I did an assignment at university about march-testing memory and I proved that it could find a lot of classes of faults. I used full deduction and the professor who didn't seem to know that method, gave me (and my partners) a 100% score!
Rule 0: Thou shalt check voltages. Especially when you get a fault which seems to be difficult to track down. High impedance capacitors can easily let a lot of AC noise through, and under load can cause power to drop out entirely. Glad you got to the bottom of it all!
Though, the issue might not have shown up with a simple voltage test, or even looking at static ripple. Maybe it only showed up under load, or certain patterns of load. It might have also been a transient issue.
@@big0bad0brad You would probably see it on a multimeter in AC voltage mode, at least if it's a consistent ripple. Which is usually the symptom of failing caps.
@@Stoney3K Good point, though that can be very heavily filtered and might not pick up higher frequency. It can confirm noise but it can't rule it out, unless you know exactly how your meter works.
This is what is awesome about Adrian's channels. He explains the computer science behind it. Also, the current maintainer of RDR was brilliant in helping me restore a 5150 from the trash. An Aussie legend
I can absolutely confirm that 'dirty' power can cause strange computer issues. Long time ago a roommate had his computer crashing all of the time. It was plugged directly into the wall, and once I talked him into getting a UPS his computer starting staying stable all of the time. Bad caps would cause fluctuations in the power waveforms, which I can totally see causing RAM issues in older stuff. I would guess that only those two parts were affected due to their not liking the variance in the power more than all of the other stuff.
In the 90's I did a support call to an office where the computer had completely random problems. I noticed it always seemed to coincide with the nearby elevator landing on the floor. So I ran back to the office and grabbed a battery backup and from then on the problem went away.
Grid power is so dirty where I live that when I turn off the lights and other stuff other places in the apartment, my tv brown-outs. All my devices have contracted serious stability issues.
@@asbjo That's not dirty power, that's a really dodgy electrical installation with high effective series resistance. Get an electrician out there to give it a checkup sooner rather than later, this could be a fire hazard if caused by oxidized connections somewhere (e.g. at the breaker panel). Your wiring would have to be severely undersized to cause such issues when everything is connected properly. I mean, it could still be old and crappy regardless, wouldn't be a rarity, but things like you describe just shouldn't be happening.
@@asbjo I chased an issue with a synth that sounded terrible at times without any pattern to it whatsoever. Sometimes failed when warm, sometimes when cold. I found a horrible ripple coming in at times so assumed that there was too much drain on it. The supply was the basic design with transformer, rectifier, and 7805/7812 regulators. Turned out to be that the 120VAC mains were dropping to 110VAC. Also explained a lot of other issues I've had Question everything!
@@SixOThree😂 I will always remember that damn computer that worked flawlessly at the shop but keep on crashing at our client home and corrupting the OS installation. It was the mains off course, every time the washing machine put a load on the system the lights dimed and the computer restarted. I didn't want to tell my boss the crappy PSUs he chose wasn't helping at all but he "solved" it by selling another UPS. 😅 Bad quality PSUs can make your motherboard components go bad quickly, it was an intuition for me that got confirmed and explained by PSU expert Jonny Guru at an old interview. I have seen two computers behaving differently in the same room connected to the same installation without UPS one with an OK quality PSU (Antec, Enermax...) and another with a no brand light as a feather chinese fire hazard, and any stability issues at mains would result in crashes and reboots on one of them, guess which one. 😅
Hi Adrian, apart from the bad caps in that Old PSU that were 100% causing your memory stability issues, there is also a minimum power draw per voltage rail to stabilise the voltage outputs. I would recommend replacing all the caps in that PSU as if two were bad, then most likely, all caps are starting to fail. Check the smaller 1uf, 10uf & 47uf caps. You're like the Old Ram God. Love your videos 👏
Such a "Slow Refresh" test would be interesting, that the timer value would be extended step by step until the memory chip finally gives an error. The test program could show on the map/matrix the milliseconds when each memory chip was still working. Based on the results, you could bin the memory chips and use them on a motherboard that bios allows a slow refresh rate.
@@m.wajihuddinkhan1857 Memory need to periodically refreshed to maintain their data and when this occurs CPU needs to wait. The slower you get this, the faster the system is. How slow you can set refresh depends on the individual memory chip's ability to store information.
My pet theory: The power supply issue could be the classic "swiss cheese" problem: loss of redundancy, i.e. on other motherboards the various decoupling (and other) capacitors were able to compensate for the dodgy PSU voltages. Perhaps on the troublesome motherboard the capacitors there are marginal or effectively absent, and that's enough to tip the scales.
Noisy power supplies fall in to the larger category of "Electromagnetic Compliance" problems. I remember how horrified our EMC engineers were when they saw PCs running without their metal cases. The purpose of well earthed metal enclosures is to confine stray electric fields and prevent them affecting nearby circuits. Note that the PCs you demonstrate would almost certainly fail their compliance tests without their metal cases. And it's also likely that you have some noisy equipment nearby (Air-Conditioners, Solar Inverters, Industrial power supplies) which are generating a high level of interfering hash. It goes two ways of course: If equipment is suffering from stray interference, it almost certainly is also causing interference to nearby radio sevices.
I'm not surprised in the least about the PSU causing the RAM issues. Never experienced that specific problem myself, but I've seen tons and tons of other issues that had me pulling my hair out - only to discover it was a faulty / weak PSU. As far as why one mobo would have problems while others didn't, I think it all comes down to the tolerances of the hardware. Some hardware can still function when tolerances are exceeded, others not so much. I've seen that in overclocking - same RAM brand and make, two separate packages - one can handle the OC, the other can't.
I worked as PC Repair technician for 9 years - If it wasn't the hard drive / SSD, in 80% of cases problems with the PC went away when we replaced the power supplies. At the beginning, I checked with the scope, but with time, I skipped that step and just threw in a fresh power supply when in doubt. Almost always fixed all the issues :D Interesting in that context: PCs would sometimes be rock solid under load (Benchmarks, Stress-Tests, etc.), but letting it sit idle in Windows = Bluescreen^^
Ten years ago my friend PC serviceman called me to observe one peculiar malfunction on moden laptop. When it worked on batteries, it worked fine, but on original power brick it would slow down to a crawl. It would boot like that and/or switch to that speed when it was plugged in to AC power from that adapter (simple 2-pole adapter, mind you). Battery charged fine and voltage was ok. With replacement adapter laptop worked fine. So, there was something, probably some unfiltered noise that drove that laptop crazy. Same here-who knows what noise frequency drove that PC stupid.
@@kirusyaga Yeah you tend to see this more on chargers with a 3 conductor connector to the laptop, ie a center pin and inside/outside conductors on the cylinder part
The issue is rails on some motherboards have less bypass caps being more reliant on the PSU. Plus bypass caps probably dead open tants no longer sorting out ripples. It also depends on makes of IC on the boards generating logic switching spikes!
My first thought when he said "intermittent random failures" was power supply or cold solder connections. Especially the TTL ICs can behave very strange if the voltages are fluctuating. I build my circuits with old extracted ICs and have such problems now and then. I will always test with at least two power supplies. Sometime is can't even see a difference with an oscilloscope, but one works right the other not.
I had a weird problem with a PC years ago that turned out to be caused by the PSU. When initially turned on, it would just give a BIOS beep code error complaining it had no RAM. But it would only do this sporadically, and when it did, hitting reset would get it going. I replaced the RAM but that didn't help. It seemed suspicious to me that it was only at cold boot that the problem occurred, so I thought perhaps some kind of instability until things warmed up. On a hunch I borrowed a PSU from another system and the problem disappeared. Looked inside the bad PSU and some caps were leaking. Recapped the entire thing and now it's good as new!
Two possibilities I can think of: 1. Not all caps are the same some are cheaper and don't smooth ripple as well so your boards that worked with it probably have caps with good specs while the two that didn't have crap caps. 2. Chips have their own internal capacitance as well that could effect ripple.
