and searched for “pro200n.”
I found a ton of results, and thankfully, a bunch of scanned and archived PC Mag issues. The latest advertisement I could find with system specs was from the June 24, 1997 edition:
Some of the specs match up, but this system does not have two USB ports, so it must be an earlier configuration.
An advertisement in the April 22, 1997 edition looks like it matches up except for the CD-ROM drive. Let’s look a bit later…
This matches my system except for the hard drive capacity. So it’s likely the system was purchased somewhere in April or May 1997, but no later since Dell added USB ports as standard some time in June. I know Dell offered options when ordering a PC, so it’s entirely possible that my dad opted for a smaller hard drive, or upgraded to the faster CD-ROM at the time.
No NIC is listed in any configuration, so that confirms I installed this later on. We’ll disregard it for now.
I want to retain as much of the original system as possible, but also keep it true to Dell’s spec. That means we’ll work off of the above advertisements to get the system up to its most likely original configuration. We know, for instance, that there is no modem or sound card. The sound card was standard, so that’s definitely missing. The modem was an option, and I know the system had one at some point, so we’ll make the assumption it had the modem listed in the advert.
So here’s a run-through of what I have in mind:
Evaluate the condition of all components
Dust removal and clean-up
Repair and refurbish all components as needed
Replace missing or non-original components
Test and first power-on
Bask in 90s glory
Much of what I plan to do isn’t necessary for a running system. Hell, it probably runs just fine right now. I’m sure I could power it up and expect a working system. The point of this whole exercise though is the journey. We’ll be getting down to the most minute details in this restoration, and hopefully learning lots of history, technical and engineering knowledge, and maybe even discover a few surprises along the way.
By the time I was forming memories as a young child, we always had a computer in the house. My dad was a software and database engineer; he worked in several industries, including in the aerospace field where he wrote software for Pratt & Whitney rocket engines. The first computer I remember was a Tandy TRS-80 Model II with a Motorola 68k CISC and two 8″ floppy drives. I have memories of seeing 8″ floppy disc mailers arrive in the mail, and I was fascinated by the idea that information could be conveyed by something that had no discernible features on it. Most of my early computer exposure was through observation: I would watch my dad and my older brother type on a keyboard and interact with the computer.
Our first real family computer was an Intel i386DX 33 MHz PC that we got for Christmas one year. It was built by a local computer store called One Step Computers located in MacGregor Village in Cary, NC. I have a distinct memory of walking down the stairs that Christmas morning, and seeing the brilliant, vivid colors on the CRT monitor as it sat on the floor waiting for my siblings and I to discover it. A fish screensaver on Windows 3.1 was on, and my sister and I stared in amazement. That was my first real “wow” moment at what computers could do– as a kid, you’re not impressed by rocket engine software, but by a crude representation of fish.
I was pretty much hooked at that point. I lived and breathed as much as I could about computers from that point forward. My dad was still involved in software engineering, and a few years later he needed an upgrade, and then another after that. Eventually, one came in the form of a top-of-the-line Dell Dimension XPS Pro200n. This was the absolute pinnacle of home PCs at the time, and it sported a 200 MHz Intel Pentium Pro CPU. For awhile, this was strictly his computer, and we weren’t allowed to use it.
Those were exciting, accelerating times though. It wasn’t long before he needed *another* upgrade, and I finally got my hands on the Pentium Pro I had been longing for. Even for me though, this new found speed didn’t last long, and I eventually moved onto newer and faster hardware.
I don’t know why, but we ended up hanging onto the Dell and for whatever reason, never quite decided to part ways with it. It persisted through so many moves, rearrangements, and so on. Even during attempts to reduce clutter and throw the old things out, the Dell remained. There’s so much from the 90s I wish I had kept: *so* much computer hardware and software, music CDs, other electronics, the list is huge. Fast forward decades later, and in my sixth or seventh move, the Dell is still here.
