Seeing this old equipment calls to mind the time a colleague told me about working for a semiconductor manufacturer in the 70s. He ended up inheriting someone else’s logic diagnostic routine that would get stuck in an endless loop until a manual break command was issued. Anyway, the micro code (and silicon) was so primitive that it didn’t have an exception built into the code in case of a divide overflow, so of course every time the routine would run it would get stuck in an endless cycle trying to divide by zero, catastrophically overheat in the span of a few milliseconds, and destroy itself (and the ancient machine it was running in) spectacularly by blowing the lid right off the chip leaving a little charred cavity presumably where the ALU was. Took a while to fix that bug but he figured it out and everyone went on to live full and somewhat meaningful lives.
the fix probably was : check if the divisor is not 0. go live your life, but you wouldn't believe how hard is to make programmers check things (they'll excuse "performance", as if a single check would matter).
Basically, anything above 3.7V is out of range of lx8383a because dropout 1.3V, MB's pdf says 4.0V for cyrix, it is possible only for low-dropout regulators with dropout 1.0V
Yeah, that's true. The dropout is extremely hard on the edge, for LX8383A it is typically at 1V, but I think, even for Cyrix I would use it rather at 3.6V. It usually works just fine if you don't overclock it.
Still have a soft spot for late 486 motherboards - while all of my friends at LAN parties could afford pentiums, a very similar motherboard served me well enough for years with a Cyrix 5x86-100 overclocked at 120MHz.
I had 486DX2/90 when friends bought new and shiny P60 machines. With my S3 Vision864 VLB and bus speed @45MHz it run Duke Nukem 3D better that friends P60 😆 Of course Quake was totally different story...
@@djpirtu2 There was an official DX2/90? 80 and 100, yes -- the DX2/100 was short-lived because very few VLB cards were happy at a 50 MHz bus speed, but the DX2/80 did fine because almost all cards that were specced for 33 MHz also ran at 40, and it was only the VLB card(s) you had to worry about (assuming the ISA clock divider was changed from CLK/4 to CLK/5).
@@mal2ksc No not official. If I remember right the CPU was Am486DX2/66 and I just overclocked it to 90MHz by swapping oscillator. Never had any problems with VLB cards at 50MHz. Now I have one Am5x86 system @150MHz so 3x50MHz, and everything in BIOS is set to tightest, Cirrus 5434 VLB flies on it. 🔥
For those who may be confused by why the resistor values did not fit: Some regulators have the 1.25 V reference voltage between output and ADJ pin, like the LX8383A, which follows the classic LM317 school. Other like the SHARP regulator (it seems to be a Japanese thing) went to 1.25 V between ground and ADJ pin, which also means they have an explicit ground connection at the IC. In any case trying to swap one for the other will result in the roles of R1 and R2 effectively being reversed, so the voltage output ends up wrong as observed. This is why swapping two resistors worked for the 3.45 V setting.
It's really true what a scholar once said, if you can't handle the fatigue of learning, you'll have to endure the pain of ignorance later. Like me, if I hadn't wasted time in school, maybe now I'd be making cool content like you.
Hi, great video as usual. Just small note: that voltage regulator with higher current capability will be heated to same temperature as that other, because same amount of heat needs to be dissipated. 🙂👍
Hello, clueless person here, so in order for that to occur you would need to increase the surface of the component or use more than one component so the heat is distributed? 😅
@@RetroTinkerer Hi, yes, it depends on temperature resistance from chip to package, from package to ambient air... so yes, using cooler, or other package can enlarge that surface (that is, why heatsinks are ribbed, to make bigger surface with contact to ambient air) and make it not so warm. 🙂
Also, it is exceptionally rare that a part won't work _because_ you put a heatsink on it. It's not impossible, but hardly worth worrying about. I think I'd try to mount it (and its wraparound heatsink as shown in the website photos) in a vertical position rather than parallel to the board and screwed to it, if it's really getting that hot, or maybe still place it parallel but as high above the board as its little legs will allow so it gets some cooling from the back. The ground plane it was bolted to does distribute heat, but then it hangs onto it because it wasn't designed for radiating heat, it was designed to be a ground plane. (It's one of the two inside layers of a typical four-layer board, the other is generally power.)
In essence, you've taken a "parts board" for someone else, and made it into a "working board" for you! It's amazing to take something where most people back in the day would just want to rip the guts out of this motherboard to fix a barely broken motherboard, and instead actually fix it!
One generation after this, the Pentium 1, ignited my love for PCs, back when I was just a elementary school kid, so nice to see such old hardware being well cared and kept alive
Omg you are hardware genius, God. So much advanced stuff people don't know. It is pleasure to watch this old pc parts get restored and works improved. Just simply amazing
This was really satisfying to watch and a nice rebuild, a lot of effort put in but really well explained too - great detail! Makes me think I should resurrect some of my junk pile.