Excess ripple is a pesky thing at times. Since digital logic tends to have enough bypass capacitance to almost not need the bulk capacitance in the PSU, unless one is dealing with more cost optimized boards. 100nF here and there does add up, and there is usually also a few larger capacitors as well in the mix. So a half dead bulk cap for the rail in the PSU can somewhat slack on the job without much notice. Some people might argue that "SMPS switching currents would fry the bypass caps on the logic boards!", but nah... ICs takes quite hefty gulps of current each transition and they also do it far far more often. Now it is true that most logic boards do not have enough capacitance to really work without additional bulk capacitance in the power supply, since the supply voltage can end up too low for the chips to run correctly, especially when one tosses on their own ripple. Though, a surprising amount of ICs do "work" far bellow recommended input voltages, though usually not always as intended. One reason that bypass capacitance usually does an unexpectedly decent job here is mainly thanks to the practice of having bypass capacitance for all chips. Chips don't generally need their own dedicated bypass capacitors. As one moves up in the values it somewhat becomes increasingly unimportant, but that doesn't stop us from designing them in regardless. Capacitors are "cheap". So a board with a handful of ICs often ends up with more capacitance than technically needed. And some people throw on a few extra bulk ones just to save themselves from any ripple related issues down the line. The bigger issue of inadequate bulk capacitance in a switch mode supplies is the excessive EMI that now has less suppression. This can lead to all sorts of wonderful issues as far as signal integrity is concerned. If particularly bad then it can even start jamming other things nearby. But then the bypass capacitance on the logic boards likely won't suffice. In my experience, PSUs tends to be where cost cutting on capacitors happens the most, since the majority of people do like to pick the cheaper option here even if we all know better. Oftentimes a rail only has 1 bulk capacitor doing the job. Ideally speaking as far as aluminium electrolytic ones are concerned, more smaller capacitors is better than 1 large one, since the ESR per capacitor is about the same regardless of size. (not really, but 4 small 120 µF @ 0.15 ohm in parallel is better than 1 large 480 µF @ 0.08 ohm. But it is more expensive to place in 4 components than 1 large one. So cost cutting manufacturers will go for fewer components. And more smaller ones typically also have more surface area to act as a cooler, so another advantage towards reliability.)
Seems to me you do enough PC/XT/AT testing that it could be worthwhile to make a "safety wedge" for the PS... just a PCB with power input (AT and ATX), conversion circuits for the ATX such as the power switch, extra filter caps so the supply is extra smooth, a good crowbar circuit in case the PS regulation fails, and a little extra that kills a rail if the amperage exceeds an expected value too quickly (tantalum shorts, bad chips, shorted connectors, solder bridges, etc.).
Using the video RAM as a stack page is ingenious! Especially since it's easy enough to get a working video card by testing that in a working machine. Kinda reminds me of the Exidy Sorcerer. If that machine's main RAM is bad (or the circuitry connecting it to the CPU), it tries to load the OS into video RAM and you get (besides graphics glitches and crashes) a hilariously small amount of free memory.
@@PaulLoveless-Cincinnati It's a PC from 1979. Exidy was somewhat biggish into arcade machines at the time, it was kinda weird their PC design was very business-oriented, missing features that would cater to games, like color or sound, instead it had lots of RAM and (for the time) very High-Res graphics. A great machine for productivity. A poor machine for games. Still better for games than the original IBM PC though.
I believe I commented the first time around that the RAM patterns being seen almost looked like what would happen with failing capacitors, though admittedly I wasn't sure how that was possible. As for capacitor plague, the thing I've noticed about it across all of the things I've ever owned which were struck by it is that it's a gradual process and can start at any time, thus the caps in your power supply may have only recently started going bad and it was gradually getting worse to the point where it was just barely affecting those particular pieces of hardware. I had a year-2000 motherboard with faulty power filter caps for the CPU which I was using for years without issue, as it wasn't until early 2006 where the failing caps first manifested as random system crashes about once a day, but within only a few months it had gotten so bad the system would sometimes halt mid-POST or last only 10 to 15 minutes before locking up. :(
I've worked in a computer store in the 2nd half of the nineties. Two things I recall. 1. A client complaint PC crashing. Stable for hours in the store. I ended up making a house call. Sure enough the thing was super unstable. I ended up wanting to unplug something and got a shock from the case. It ended up that a desk lamp was leaking current onto the ground wire. 2. Our top line model always used asus motherboards and creative labs sound blasters. One night we are building 5 or 6. All of then failing to produce sound. Tested all of the sound cards in the showroom model, all good. Took the sound card from showroom into the new builds, all good. Somehow the incompatibility was the combination of the different production runs.
Back during the Pentium 4 era, I had an Asus P4C800-E Deluxe which had its onboard gigabit Ethernet fail after a couple years. Not long afterwards, the video card starting having problems and then the motherboard had problems starting. Replacing the power supply not only fixed the starting problem and the graphics card problem, but allowed the Ethernet to work again.
Yes Adrian, i have a home brew power supply from my bench that i used with that exact motherboard that you have as well. The same xt one from taiwan in 1985. It is from my old VIZ 708 powersupply with 5v and 12v, with long wires to the connector. I had no boots, some very odd sound issues with the speaker and even though my post card shows fine supplys it was a nightmare for me trying to repaire it. I went out of my way to get my old win98 system opened up and then connected that motherboard and with a nice AT powersupply (one that i repaired with new caps years ago) and like you say, it freakin worked!! Having an engineering background i do belive you are so correct. The powersupply must have high current, super low ripple and to be very efficient. It happened to me this week! Thanks for sharing this wonderful story.
It is good to have all these test ROMS in the toolbox. I just recently was working on a Commodore PC-10 III system (8088 system running DOS). It wouldn't post. So, I did the Ruud ROM, and it failed at the 8253-timer channel 1 and halted the test. The PC-10 uses the Faraday FE2010A chipset which has a lot of logic like the 8237, 8253 and others. If Ruud didn't stop testing because of the 8253-timer channel 1 failing it would be able to test the other sub-systems. I don't believe the timer 1 error is valid just an issue between how the FE2010A handles it versus the 8253. The Landmark ROM does have the same timer channel 1 error, but it does continue on, and it didn't diagnose my particular RAM problem. It identified bit 7 as being bad on both banks of the 256Kx1 DRAM's. Swapped those out and everything was good. Not sure if this new diagnostic rom will work fine with the PC-10 and the FE2010A chipset but when I get a chance I will have to see if it works with it. Great video as always and thanks to you and dgiller for working on it!
For PC systems with the actual 8253, timer-1 is absolutely necessary for any real testing, because it's what activates the DMA refresh cycle. Without refresh, you have no RAM. That limits what you can test, to what you can hold in the processor registers. So it makes sense that the ROM would assume that it can't do anything after that. Basically, until it creates functional RAM, complex operations are pointless.
The first thing my electronic teacher told me was to never assume the power supply in a system is ok. And if something is not working right. No matter the test, always test the power supply FIRST. The base of ant circuitry is the supply. Even power supply has a supply of AC, and it needs to be checked then the rectifier, then the filtration , then the regulation starting voltage circuit, ect, ect, ect.
Funny, when watching the beginning of the video about the fact you found the issue for both, I basically screamed : power supply!!! It seems I was right ...
Cool! I once had a laptop whose touchpad was junpoing around and it was caused by the power supply. The issue can be caused by a high frequency COMMON-MODE noise while there's no obvious voltage ripple on the osciloscope. yeah! In your case the non-working smoothing caps in your power supply meant it was pushing a lot of switching noise along the wires to the board which may have cause COMMON-MODE noise again.
I also have an original IBM-PC for sale, but we have put a 286-board in it for practical reasons and removed the back-grid, because the PC had a different distance between the slots. And my father changed the front panel a bit. Adding a 3.5 inch floppy drive for example.
Pretty sure you nailed it with the PSU. If the caps were visibly bulging and leaking, then the ripple must have been pretty bad at that point. Maybe its worth building a PSU tester. Just some load resistors on a board with a connector to load up the supply's rails and measured the rails on a scope. It might save a load of problems in the future or just simply rule out the PSU as the problem.
Awesome new ROM, nice work by David and you. Ive had weird ran issues on my Electron that ive put down to erratic power supply issues so good to see that that is a thing.
Can we just take a pause and reflect on just how cool it is that the build dates on the software being used to test this 30 year old hardware is 4 days old? (Video released on Aug 4 2024 and the build date for XTRATEST is Aug 1 2024!)
I learned my lesson early when I started building computers. The life and reliability of ANY computer STARTS with a high quality Power Supply. Even if the rest of the machine is cheap. Been there, done that, got the T-Shirt. BTW, shout out to Seasonic! Never a problem! 👍
Seasonic is actually the only brand I've had that damaged attached hardware. The main supply wouldn't come on but I was using the 5v standby supply to run my tablet. Eventually the cap bulged and it put out something like 9 volts.
@@eDoc2020 Probably because the standby 5v is a very low current supply never meant to power anything more than a logic switch. Don't blame a brand for mis-using it. However, ANY power supply can blow up. Some are just far less known to do so. And yes, depending on the age, the capacitor blight was pretty bad. I have replaced thousands of bad caps.
@@Runco990 The +5Vs rail isn't very low current, it was rated 2 amps. I don't really blame Seasonic because it's my fault for using a known faulty power supply. Though it still should have had overvoltage protection on this rail. If it was left in a seldom-used system it could have caused more damage. If you're curious it was a model SS-300FS.