I love nostalgia, and especially during these times of climate change, a pandemic, massive inequality, and so on– the experience of being an ignorant, naive kid during the 90s is something I mentally visit from time to time as a respite. I would’ve LOLed at the thought of that had you told me during the time, of course.
When I saw the Dell sitting in my office closet a few weeks ago though, I realized it was probably the greatest link I still have to that era. A crazy idea popped into my head: what if I restored the system back to its original glory? I’m sure the system would probably run just fine again, but what if I restored the system to absolutely new condition as an homage to that era of computing? I decided I would keep the Dell most likely forever, and I want it to be in working condition as long as possible. As an adult you often realize the importance of history, and of preserving that history. When we lose or forget history, no matter how insignificant it may seem, we lose a piece of our collective human journey through this absurd universe.
So my plan is to document what I learn about the system, the how and why of restoration, and maybe relive some of the 90s through that process. I have no real end goal other than to end up with the computer as it would’ve been when it was new. Some parts of the restoration may be ridiculously in-depth, and maybe even seem pointless to some, but ultimately it’s not really about the end goal. It’s about the process, and what I (or we) learn along the way.
My Brother HL-L2360DW laser printer went into a reboot loop after a power outage. The LCD initializes and displays solid blocks, followed by a blank screen, and the Wifi button LED blinks once, and then the cycle repeats. Holding the “GO” button either during startup or while it’s looping puts the printer into “Users Mode” which displayed on the LCD. The service manual for this particular model does not mention this mode, but I found another service manual that did. I tried every possible option, but always ended up in the same place.
I removed the access panels on the printer and quickly checked voltages from the low-voltage power supply. I didn’t suspect a problem here, and I didn’t find one… all the voltages measured OK, but that’s always the first step. I then turned to the main PCB (p/n B512386-4) as the issue was almost definitely there. I tried disconnecting all external connections: the HV power supply, laser unit, motor, etc but this had no effect on the issue.
The service manual has a simplified block diagram showing the main PCB components and their connections. There is a Renasas ARM MPU in a 144-pin QFP package that serves as the main controller. This is powered by a 3.3V rail that is generated by a DC-DC converter. This voltage checked OK, as expected. The MPU uses I2C to communicate with an EEPROM as well as an IC which is labeled “HYPNOS” in the diagram. Among other things, the HYPNOS appears to be responsible for sending the AC zero crossing signal to the MPU which is probably for controlling a triac or SCR. I checked the SCL and SDA lines on my scope, and confirmed that the MPU is sending clock as well as data. Entering into “Users Mode” immediately ceases I2C communication, and it doesn’t begin again until the machine cycles.
The MPU also uses SPI to communicate with a serial flash IC (a cFeon QH64) labeled U7 on the board, which is likely used for the firmware. I was able to see SPI clock and data on the scope, so it seemed like the main MPU was intact.
I don’t think the EEPROM (a ST M24C32-W) stores anything absolutely necessary for startup and some of the parameters can actually be set in a maintenance mode according to the service manual. There are also a number of error codes related to EEPROM, DRAM, and flash ROM failure. I haven’t received any error codes, though. At this point I was thinking that the firmware was most likely corrupted given the behavior of the system.
I knew I would need a replacement board due to the current state of repair parts and software availability. I was unable to find an exact replacement board, but I was able to find a board for a Dell E310DW which is essentially a re-badged HL-L2360DW. The part number for the board differs by one digit (B512386-5), and it visually looks identical.
Since a replacement was on the way, I decided to go a bit further to see if I could figure out exactly what was at fault. I desoldered the EEPROM (U1) and there was no change in the behavior. I desoldered the QH64 flash IC (U7), and bam– no output on the LCD. Does this mean that the firmware is actually intact and that the issue is with the EEPROM? I soldered the QH64 back in, and was back to where I started. Observing the I2C and SPI lines, it is clear that the system is simply looping. It attempts to boot, something fails, it restarts, repeat.