This is the first video of yours I've seen! I really enjoyed how you explained the changes you were making! It was very clear and enjoyable to follow along with 😃
As always good work. I have 2 soyo's to bring back to life. Apparently it would just be changing the swollen capacitors, but no. Socket 370, which was not booting, now heats up the processor and stops reading memory. The other socket 7 that sometimes booted and crashed, after I changed the capacitors the processor heats up and does not leave FFFF.
oh, WOW! I had exactly such a motherboard from SOYO! And I loved it for overclocking it with some AMD 5x86DX133 and clock of 50 MHz BUS (no, not 25 MHz, but real 50 MHz!) and multiplier of 3... This beast outruled every Pentium up to Pentium 90 including! 😀
Yes. Specially, it's a Schottky diode, and will have a significantly lower forward voltage drop than a regular silicon junction diode. Current and reverse voltage capabilities are fine for this application.
A note about the regulator: A linear regulator is basically a variable resistor that turns all the excess voltage into heat. This means how hot a linear regulator gets is never related to how powerful it is, but rather directly to how much voltage it's dropping. if it's dropping 5V down to 3.45V, it will be dropping 1.55V, which means that at its limit of 7.5A, it will have to dissipate 11.6W of power, and at 3A any regulator is going to dissipate 4.65W of power regardless of their rated maximum (give or take a little bit depending on efficiency). This is different in switching regulators, which turn the input on and off for different periods (PWM) and then smooth out the output, meaning there's very little excess power at the cost of having a noisy output that needs very good filtering.
(Wrote this comment because I didn't see anything related to it, and then just after sending, I see that just a little further down than I scrolled, someone already mentioned it)
8:16 all linear regulators emit the same exact amount of heat at a given input and output voltage! The output current rating is only the rating for load transient maximum current. Real current through the regulator is the current consumed by the load (use average value), and the dropped voltage is the voltage difference between input and output voltage. So if you have a 1A load, a 3.3V regulator running off 5V will always drop 1.7V and will always consume and convert to heat 1.7W of power. Higher current rating generally implies lower effective dropout voltage at high current, meaning the minimum necessary input voltage would be lower, but no advantage when you can't change the input voltage. There is one other thing, how well it removes heat. the 3A regulator has 3°C/W thermal resistance to tab, while your 7.5A one has 2.7°C/W. So it does run maybe a little cooler but not by much! You can afford to dissipate up to 10% more power on it before it hits the thermal limit, which means up to 5% more current, under ideal conditions. But to be kept in mind that the total thermal impedance also contains thermal impedance from tab to board, from board to air, etc, so in reality this small difference is even smaller yet, realistically it is zero. I don't suggest skipping the heatsink if you want to remain compatible with high current loads, and you should use high temperature rated inseparable thermal paste, not CPU paste. I keep HY410 for such use. But soldering the tab to PCB should also be pretty good. Putting a heatsink on top of plastic casing of a regulator is nearly pointless.
Very interresting content, especially about CPU voltage regulator. I like also informations about 486 dx/2 80MHz CPU. Most people use dx/2 66 or dx4/100 but dx/2 80 is forgotten. This CPU with 40MHz FSB in my oppinion is more interresting than dx4/100 with slow 33MHz FSB.
I discovered recently that you can get SMD sample packs of the full E96 range of resistance values for relatively little money (£16). A set of 0603 and 0804 would cover most eventualities.
Higher current specification of a regulator doesn't mean that it will be running colder. It has to dissipate the same amount of power regardless of its amp rating. Dissipated power depends on the voltage that the regulator has to drop and the current that it passes (P = V * I). That's why LDO regulators can supply quite more current (when supplied as low input voltage as possible) than non-LDO variants. And that's the same reason why switching regulators are used. For those input to output voltage difference doesn't really matter.
16:00 who cares if it looks a bit ugly? It does the job, it is soldered down properly and it allowed you to get on with the repair instead of spending more money and waiting for a few days for the parts. As a bonus, new parts don't need to be shipped to you which makes this repair even more environmentally friendly ;)
It might make you feel better, but whether or not you have new parts shipped is unlikely to have any meaningful effect on the environmental impact. It's not like they're sending a whole plane, truck, etc just for your order:
I also have that BenQ lcd monitor :) Good to know: besides VGA It accepts 15khz rgb signals, so you can hook up Amiga and other retro platforms that use rgb scart connectors with an adaptor!
Just a tip: LDO's dissipate power only relative to their voltage regulation drop and current supplied to load. This means no rating on the LDO will change this and every LDO put in the board will be just as hot.
Would be neat if fastdoom was added as a benchmark option. While original Doom is a good representation of how it would run back in the day, fastdoom would be a better representation for people who want to do some retro PC gaming on hardware and want to eke out a bit more performance on a less-ideal PC. Wonder if you could get Quake running acceptably (by the standards of the day of someone forced to play on a previous-gen PC :p) on that. I like the kinda chill almost liminal background music. :)
Actually if you look at the vast majority of DRAM specifications you'll find that they all pretty much settled on a refresh period of 15.6uS. This even carried over somewhat in to SDRAM. You have to go back a loooong way to find DRAMs that need to be refreshed much more frequently than that. But I think what you're seeing here is that DRAM refresh is already quite a small percentage of overall bus utilisation, and that a slower refresh rate is probably more of a power saving feature for battery powered devices than a performance enhancing feature.