@@eDoc2020 Wow that's old. Yeah, that's in the "bad cap" era. Seasonic also got a lot better since than. At least it wasn't a brand like sparklepower....
Conclusion: when Adrian can't fix something, good things happen for the community. Corollary: we should send Adrian more stuff that's difficult to fix. 😁
Power supply! I knew it! I think I even commented on one of the vids to check power. As soon as you get "weird" things happening, think power. Not necessarily the supply, also check caps on the board/card that smooth the ripple. Wouldve been nice to have seen the dodgy rail on the oscilloscope to see just how bad it was.
We have found that a pseudo random memory test will find errors that the march test doesn't. This was used on digital tape and hard disk systems and memories were too large to do march tests anyway.
About the 8288 being an essential part of the PC architecture: Intel basically intended the 8086/8088 to be always used in conjunction with the 8284 clock/ready generator and the 8288 bus controller IC. Interfacing the trio of these three chips is way easier than trying to interface the 8088 alone. I suppose Intel split the stuff into multiple chips because they ran out of pins on the 8086/8088, and didn't want to use an unusual package with more than 40 pins.
Yeah, the manufacturers probably didn’t have a large selection of 42 or 44 pin IC sockets back then, so makes sense to stick with known packages during the design/prototyping phase.
This is exactly how it was done with the old 8080 CPU, so they probably copied that scheme. The 8080 required the 8224 clock generator and divider and the 8228 bus controller and driver IC in order to work properly. Given how similar the numbers are, I'd say that the 8284 and 8288 were essentially copies of the earlier 8080 versions, just as the 8088 was essentially an (internally) 16-bit copy of the 8080.
I've seen video cards flake out with failing power supplies. Usually would show up during gaming or stress tests. So, yes, it's very possible and probable that a power supply with failed caps would certainly cause motherboards to act weird like that. Glad it didn't take the boards out for good!
I've worked on a fair few arcade machines and as soon as you said that they just started working, I started to think about the power supply / voltages & then you confirmed it at the end of the video... 🙂
They may well have smelled bad. Exploded caps often do, but the ones dating from the Capacitor Plague of the early 2000s smell even worse than normal (very fishy) which is related to the bad chemistry that makes them fail in the first place.
@@mal2ksc Yeah, and the bad chemistry was a result of failed industry espionage. The company made the chemical instructions slightly wrong as so copycats couldn't know exactly how exactly it was made, yet the copycats stole it anyway and here we are. It's like if Willy Wonka intentionally gave out the wrong chocolate formula for people like Mr Slugsworth to steal, and the chocolate ended of giving people cancer.
It would be interesting to create a device that could simulated your bad power supply and then go back and find the components in various motherboards that were sensitive to the power problems. I remember in the 90s always being told there were systems that needed pristine power or you would get problems. I never actually ran into the problem but because of what I had been told I always used high quality power supplies.
Hey, just to let you know, your XT RAM tester rom was instrumental in helping me solve a faulty XT clone motherboard. One of the TTL logic chips near the memory chips was marginal, not bad enough to show bad in the EEPROM programmer test function. Motherboard wasn't functional enough to boot normal BIOS, but functional enough to run the memory test ... where I would eventually get lots of bus errors.
The Slow Refresh test can be quite useful. Memories lose charge faster at elevated temperatures. Therefore, tests on a freshly turned on computer may not show anything. However, slow refresh tests will show that the memory may potentially have problems if it is at a higher temperature.
Your ROM obviously needs the RAM on the video card for subroutine calls, ie a stack, and other variables. Another way to do this without a video card, would be to use a serial port to spit out information about the test while it is running. Then you could use an external "terminal" to provide the display. It would be harder to implement because of no working RAM. Or use the display of a POST card to show what you are doing. My original 1982 IBM 5150 has some strange RAM problem that I am hoping to try this on. Thanks for the great work.
Tweaking voltages is sometimes used for overclocking / underclocking. Seems like the bad capacitors was introducing this state and was pushing the board into the failure zone. If the rail(s) were unstable the erratic changes might be explained by the various fluctuations.
@adriansdigitalbasement Thank you for sharing with us! Keep up the good work :) As far as the PSU goes, I would probably go ahead and replace the rest of the caps in it just to make sure you are rid of any other possible bad cap issues.
Interesting video, love the new test rom! I’ve had some power noise issues with xt ide in a 5160. The PSU is fine but not up to standard of todays. Manifested itself with random data corruption, quite intermittent but would crash before the end of 8088mph! After much analysis and troubleshooting one day I moved the card into a slot further away from PSU and it became _slighty_ better. Eventually adding a bigger filter cap on the XT ide giving power to the CF/SD cards fixed it. So I think related to tolerance of the modern components - not the same as your issues but interesting what noisy power/interference can do.
My dad's got a Tascam SX-1 in his recording studio which had its PSU recapped years ago because it broke down, now it broke down again and the repair shop couldn't get it to work. I went and replaced the SMD caps on the special PC plugin cards and it went back to working perfectly.
When you said I'll tell you later why they are working, I was suspecting the power supply. Digital is perfect until it isn't, so I have seen marginal issues plenty of times in the last few decades as a technician. The two boards may be nearly twins, but they will have some variation. The Twin board may have been on the edge of failure, but just not there yet with the flakey power. I very often swap out a power supply to rule that out when odd things are happening.
My theory is that make that other mobo had slightly better local filtering allowing it to still work despite the flakey power supply. I have myself had a psu fail once where the diodes for the 3.3V rail shorted and sent the full switchmode AC waveform into the system, the system did not just instantly die though, no it did keep running seemingly normal except that the audio was playing at like half speed for a few seconds before the system finally shut down.
I bought a battery -powered two-inch Thandar oscilloscope (single channel, 10 MHz) in year 1984 for field troubleshooting of memory problems in PC. The noisy power supply have been the cause of RAM memory problems 50% of the times. Other experts just swapped the power supply; I preferred to visualize the 100 KHz noise on the +5 Volt line. By looking at your video someone may believe that problems from the power supply are a black swan event; in reality, following the 80486 era, the power supply become the second cause of random errors and failures - the first, being the processor fan. When the PCI bus was implemented, after the short period when Local Bus dominated, power consumption from processor and video card increased ten times...
Consider that the previously weak power supply meant that the ram currently diagnosing as fail in the slow refresh cycle test was previously misbehaving so badly that the ram errors were manifesting and causing errors. Old caps compromised the ram refresh cycle. New caps, marginal chip performs within spec again.
I had something vaguely similar happen with an Xbox One my son had. Everything on the system worked flawlessly, except one day, anything plugged into the USB ports started failing intermittently. I cracked open the system and couldn't find anything wrong. I cracked open the power supply and found a bulging cap. Everything is packed in tightly in those things, so it was a bit of a trick to replace the cap, but it fixed the problem.
If the PSU was sending bad rippled 12 & 5 volts and some boards worked but other boards were flakey, that means you have some flakey caps on the board as well. Tracking them down and replacing them would fix the issue of the board.
this is such a great video and ROM idea. Xt class machines are such a pain to diagnose. I have a ...clone ish of sorts of the AST six pack that has a bad ram issue and i cant figure it out. Unfortunatly i have no way of writing a rom chip for it to try it out.
Ripple can be a real nightmare to try and pin down as a cause. That standalone battery-powered scope should be great to check for it in the future. (I really should get one myself considering I've had ripple issues on some tech myself lately.)
This is why I really like programming AVR microcontrollers. There are several chips that have 0 bytes of RAM, but they have 32 registers. If you're writing simple hardware interface applications in ASM, you can get a lot done with 32 registers. Working on 70s/80s CPUs feels sooooo cramped by comparison.
11:33 Holy crap, THAT'S what those cards are! I've got 2 of that exact card you're holding sitting in a box with a few other early AT/XT 8bit ISA cards .. I don't know what they all are, I knew 2 were video (same as the one Adrian's holding at 11:33) and the one I HAD identified is in fact a Hercules card.
It could be that the problem boards have slightly undersized or failing caps on them and working ones are better at handling the ripple or out of spec voltages. An unrelated issue that I've witnessed is when datacenters use Automatic Transfer Switches (ATS) for single PSU systems. If the caps aren't large enough, the milliseconds it takes to switch from mains to backup through an ATS can cause a system to reset. Replacing them with a "better" PSU with larger caps that can handle a few extra milliseconds of dropout would resolve the issue.
my first IBM compatible PC was a Heathkit/Zenith that included an AST memory board. it would crash about once a week or once a month. i upgraded to a 286 and kept the AST memory board, and it would crash once a month. when i upgraded to a 386, somebody wanted to buy the memory board, so i took a close look at the memory chips, and found all the chips were 256k, except one, which was a 64k chip. i replaced that chip, and the buyer of the card never had a problem. my theory before finding the 64k chip was "cosmic rays" were occasionally interacting with marginal ram chips and flipping bits (i live in Denver, so at the time it wasn't outside the realm of possibility).