Interestingly, the previously mentioned HYPNOS has an output pin labeled “CPURST” as in “CPU reset.” There were several ICs on the board that I was unable to find any datasheets for, so at first I wasn’t sure exactly which one was the HYPNOS. The block diagram showed 7 pinouts, one of which was tied to the AC signal for the zero crossing detection. I was able to trace the pin header for the AC signal from the power supply, and eventually found a 20-pin SMD that looked like it might be the one. I probed around with the scope, and found I2C communication on one of the pins, so I’m fairly confident it was the right one. The IC itself is labeled “510A” with “DN5” on the second line, but otherwise no other identification. I desoldered it, and the system would not boot, so nothing gained there.
I received and installed the replacement board, and the printer is back up and running, albeit as a Dell now:
This at least confirmed what I already knew. The printer was intact other than something on the main PCB. On closer inspection, the Dell board is not quite identical to the Brother, as the flash IC is different. It is a GigaDevice 25Q64CSIG and there is a 10k pull-up resistor on the CS pin which is absent on the original board. The EEPROM is different as well. This new board uses a slower SPI clock at 25MHz, compared to 50MHz for the QH64. I’m thinking that the Dell board might be a slightly newer revision given the part number. I wonder if Brother had issues with the older design given they left CS floating.
I de-soldered the flash IC and whacked it onto a breakout board. I hooked it up to a Bus Pirate:
I used spiflash to extract a .bin of the firmware. binwalk showed that the firmware consists of ARMEB instructions, which is ARM’s “old” application binary interface (ABI) that uses big-endian:
strings output showed an entry for “Users Mode” so it appears that this is indeed part of the firmware. Hmm… does this mean the firmware was actually intact all along? For grins and giggles, I also found entries for “Dell Printer E310dw.” I suppose I could change the name of the printer back to the original Brother model if I wanted… or something entirely custom!
I went ahead and wrote the .bin from the Q64 from the Dell board to the QH64 from the Brother board.
That completed successfully, so I re-soldered the QH64 back onto the original Brother board. I reinstalled the board and turned it on.
Success! Sort of…
The printer booted into maintenance mode. I checked the service manual for the different commands, and printed a test page using command “09”. It worked! With that success, I chose command “01” which automatically set the EEPROM parameters. I then printed a “maintenance information” report using command “77” which confirmed that this Dell firmware is indeed at least newer than the version shown in the service manual:
The “SW CheckSum” shows “NG” which I assume means “not good,” but I’m not sure what this really indicates or what it is calculating a checksum for. After exiting maintenance mode, everything seemed okay. Unfortunately, I then noticed there was no network option in the main menu, and the Wifi button did not respond.
The board uses a Realtek PHY, and after checking the IO line to the MCU, it was clear that something was going awry and the MCU was choosing to shut down the PHY. I went back into maintenance mode and selected option “80” which displays machine log information including the MAC address. No MAC address was listed, so that was almost definitely the issue.
The service manual indicated that the MAC address is stored on the EEPROM, but it is not an initialized parameter. In other words, it is pre-defined on the EEPROM and there’s no way to update or change it from the printer itself:
So, I de-soldered the EEPROM again in hopes of seeing if I could either manually edit the MAC address, or perhaps dump the contents of the Dell board EEPROM onto this one. During this process, the original EEPROM died. It developed an internal short which killed the chip when I was trying to read it. I probably exposed it to too much handling and heat.
I slapped the Dell board back in and the printer (and networking) works fine of course. At some point I may order another EEPROM IC and try the above read & write sequence. For now though, I’m done with this project and mostly accomplished what I set out to do: repair the printer, and determine what went wrong with the original. I think both the firmware and EEPROM were corrupted during the power loss / surge event.
A friend of mine designed a simple phono preamp for me to use with my newly purchased turntable. Once he was done with the design, I used EAGLE to design PCB layouts for both the power supply and amplifier boards. They were fabricated by OSH Park.