A popular RAM test in the 1980s was the "barber pole". You wrote 0x01 in address 0, 0x02 in 1... 0x80 in 7, 0x00 in 8, 0x01 in 9, 0x02 in 0xA and so on. Repeating a 9 byte pattern helped detect any addressing problems since it isn't a power of 2. In the next pass you checked that the values were as expected and did the same thing with the pattern advanced by one. After 9 passes most errors would have been detected.
The black&white Macintosh clearly (as you can see from the pattern in video memory during startup) was doing a 3-bit barber pole test. Since 3 is relatively prime to all powers of 2, it is definitely going be out of sync with memory chip sizes. And it only needs three passes (six if you test the inverse too), so it was fast, which you want in a power-on memory test. I actually implemented this on a board I was helping to bring up at work back in 2001 or so, and it caught a memory problem with the board right away. (Fixing it was happily Not My Problem.)
Barber pole tests a great because many faults are something like a solder blob and can be found almost instantly. After all the simple errors are fixed run a longer sophisticated test to find internal chip faults.
The reason that most stuff might have worked, but those 2 failed before the caps were replaced, could just be down to power draw. It could be they draw more than other things you've tested, and as the Amp draw increases, so does the ripple on the power lines. And maybe they drew enough to put the power into a marginal state for those boards. Maybe those particular boards are shy of a few capacitors too, whereas others maybe had more caps on the power rails, making up for the faulty ones in the PSU.
I worked at AST, and I had that Ram card in the computer they set me up with. And it worked great. Then shortly we upgraded our PC's and it left. I did purchase one later. But could not find it when we moved. I was not happy at all.
You can tell the actual test routines aren't doing much if any stack activity as that would result in constant snow. I've used not-visible screen RAM on Z80 based systems at times so it's interesting to me to see someone else doing it too even if it's a completely different CPU.
I think this mobo (and RAM card) have some faulty decoupling capacitors which is why they are more susceptible to bad PSU, while mobos with good decoupling capacitors were able to hide the problem. You could still put an oscilloscope on top of the tantalum caps and see if some of them show sharper edges / more noise than others. Hard to say what to expect.
Ah the era of the crapacitor plague, I've been working on replacing crapacitors on a Mini-ITX system (VIA EPIA M10000 board in a Travla C134 case, cute thing, built one new about 20 years ago, got thrown out without my permission when I was trying to repair it back then!) that I got recently, proper bulged-out and crusty things on the 5v rail on both motherboard and power supply board (uses an external 12v brick with an internal board to do the other voltages), and oddly enough, it was the same issue on my original system back in the day, caps failed, computer conked out, but this one I have now I was just working on reassembling, and it's back to life as well, and apparently nobody formatted the HDD in it cos it has XP Pro on it and seems it was used for writing PIC chips and some other LED-related electronickery... :)
So here's an idea. Try to replicate those caps. Put in some small uF caps that match what you read in the test and also lift a leg and insert a very low value resistor.
I would bet the 2 motherboards have different current draws. the more current draw the more the ripple will show. even with the mem. exp. instilled it would increase the current draw making it test bad. I love his content, talks a lot about how it works better then reading it in a dry technical book lol.
Yeah. I generally just chuck any power supply I come across from that era, unless it's a specialty item. For a bench supply AT/ATX supply, that absolutely has to work and not fry stuff, I'd honestly use a modern Corsair (or similar) modular.
Not unheard of for a bad PSU to cause Windows to randomly BSOD. If experiencing random problems it is always a good idea to suspect that it could be the PSU and if you have a spare, change it. Ability to adjust DRAM Refresh timings is an odd feature of the IBM PC architecture. A feature that can help to make those amazing modern CGA demos. @PCRetroProgrammer has done a little bit of investigation into pushing the limits of dram refresh
The stack is still important on modern amd64 CPUs and you still have to setup a stack pointer if you want to run any C code on the CPU or use the call/ret instructions, although the UEFI does that for you so by the time the bootloader runs a stack is already present.
Great episode Adrian, tons of awesome information on the XT ram testing and software to use.. however, the moral of this story to ANYONE watching who may have some random - especially sudden - issues is to TEST your POWER SUPPLY if you are capable. I'm going to guess that anyone following you and who may have issues listed above does have the ability to test their caps in their power supply. We all know, however, if you're NOT competent and are unsure of what you're doing, STAY OUT of ANY power supply or CRT monitor.. seriously, you WILL get hurt.
The first thing I do when trying to fix any electronic device, regardless of what I think the issue may be, is check that the PS voltages are correct and clean.
For EGA/VGA cards, could this diagnostic be run in the Option ROM region? I saw another video of yours where you got this to work. Very handy diagnostic ROM. Excellent video too. That faulty PSU would cause issues. Good find!
Oh ATX power supplies can be so noisy, but then again so can AT power supplies. There was a period when I didn't have an original Colecovision PSU (Linear) to power my Colecovision. When I used an ATX PSU to power it the game had so much noise on the video out and the audio out. I used two big capacitors across the 12v and 5v and that cleared up the noise for the most part. That said not all switching AT and ATX PSUs are alike and it really is good to test them out first before relying on them, in particular ones with some random Chinese brand name. I made that mistake with one of my Athlon computers I built back in the day. It killed everything in my system, even the mouse!
Only thing I can guess at for why your original board was less tolerant of bad power - it looks like the second motherboard was maybe an earlier "full fat" version, where it still had the "extra" power connector, everything had sockets, etc. Your OG board maybe a later cost reduced version - often times later revs will start omitting or downgrading filtering components to save cost. Just a guess - depending on how much your curiosity/OCD demands it, would have to check /every/ cap and resistor value on both board and see what changed or was omitted. Also, while i know you're not a fan of just replacing capacitors "just because" for any Pentium 4 era PSU that hasn't already blown up yet... I'd at least pull the remaining caps out and test them! Many of those era supplies (some Bestec models i think?) were infamous for blowing up like your old AT supply did, where it killed the connected computer on it's way out!
Not a faulty PSU, but I once had a dodgy LED lamp that constantly tripped the OCP/OVP in one of the power supplies in that room. Took me ages to get that one figured out.
My AM3 board would refuse to boot if I had too much crap attached to the USB bus. I had to put all the slow stuff (like keyboard and mouse) behind a powered hub to present less of a drain, and still had problems trying to boot from flash drives in a USB 3 port if they were power-hungry. Switching them to a USB 2 port made sure they _couldn't_ get all the power they wanted, but at a huge speed penalty. I'm still running all the slow peripherals through a USB 2 hub after upgrading to a board that has _only_ USB 3 ports, and nine of them, because it's already zip tied to the desk.
I did an assignment at university about march-testing memory and I proved that it could find a lot of classes of faults. I used full deduction and the professor who didn't seem to know that method, gave me (and my partners) a 100% score!
new meaning for PSU: problem supply
Problem Supply Unit...
20 lat temu pewien znajomy powiedział mi takie prawidło, które działa do dzisiaj: jeżeli nie wiesz co powoduje problem - zawsze sprawdź zasilacz! 😎
@@mwk1 Prawda
Rule 0: Thou shalt check voltages. Especially when you get a fault which seems to be difficult to track down.
High impedance capacitors can easily let a lot of AC noise through, and under load can cause power to drop out entirely.
Glad you got to the bottom of it all!
Though, the issue might not have shown up with a simple voltage test, or even looking at static ripple.
Maybe it only showed up under load, or certain patterns of load. It might have also been a transient issue.
@@phirenz You'll probably see it on a scope but not a multimeter.
@@big0bad0brad You would probably see it on a multimeter in AC voltage mode, at least if it's a consistent ripple. Which is usually the symptom of failing caps.
@@Stoney3K Good point, though that can be very heavily filtered and might not pick up higher frequency. It can confirm noise but it can't rule it out, unless you know exactly how your meter works.
This is what is awesome about Adrian's channels. He explains the computer science behind it. Also, the current maintainer of RDR was brilliant in helping me restore a 5150 from the trash. An Aussie legend
I can absolutely confirm that 'dirty' power can cause strange computer issues. Long time ago a roommate had his computer crashing all of the time. It was plugged directly into the wall, and once I talked him into getting a UPS his computer starting staying stable all of the time. Bad caps would cause fluctuations in the power waveforms, which I can totally see causing RAM issues in older stuff. I would guess that only those two parts were affected due to their not liking the variance in the power more than all of the other stuff.
In the 90's I did a support call to an office where the computer had completely random problems. I noticed it always seemed to coincide with the nearby elevator landing on the floor. So I ran back to the office and grabbed a battery backup and from then on the problem went away.
Grid power is so dirty where I live that when I turn off the lights and other stuff other places in the apartment, my tv brown-outs.
All my devices have contracted serious stability issues.
@@asbjo That's not dirty power, that's a really dodgy electrical installation with high effective series resistance. Get an electrician out there to give it a checkup sooner rather than later, this could be a fire hazard if caused by oxidized connections somewhere (e.g. at the breaker panel). Your wiring would have to be severely undersized to cause such issues when everything is connected properly. I mean, it could still be old and crappy regardless, wouldn't be a rarity, but things like you describe just shouldn't be happening.
@@asbjo I chased an issue with a synth that sounded terrible at times without any pattern to it whatsoever. Sometimes failed when warm, sometimes when cold. I found a horrible ripple coming in at times so assumed that there was too much drain on it. The supply was the basic design with transformer, rectifier, and 7805/7812 regulators.
Turned out to be that the 120VAC mains were dropping to 110VAC. Also explained a lot of other issues I've had
Question everything!
@@SixOThree😂 I will always remember that damn computer that worked flawlessly at the shop but keep on crashing at our client home and corrupting the OS installation.
It was the mains off course, every time the washing machine put a load on the system the lights dimed and the computer restarted.
I didn't want to tell my boss the crappy PSUs he chose wasn't helping at all but he "solved" it by selling another UPS. 😅
Bad quality PSUs can make your motherboard components go bad quickly, it was an intuition for me that got confirmed and explained by PSU expert Jonny Guru at an old interview.
I have seen two computers behaving differently in the same room connected to the same installation without UPS one with an OK quality PSU (Antec, Enermax...) and another with a no brand light as a feather chinese fire hazard, and any stability issues at mains would result in crashes and reboots on one of them, guess which one. 😅
As usual thanks for coming up with new, invaluable, test Roms.
Adrian has become the go to when it comes to open source diagnostic ROMs.
Love how you're roasting how awful those other RAM tests are. Telling it how it is.
good to see you running at 100 precent aggan
Hi Adrian, apart from the bad caps in that Old PSU that were 100% causing your memory stability issues, there is also a minimum power draw per voltage rail to stabilise the voltage outputs. I would recommend replacing all the caps in that PSU as if two were bad, then most likely, all caps are starting to fail. Check the smaller 1uf, 10uf & 47uf caps. You're like the Old Ram God. Love your videos 👏
Such a "Slow Refresh" test would be interesting, that the timer value would be extended step by step until the memory chip finally gives an error. The test program could show on the map/matrix the milliseconds when each memory chip was still working. Based on the results, you could bin the memory chips and use them on a motherboard that bios allows a slow refresh rate.
What advantage would running the superior ram at a slower refresh provide tho?
@@m.wajihuddinkhan1857 Memory need to periodically refreshed to maintain their data and when this occurs CPU needs to wait. The slower you get this, the faster the system is. How slow you can set refresh depends on the individual memory chip's ability to store information.
@@m.wajihuddinkhan1857 You get very slightly higher memory throughput as more of the bus cycles are available to the CPU.
My pet theory: The power supply issue could be the classic "swiss cheese" problem: loss of redundancy, i.e. on other motherboards the various decoupling (and other) capacitors were able to compensate for the dodgy PSU voltages. Perhaps on the troublesome motherboard the capacitors there are marginal or effectively absent, and that's enough to tip the scales.
makes sense
Noisy power supplies fall in to the larger category of "Electromagnetic Compliance" problems.
I remember how horrified our EMC engineers were when they saw PCs running without their metal cases. The purpose of well earthed metal enclosures is to confine stray electric fields and prevent them affecting nearby circuits.
Note that the PCs you demonstrate would almost certainly fail their compliance tests without their metal cases. And it's also likely that you have some noisy equipment nearby (Air-Conditioners, Solar Inverters, Industrial power supplies) which are generating a high level of interfering hash. It goes two ways of course: If equipment is suffering from stray interference, it almost certainly is also causing interference to nearby radio sevices.
1:02:06 - Time to start a 'Dead Caps Box', Adrian! :)
I am so glad that we found the smoking gun, or swollen caps in this case.
I'm not surprised in the least about the PSU causing the RAM issues. Never experienced that specific problem myself, but I've seen tons and tons of other issues that had me pulling my hair out - only to discover it was a faulty / weak PSU. As far as why one mobo would have problems while others didn't, I think it all comes down to the tolerances of the hardware. Some hardware can still function when tolerances are exceeded, others not so much. I've seen that in overclocking - same RAM brand and make, two separate packages - one can handle the OC, the other can't.
"Do you suffer from short term memory loss?"
"I don't remember---"
Chumbawamba? 😀
Amnesia
To the people that replied. Yes and yes.
Am I ignorant and apathetic? I don't know and I don't care.
I worked as PC Repair technician for 9 years - If it wasn't the hard drive / SSD, in 80% of cases problems with the PC went away when we replaced the power supplies. At the beginning, I checked with the scope, but with time, I skipped that step and just threw in a fresh power supply when in doubt. Almost always fixed all the issues :D
Interesting in that context: PCs would sometimes be rock solid under load (Benchmarks, Stress-Tests, etc.), but letting it sit idle in Windows = Bluescreen^^
Ten years ago my friend PC serviceman called me to observe one peculiar malfunction on moden laptop. When it worked on batteries, it worked fine, but on original power brick it would slow down to a crawl. It would boot like that and/or switch to that speed when it was plugged in to AC power from that adapter (simple 2-pole adapter, mind you). Battery charged fine and voltage was ok. With replacement adapter laptop worked fine. So, there was something, probably some unfiltered noise that drove that laptop crazy.
Same here-who knows what noise frequency drove that PC stupid.
With broken or unrecognised power adapter some notebooks are entering emergency mode either slowing down to the snail or powering off after some time.
@@kirusyaga Yeah you tend to see this more on chargers with a 3 conductor connector to the laptop, ie a center pin and inside/outside conductors on the cylinder part
The issue is rails on some motherboards have less bypass caps being more reliant on the PSU. Plus bypass caps probably dead open tants no longer sorting out ripples. It also depends on makes of IC on the boards generating logic switching spikes!
My first thought when he said "intermittent random failures" was power supply or cold solder connections. Especially the TTL ICs can behave very strange if the voltages are fluctuating. I build my circuits with old extracted ICs and have such problems now and then. I will always test with at least two power supplies. Sometime is can't even see a difference with an oscilloscope, but one works right the other not.
I had a weird problem with a PC years ago that turned out to be caused by the PSU. When initially turned on, it would just give a BIOS beep code error complaining it had no RAM. But it would only do this sporadically, and when it did, hitting reset would get it going. I replaced the RAM but that didn't help. It seemed suspicious to me that it was only at cold boot that the problem occurred, so I thought perhaps some kind of instability until things warmed up. On a hunch I borrowed a PSU from another system and the problem disappeared. Looked inside the bad PSU and some caps were leaking. Recapped the entire thing and now it's good as new!
Two possibilities I can think of:
1. Not all caps are the same some are cheaper and don't smooth ripple as well so your boards that worked with it probably have caps with good specs while the two that didn't have crap caps.
2. Chips have their own internal capacitance as well that could effect ripple.
Excess ripple is a pesky thing at times. Since digital logic tends to have enough bypass capacitance to almost not need the bulk capacitance in the PSU, unless one is dealing with more cost optimized boards. 100nF here and there does add up, and there is usually also a few larger capacitors as well in the mix. So a half dead bulk cap for the rail in the PSU can somewhat slack on the job without much notice.
Some people might argue that "SMPS switching currents would fry the bypass caps on the logic boards!", but nah... ICs takes quite hefty gulps of current each transition and they also do it far far more often. Now it is true that most logic boards do not have enough capacitance to really work without additional bulk capacitance in the power supply, since the supply voltage can end up too low for the chips to run correctly, especially when one tosses on their own ripple. Though, a surprising amount of ICs do "work" far bellow recommended input voltages, though usually not always as intended.
One reason that bypass capacitance usually does an unexpectedly decent job here is mainly thanks to the practice of having bypass capacitance for all chips. Chips don't generally need their own dedicated bypass capacitors. As one moves up in the values it somewhat becomes increasingly unimportant, but that doesn't stop us from designing them in regardless. Capacitors are "cheap". So a board with a handful of ICs often ends up with more capacitance than technically needed. And some people throw on a few extra bulk ones just to save themselves from any ripple related issues down the line.