The power supply outputs 27V and uses a LM317LZ regulator. The amp section uses 2N5088 transistors.
The build was relatively unremarkable. I used an aluminum Hammond chassis, and point-to-point wired everything before, between, and after the PCBs.
It sounds great and has been trouble-free for years!
I bought a Fisher CA-272 100w/ch amplifier that had very noisy controls. Specifically, the amplifier had a built-in EQ and the sliders were deteriorated to the point of being unrepairable. Instead of ditching the amp, which sounded relatively good, I decided to transplant it into a new chassis.
The cool thing about this amplifier was the beautiful pair of heatsinks for the output transistors. I decided to show those off a bit in the new chassis. Since I was ditching the front panel, I had some power to play with as well.
I replaced any capacitors in the audio path with Nichicon KL series as these are known in audiophile circles to have the best performance in these applications.
For the volume potentiometer, I used an ALPS “Blue Velvet” RK27112A00CC.
For the chassis I decided on a steel Hammond 1441. The power transformer of this amp was pretty hefty, so I needed a relatively stout panel to mount it. Also, working with steel is a bit easier than aluminum when it comes to machining.
After drilling holes for the board mounts, transformer, RCA jacks, and power cord, I mounted everything in place.
To show off the heatsinks, I cut rectangular slots in the top removable panel using a dremel.
The front panel of the original amplifier was fed using +17.2VDC, so I used this source for LED lighting. I decided to illuminate the heatsinks that would be protruding through the top panel. Eight amber LEDs were connected in series along with a 470 Ω current-limiting resistor. The array draws about 30mA.
To position the board and heatsinks at the correct level, I used Keystone standoffs to offset the board from the bottom of the chassis.
Putting it all together
I was pleased with the end result, but I wish the lighting was a bit more diffuse. At some point I plan on addressing this. I’d also like to clean up the edges of the heatsink cutouts using grommet edging.
The Compaq Portable was one of the world’s first 100% IBM PC compatible computers, and definitely one of the first real “portable” computers. It’s not really portable by today’s standards as it is nearly 30lbs, but in 1983, it was groundbreaking. A co-worker posted one for sale on a company bulletin board as non-working. He bought it new when it was first released, and hadn’t powered it on until just recently. When he did, there was apparently a loud noise and smoke followed.
Working on old irreplaceable equipment is always a little daunting. Parts availability can be questionable or non-existent, and it’s easy to get in over your head on how far a restoration should go. In this case, I decided that I would restore the system back to working order, but would not go as far as a complete cosmetic restoration. I love vintage hardware, but there are others much, much more dedicated and knowledgeable to preserving these historical artifacts than I am.
I disassembled the entirety of the unit including the CRT. The issue that the previous owner ran into was easy to spot: a tantalum capacitor had shorted and exploded on the power supply board.
After cleaning the surrounding area, it appeared that no other components were damaged, at least from what I could tell visually. Given tantalum capacitors’ propensity to fail in this manner, and the age of the system, I decided it would be justifiable to replace all of them.
Additionally, the keyboard used capacitive foam & foil discs that had long since disintegrated. Luckily, many other vintage systems used similar keyboards (made by Keytronic) so replacements are available.
Replacing the tantalum capacitors was straightforward. Before replacing the keyboard foil discs, I slowly powered the system on using a variac and dim bulb tester. After ensuring voltages were at the expected values, I slowly increased the variac to maximum voltage. The system booted!
Since the system was now working, I went to work on replacing the keyboard discs. This wasn’t a fun or interesting job, but about an hour later, I had a working, usable system on hand.
The 5.25″ drives seem to have a little trouble reading the included MS-DOS (version 1.11!) diskettes. I did not attempt any further repair, as I had mostly accomplished what I set out to do, and I did not want to risk damaging these diskettes.