The bigger issue of inadequate bulk capacitance in a switch mode supplies is the excessive EMI that now has less suppression. This can lead to all sorts of wonderful issues as far as signal integrity is concerned. If particularly bad then it can even start jamming other things nearby. But then the bypass capacitance on the logic boards likely won't suffice.
In my experience, PSUs tends to be where cost cutting on capacitors happens the most, since the majority of people do like to pick the cheaper option here even if we all know better. Oftentimes a rail only has 1 bulk capacitor doing the job. Ideally speaking as far as aluminium electrolytic ones are concerned, more smaller capacitors is better than 1 large one, since the ESR per capacitor is about the same regardless of size. (not really, but 4 small 120 µF @ 0.15 ohm in parallel is better than 1 large 480 µF @ 0.08 ohm. But it is more expensive to place in 4 components than 1 large one. So cost cutting manufacturers will go for fewer components. And more smaller ones typically also have more surface area to act as a cooler, so another advantage towards reliability.)
Seems to me you do enough PC/XT/AT testing that it could be worthwhile to make a "safety wedge" for the PS... just a PCB with power input (AT and ATX), conversion circuits for the ATX such as the power switch, extra filter caps so the supply is extra smooth, a good crowbar circuit in case the PS regulation fails, and a little extra that kills a rail if the amperage exceeds an expected value too quickly (tantalum shorts, bad chips, shorted connectors, solder bridges, etc.).
Using the video RAM as a stack page is ingenious! Especially since it's easy enough to get a working video card by testing that in a working machine.
Kinda reminds me of the Exidy Sorcerer. If that machine's main RAM is bad (or the circuitry connecting it to the CPU), it tries to load the OS into video RAM and you get (besides graphics glitches and crashes) a hilariously small amount of free memory.
Is this Exidy Sorcerer a program or something?
@@PaulLoveless-Cincinnati It's a PC from 1979. Exidy was somewhat biggish into arcade machines at the time, it was kinda weird their PC design was very business-oriented, missing features that would cater to games, like color or sound, instead it had lots of RAM and (for the time) very High-Res graphics. A great machine for productivity. A poor machine for games. Still better for games than the original IBM PC though.
I believe I commented the first time around that the RAM patterns being seen almost looked like what would happen with failing capacitors, though admittedly I wasn't sure how that was possible. As for capacitor plague, the thing I've noticed about it across all of the things I've ever owned which were struck by it is that it's a gradual process and can start at any time, thus the caps in your power supply may have only recently started going bad and it was gradually getting worse to the point where it was just barely affecting those particular pieces of hardware. I had a year-2000 motherboard with faulty power filter caps for the CPU which I was using for years without issue, as it wasn't until early 2006 where the failing caps first manifested as random system crashes about once a day, but within only a few months it had gotten so bad the system would sometimes halt mid-POST or last only 10 to 15 minutes before locking up. :(
I've worked in a computer store in the 2nd half of the nineties. Two things I recall. 1. A client complaint PC crashing. Stable for hours in the store. I ended up making a house call. Sure enough the thing was super unstable. I ended up wanting to unplug something and got a shock from the case. It ended up that a desk lamp was leaking current onto the ground wire. 2. Our top line model always used asus motherboards and creative labs sound blasters. One night we are building 5 or 6. All of then failing to produce sound. Tested all of the sound cards in the showroom model, all good. Took the sound card from showroom into the new builds, all good. Somehow the incompatibility was the combination of the different production runs.
Back during the Pentium 4 era, I had an Asus P4C800-E Deluxe which had its onboard gigabit Ethernet fail after a couple years. Not long afterwards, the video card starting having problems and then the motherboard had problems starting. Replacing the power supply not only fixed the starting problem and the graphics card problem, but allowed the Ethernet to work again.
Yes Adrian, i have a home brew power supply from my bench that i used with that exact motherboard that you have as well. The same xt one from taiwan in 1985. It is from my old VIZ 708 powersupply with 5v and 12v, with long wires to the connector. I had no boots, some very odd sound issues with the speaker and even though my post card shows fine supplys it was a nightmare for me trying to repaire it. I went out of my way to get my old win98 system opened up and then connected that motherboard and with a nice AT powersupply (one that i repaired with new caps years ago) and like you say, it freakin worked!! Having an engineering background i do belive you are so correct. The powersupply must have high current, super low ripple and to be very efficient. It happened to me this week! Thanks for sharing this wonderful story.
It is good to have all these test ROMS in the toolbox. I just recently was working on a Commodore PC-10 III system (8088 system running DOS). It wouldn't post. So, I did the Ruud ROM, and it failed at the 8253-timer channel 1 and halted the test. The PC-10 uses the Faraday FE2010A chipset which has a lot of logic like the 8237, 8253 and others. If Ruud didn't stop testing because of the 8253-timer channel 1 failing it would be able to test the other sub-systems. I don't believe the timer 1 error is valid just an issue between how the FE2010A handles it versus the 8253.
The Landmark ROM does have the same timer channel 1 error, but it does continue on, and it didn't diagnose my particular RAM problem. It identified bit 7 as being bad on both banks of the 256Kx1 DRAM's. Swapped those out and everything was good.
Not sure if this new diagnostic rom will work fine with the PC-10 and the FE2010A chipset but when I get a chance I will have to see if it works with it.
Great video as always and thanks to you and dgiller for working on it!
For PC systems with the actual 8253, timer-1 is absolutely necessary for any real testing, because it's what activates the DMA refresh cycle. Without refresh, you have no RAM. That limits what you can test, to what you can hold in the processor registers. So it makes sense that the ROM would assume that it can't do anything after that. Basically, until it creates functional RAM, complex operations are pointless.
The first thing my electronic teacher told me was to never assume the power supply in a system is ok. And if something is not working right. No matter the test, always test the power supply FIRST. The base of ant circuitry is the supply. Even power supply has a supply of AC, and it needs to be checked then the rectifier, then the filtration , then the regulation starting voltage circuit, ect, ect, ect.
It took me many years to learn that lesson...
Funny, when watching the beginning of the video about the fact you found the issue for both, I basically screamed : power supply!!!
It seems I was right ...
A fascinating video, and a triumph for the new test chip.
Cool! I once had a laptop whose touchpad was junpoing around and it was caused by the power supply. The issue can be caused by a high frequency COMMON-MODE noise while there's no obvious voltage ripple on the osciloscope. yeah! In your case the non-working smoothing caps in your power supply meant it was pushing a lot of switching noise along the wires to the board which may have cause COMMON-MODE noise again.
I also have an original IBM-PC for sale, but we have put a 286-board in it for practical reasons and removed the back-grid, because the PC had a different distance between the slots. And my father changed the front panel a bit. Adding a 3.5 inch floppy drive for example.
Pretty sure you nailed it with the PSU. If the caps were visibly bulging and leaking, then the ripple must have been pretty bad at that point. Maybe its worth building a PSU tester. Just some load resistors on a board with a connector to load up the supply's rails and measured the rails on a scope. It might save a load of problems in the future or just simply rule out the PSU as the problem.
Awesome new ROM, nice work by David and you. Ive had weird ran issues on my Electron that ive put down to erratic power supply issues so good to see that that is a thing.
Can we just take a pause and reflect on just how cool it is that the build dates on the software being used to test this 30 year old hardware is 4 days old? (Video released on Aug 4 2024 and the build date for XTRATEST is Aug 1 2024!)
I learned my lesson early when I started building computers. The life and reliability of ANY computer STARTS with a high quality Power Supply.
Even if the rest of the machine is cheap. Been there, done that, got the T-Shirt. BTW, shout out to Seasonic! Never a problem! 👍
Still rocking the original Corsair HX520 - their first PSU was actually a Seasonic design. Still runs great in 2024.
Seasonic is actually the only brand I've had that damaged attached hardware. The main supply wouldn't come on but I was using the 5v standby supply to run my tablet. Eventually the cap bulged and it put out something like 9 volts.
@@eDoc2020 Probably because the standby 5v is a very low current supply never meant to power anything more than a logic switch.
Don't blame a brand for mis-using it. However, ANY power supply can blow up. Some are just far less known to do so.
And yes, depending on the age, the capacitor blight was pretty bad. I have replaced thousands of bad caps.
@@Runco990 The +5Vs rail isn't very low current, it was rated 2 amps. I don't really blame Seasonic because it's my fault for using a known faulty power supply. Though it still should have had overvoltage protection on this rail. If it was left in a seldom-used system it could have caused more damage.
If you're curious it was a model SS-300FS.
@@eDoc2020 Wow that's old. Yeah, that's in the "bad cap" era. Seasonic also got a lot better since than. At least it wasn't a brand like sparklepower....
I'm gonna guess this video has a lot to do with caffeine.
Conclusion: when Adrian can't fix something, good things happen for the community. Corollary: we should send Adrian more stuff that's difficult to fix. 😁
Power supply! I knew it! I think I even commented on one of the vids to check power.
As soon as you get "weird" things happening, think power. Not necessarily the supply, also check caps on the board/card that smooth the ripple.
Wouldve been nice to have seen the dodgy rail on the oscilloscope to see just how bad it was.
We have found that a pseudo random memory test will find errors that the march test doesn't. This was used on digital tape and hard disk systems and memories were too large to do march tests anyway.
About the 8288 being an essential part of the PC architecture: Intel basically intended the 8086/8088 to be always used in conjunction with the 8284 clock/ready generator and the 8288 bus controller IC. Interfacing the trio of these three chips is way easier than trying to interface the 8088 alone. I suppose Intel split the stuff into multiple chips because they ran out of pins on the 8086/8088, and didn't want to use an unusual package with more than 40 pins.
Which is kind of funny, because they certainly didn't have an issue with putting CPUs in unusual packages later on.
Yeah, the manufacturers probably didn’t have a large selection of 42 or 44 pin IC sockets back then, so makes sense to stick with known packages during the design/prototyping phase.
@@williamsquires3070 Not only that but the chip packaging equipment might not be capable.
This is exactly how it was done with the old 8080 CPU, so they probably copied that scheme. The 8080 required the 8224 clock generator and divider and the 8228 bus controller and driver IC in order to work properly. Given how similar the numbers are, I'd say that the 8284 and 8288 were essentially copies of the earlier 8080 versions, just as the 8088 was essentially an (internally) 16-bit copy of the 8080.
Incredible! I have learned that you should always check your power supply, but I would never have questioned an AT power supply
running the computer off the video ram is so fucking funny
It was a common trick back in the day for pirated software. The main "copy protect" code only checks the main memory.
Funny when he said it magically started working my brain said "I bet he changed the PSU" heh to many times I had PSU issues repairing computers.
I've seen video cards flake out with failing power supplies. Usually would show up during gaming or stress tests. So, yes, it's very possible and probable that a power supply with failed caps would certainly cause motherboards to act weird like that. Glad it didn't take the boards out for good!
I've worked on a fair few arcade machines and as soon as you said that they just started working, I started to think about the power supply / voltages & then you confirmed it at the end of the video... 🙂
wait wait wait, you threw out the caps rather than putting them in the dead parts bin? But the bin hungers!
They may well have smelled bad. Exploded caps often do, but the ones dating from the Capacitor Plague of the early 2000s smell even worse than normal (very fishy) which is related to the bad chemistry that makes them fail in the first place.
@@mal2ksc Yeah, and the bad chemistry was a result of failed industry espionage. The company made the chemical instructions slightly wrong as so copycats couldn't know exactly how exactly it was made, yet the copycats stole it anyway and here we are.
It's like if Willy Wonka intentionally gave out the wrong chocolate formula for people like Mr Slugsworth to steal, and the chocolate ended of giving people cancer.
Then the bin smells bad out of her mouth.
It would be interesting to create a device that could simulated your bad power supply and then go back and find the components in various motherboards that were sensitive to the power problems. I remember in the 90s always being told there were systems that needed pristine power or you would get problems. I never actually ran into the problem but because of what I had been told I always used high quality power supplies.
Hey, just to let you know, your XT RAM tester rom was instrumental in helping me solve a faulty XT clone motherboard.
One of the TTL logic chips near the memory chips was marginal, not bad enough to show bad in the EEPROM programmer test function. Motherboard wasn't functional enough to boot normal BIOS, but functional enough to run the memory test ... where I would eventually get lots of bus errors.
The Slow Refresh test can be quite useful. Memories lose charge faster at elevated temperatures. Therefore, tests on a freshly turned on computer may not show anything. However, slow refresh tests will show that the memory may potentially have problems if it is at a higher temperature.
Your ROM obviously needs the RAM on the video card for subroutine calls, ie a stack, and other variables. Another way to do this without a video card, would be to use a serial port to spit out information about the test while it is running. Then you could use an external "terminal" to provide the display. It would be harder to implement because of no working RAM. Or use the display of a POST card to show what you are doing. My original 1982 IBM 5150 has some strange RAM problem that I am hoping to try this on. Thanks for the great work.
Tweaking voltages is sometimes used for overclocking / underclocking. Seems like the bad capacitors was introducing this state and was pushing the board into the failure zone. If the rail(s) were unstable the erratic changes might be explained by the various fluctuations.
Sounds like "if all you have is a hammer everything looks like a nail" situation that can cause misdiagnosis lol
You are the only person on the Planet Earth that know how this test works. Well, have a nice day mate :)
I remember the Book 8088 was provided by "Ezra from Utah"... I still haven't found out who they are.
@adriansdigitalbasement
Thank you for sharing with us!
Keep up the good work :)
As far as the PSU goes, I would probably go ahead and replace the rest of the caps in it just to make sure you are rid of any other possible bad cap issues.
Interesting video, love the new test rom!
I’ve had some power noise issues with xt ide in a 5160. The PSU is fine but not up to standard of todays. Manifested itself with random data corruption, quite intermittent but would crash before the end of 8088mph! After much analysis and troubleshooting one day I moved the card into a slot further away from PSU and it became _slighty_ better. Eventually adding a bigger filter cap on the XT ide giving power to the CF/SD cards fixed it. So I think related to tolerance of the modern components - not the same as your issues but interesting what noisy power/interference can do.
My dad's got a Tascam SX-1 in his recording studio which had its PSU recapped years ago because it broke down, now it broke down again and the repair shop couldn't get it to work.
I went and replaced the SMD caps on the special PC plugin cards and it went back to working perfectly.
Aww, I just noticed Rammy snuggling up to the Plexus in the back 🥰
When you said I'll tell you later why they are working, I was suspecting the power supply. Digital is perfect until it isn't, so I have seen marginal issues plenty of times in the last few decades as a technician. The two boards may be nearly twins, but they will have some variation. The Twin board may have been on the edge of failure, but just not there yet with the flakey power. I very often swap out a power supply to rule that out when odd things are happening.
It makes me happy to still see the Plexus in the background. 😌
My theory is that make that other mobo had slightly better local filtering allowing it to still work despite the flakey power supply.
I have myself had a psu fail once where the diodes for the 3.3V rail shorted and sent the full switchmode AC waveform into the system, the system did not just instantly die though, no it did keep running seemingly normal except that the audio was playing at like half speed for a few seconds before the system finally shut down.
I bought a battery -powered two-inch Thandar oscilloscope (single channel, 10 MHz) in year 1984 for field troubleshooting of memory problems in PC. The noisy power supply have been the cause of RAM memory problems 50% of the times.
Other experts just swapped the power supply; I preferred to visualize the 100 KHz noise on the +5 Volt line.
By looking at your video someone may believe that problems from the power supply are a black swan event; in reality, following the 80486 era, the power supply become the second cause of random errors and failures - the first, being the processor fan.
When the PCI bus was implemented, after the short period when Local Bus dominated, power consumption from processor and video card increased ten times...
Power Supply! Like my friends at Joe's Classic Video Games say, "Power Supply, Power Supply, Power Supply!"
Consider that the previously weak power supply meant that the ram currently diagnosing as fail in the slow refresh cycle test was previously misbehaving so badly that the ram errors were manifesting and causing errors. Old caps compromised the ram refresh cycle. New caps, marginal chip performs within spec again.
I had something vaguely similar happen with an Xbox One my son had. Everything on the system worked flawlessly, except one day, anything plugged into the USB ports started failing intermittently. I cracked open the system and couldn't find anything wrong. I cracked open the power supply and found a bulging cap. Everything is packed in tightly in those things, so it was a bit of a trick to replace the cap, but it fixed the problem.
Very cool. Thanks for the new tool in the toolbox.
If the PSU was sending bad rippled 12 & 5 volts and some boards worked but other boards were flakey, that means you have some flakey caps on the board as well. Tracking them down and replacing them would fix the issue of the board.
this is such a great video and ROM idea. Xt class machines are such a pain to diagnose. I have a ...clone ish of sorts of the AST six pack that has a bad ram issue and i cant figure it out. Unfortunatly i have no way of writing a rom chip for it to try it out.
using the video card RAM to put the stack is a really smart idea
Ripple can be a real nightmare to try and pin down as a cause. That standalone battery-powered scope should be great to check for it in the future. (I really should get one myself considering I've had ripple issues on some tech myself lately.)
This is why I really like programming AVR microcontrollers. There are several chips that have 0 bytes of RAM, but they have 32 registers. If you're writing simple hardware interface applications in ASM, you can get a lot done with 32 registers.
Working on 70s/80s CPUs feels sooooo cramped by comparison.
what an emotional roller coaster!!!
11:33 Holy crap, THAT'S what those cards are! I've got 2 of that exact card you're holding sitting in a box with a few other early AT/XT 8bit ISA cards .. I don't know what they all are, I knew 2 were video (same as the one Adrian's holding at 11:33) and the one I HAD identified is in fact a Hercules card.
I kind of want this as a screensaver.
It could be that the problem boards have slightly undersized or failing caps on them and working ones are better at handling the ripple or out of spec voltages.
An unrelated issue that I've witnessed is when datacenters use Automatic Transfer Switches (ATS) for single PSU systems. If the caps aren't large enough, the milliseconds it takes to switch from mains to backup through an ATS can cause a system to reset. Replacing them with a "better" PSU with larger caps that can handle a few extra milliseconds of dropout would resolve the issue.
my first IBM compatible PC was a Heathkit/Zenith that included an AST memory board. it would crash about once a week or once a month. i upgraded to a 286 and kept the AST memory board, and it would crash once a month. when i upgraded to a 386, somebody wanted to buy the memory board, so i took a close look at the memory chips, and found all the chips were 256k, except one, which was a 64k chip. i replaced that chip, and the buyer of the card never had a problem. my theory before finding the 64k chip was "cosmic rays" were occasionally interacting with marginal ram chips and flipping bits (i live in Denver, so at the time it wasn't outside the realm of possibility).
A popular RAM test in the 1980s was the "barber pole". You wrote 0x01 in address 0, 0x02 in 1... 0x80 in 7, 0x00 in 8, 0x01 in 9, 0x02 in 0xA and so on. Repeating a 9 byte pattern helped detect any addressing problems since it isn't a power of 2. In the next pass you checked that the values were as expected and did the same thing with the pattern advanced by one. After 9 passes most errors would have been detected.
The black&white Macintosh clearly (as you can see from the pattern in video memory during startup) was doing a 3-bit barber pole test. Since 3 is relatively prime to all powers of 2, it is definitely going be out of sync with memory chip sizes. And it only needs three passes (six if you test the inverse too), so it was fast, which you want in a power-on memory test.
I actually implemented this on a board I was helping to bring up at work back in 2001 or so, and it caught a memory problem with the board right away. (Fixing it was happily Not My Problem.)
Barber pole tests a great because many faults are something like a solder blob and can be found almost instantly. After all the simple errors are fixed run a longer sophisticated test to find internal chip faults.
The reason that most stuff might have worked, but those 2 failed before the caps were replaced, could just be down to power draw. It could be they draw more than other things you've tested, and as the Amp draw increases, so does the ripple on the power lines. And maybe they drew enough to put the power into a marginal state for those boards.
Maybe those particular boards are shy of a few capacitors too, whereas others maybe had more caps on the power rails, making up for the faulty ones in the PSU.
I worked at AST, and I had that Ram card in the computer they set me up with. And it worked great. Then shortly we upgraded our PC's and it left. I did purchase one later. But could not find it when we moved. I was not happy at all.
You can tell the actual test routines aren't doing much if any stack activity as that would result in constant snow. I've used not-visible screen RAM on Z80 based systems at times so it's interesting to me to see someone else doing it too even if it's a completely different CPU.
I think this mobo (and RAM card) have some faulty decoupling capacitors which is why they are more susceptible to bad PSU, while mobos with good decoupling capacitors were able to hide the problem.
You could still put an oscilloscope on top of the tantalum caps and see if some of them show sharper edges / more noise than others. Hard to say what to expect.
RAM is a lot like printers. One of the most direct meeting point between the clean world of the binary and the messy world of the analog.
Ah the era of the crapacitor plague, I've been working on replacing crapacitors on a Mini-ITX system (VIA EPIA M10000 board in a Travla C134 case, cute thing, built one new about 20 years ago, got thrown out without my permission when I was trying to repair it back then!) that I got recently, proper bulged-out and crusty things on the 5v rail on both motherboard and power supply board (uses an external 12v brick with an internal board to do the other voltages), and oddly enough, it was the same issue on my original system back in the day, caps failed, computer conked out, but this one I have now I was just working on reassembling, and it's back to life as well, and apparently nobody formatted the HDD in it cos it has XP Pro on it and seems it was used for writing PIC chips and some other LED-related electronickery... :)
So here's an idea. Try to replicate those caps. Put in some small uF caps that match what you read in the test and also lift a leg and insert a very low value resistor.
Great job you did Adrian. Greetings from Steven from the Netherlands
I've worked as a computer technician.PSU problems often look like RAM problems, but even more random.
I would bet the 2 motherboards have different current draws. the more current draw the more the ripple will show. even with the mem. exp. instilled it would increase the current draw making it test bad. I love his content, talks a lot about how it works better then reading it in a dry technical book lol.
Yeah. I generally just chuck any power supply I come across from that era, unless it's a specialty item. For a bench supply AT/ATX supply, that absolutely has to work and not fry stuff, I'd honestly use a modern Corsair (or similar) modular.
Not unheard of for a bad PSU to cause Windows to randomly BSOD. If experiencing random problems it is always a good idea to suspect that it could be the PSU and if you have a spare, change it. Ability to adjust DRAM Refresh timings is an odd feature of the IBM PC architecture. A feature that can help to make those amazing modern CGA demos. @PCRetroProgrammer has done a little bit of investigation into pushing the limits of dram refresh
The stack is still important on modern amd64 CPUs and you still have to setup a stack pointer if you want to run any C code on the CPU or use the call/ret instructions, although the UEFI does that for you so by the time the bootloader runs a stack is already present.
Great episode Adrian, tons of awesome information on the XT ram testing and software to use.. however, the moral of this story to ANYONE watching who may have some random - especially sudden - issues is to TEST your POWER SUPPLY if you are capable. I'm going to guess that anyone following you and who may have issues listed above does have the ability to test their caps in their power supply. We all know, however, if you're NOT competent and are unsure of what you're doing, STAY OUT of ANY power supply or CRT monitor.. seriously, you WILL get hurt.
The first thing I do when trying to fix any electronic device, regardless of what I think the issue may be, is check that the PS voltages are correct and clean.
For EGA/VGA cards, could this diagnostic be run in the Option ROM region?
I saw another video of yours where you got this to work.
Very handy diagnostic ROM.
Excellent video too.
That faulty PSU would cause issues. Good find!
Oh ATX power supplies can be so noisy, but then again so can AT power supplies. There was a period when I didn't have an original Colecovision PSU (Linear) to power my Colecovision. When I used an ATX PSU to power it the game had so much noise on the video out and the audio out. I used two big capacitors across the 12v and 5v and that cleared up the noise for the most part. That said not all switching AT and ATX PSUs are alike and it really is good to test them out first before relying on them, in particular ones with some random Chinese brand name. I made that mistake with one of my Athlon computers I built back in the day. It killed everything in my system, even the mouse!
Only thing I can guess at for why your original board was less tolerant of bad power - it looks like the second motherboard was maybe an earlier "full fat" version, where it still had the "extra" power connector, everything had sockets, etc. Your OG board maybe a later cost reduced version - often times later revs will start omitting or downgrading filtering components to save cost. Just a guess - depending on how much your curiosity/OCD demands it, would have to check /every/ cap and resistor value on both board and see what changed or was omitted.
Also, while i know you're not a fan of just replacing capacitors "just because" for any Pentium 4 era PSU that hasn't already blown up yet... I'd at least pull the remaining caps out and test them! Many of those era supplies (some Bestec models i think?) were infamous for blowing up like your old AT supply did, where it killed the connected computer on it's way out!
Not a faulty PSU, but I once had a dodgy LED lamp that constantly tripped the OCP/OVP in one of the power supplies in that room. Took me ages to get that one figured out.
My AM3 board would refuse to boot if I had too much crap attached to the USB bus. I had to put all the slow stuff (like keyboard and mouse) behind a powered hub to present less of a drain, and still had problems trying to boot from flash drives in a USB 3 port if they were power-hungry. Switching them to a USB 2 port made sure they _couldn't_ get all the power they wanted, but at a huge speed penalty. I'm still running all the slow peripherals through a USB 2 hub after upgrading to a board that has _only_ USB 3 ports, and nine of them, because it's already zip tied to the desk.