As usual great job ! I love watching your videos. And just out of curiosity I'll ask if you have in your collection the motherboard I once had in my first PC of my own ? It was a Zida/Tomato 4DVS.
I have 2pcs of i486DX4/100 CPUs and the multiplier can be changed between 3x or 2x. And when these SiS497 can go up to 66MHz bus speed then 2x66 gives 133MHz. Both of those CPUs run at that speed with default voltage. Motherboard is Asus PVI-486SP3. PCI-bus don't like that post-write option anymore but when it's disabled, everything works.
Hello! I am referring to your experience in board repair. There is a rare 624415-230 board with A80501-60 (Intel Pentium 60 MHz), the board is very similar to DFI THE LEOPARD, which is on The Retro Web site, the board starts, makes one peak, but there is no image with any video card, perhaps after one peak it freezes. There is a Dallas on the board, can it behave like this because of a dead battery in Dallas? The code on the postcard is 0088. If you remove the RAM, the board swears about it with a series of peaks in the speaker.
Will 1MB of L2 cache work like 3D V-cache on AMD X3D CPUs and massively boost gaming performance? If you have a Voodoo 2 you can run Quake in this system and test 3D gaming performance 😁
The cache on this system is used to accelerate memory access. Means, that it makes sense to have a lot of this if you have a lot of memory. With 32MB of RAM the difference between 256K of cache and 1MB would be insignificant, but with 128MB of RAM it would. Voodoo 2 makes no sense on a 486, even Voodoo 1 is an overkill, since that needs a strong FPU, which 486 doesn't have. The same is valid for Quake, it also needs a good FPU and is a game made for Pentium.
@@necro_ware There are some videos by people that have paired the Voodoo 1 with an Am5x86@160Mhz just for fun. Yes, 3D games are much better with the Pentium's stronger FPU, but the point is just to do it for fun and science. SUCRA has a video with that pairing playing a few games including Quake and the results were interesting. Since this Soyo can use 1MB of L2, I thought it would be an interesting experiment. I have a Am5x86 running at 160Mhz in a motherboard with a single PCI slot, but it doesn't support 1MB of L2. I also don't own any Voodoo 1 or 2s. I just have a few Voodoo 3 models. Yes, it's silly, but it's just for fun.
I saw that you checked for unwanted battery charge current with the meter on the amp range, and apparently measured about 2mA. That's a lot for a lithium cell that is not intended to be charged. Maybe there's a meter problem...? You should ideally be using a milliamp or microamp range, and the charging current should be really a lot lower - even 10 microamps would be too much.
There was another voltage regulator on that board.. Every one uses different resistor values.. So it that out of spec with the regulator your are using, then (what already mentioned in the video) you have to change those resistors too.
I just got a Pc with an AMD DX2 66 and a QDI V4P895P3/SMT + Miro Crystal 10SD. I installed DOS 6.22 and it runs fine. I have my problems with DOS because I'm a little bit too young to know the syntax etc. Do you think it is a good idea to install Windows 95? Do I need to do anything to use the CD-Drive I've built in? The bios can't detect it. There is only the option to configure a master and a slave drive which works great for autodiscovering HDDs.
Hi, as @MKRENB already mentioned, in DOS you will need CD-ROM drivers. 486 was usually not able to boot from CD-ROM or detect it. Just google for DOS mscdex and how to use it.
honestly, i never had much luck with the SIS 496/497. it was always kinda slow... my m919 w UMC chipset is much faster (with real cache ofc). but the funniest thing is that my m912 is by far my fastest 486 board, with a good vlb card (CHIPS vlb cards are absolute gold) i get 48-49 fps in doom with an amd 5x86 at 133. i just got an opti 895 board though, curious to see how that fares.
You probably have a super fast DX2-66 system if you think so. As I said, this is nothing to make world records with, but the performance is within expectations. Probably on a faster mainboard with VLB graphics card it could be another 10-15% faster, but for an average system of that time it was still quite ok.
Wouldn't it be just easier to add a tp4056 or something similar to the battery mount meaning you can use a lithium battery without messing about with the board?
Did the resistors around the voltage regulator need to be adjusted because it's a different regulator to the original? Also, you put in a 3V CPU but measured 3.4V at the regulator and you were happy with that. Isn't the voltage too high?
Despite the 3V inscription, that CPU is actually specified for 3.3V-3.6V. I could theoretically install different resistors to get any voltage below 4V out that regulator, but this board has three jumpers predefined for 3.45V, 3.6V and 4V.
Uhm I don't fully understand the mod you did to use a CR2032. Doesn't the power supply power the RTC by default when the system is on regardless of any mod? I modded several board but I only disable charging
Per default PSU would power the RTC and recharge the battery. The mod which I made would prevent battery from recharging, but the PSU would still power the RTC when turned on.
I have even better variant or higher model then this MB but sadly it dead right now. I am troubleshoting it for years on and off and the last thing i will try is to change all tantalum caps.
Am I surprised an Intel DX4 wouldn't run at 120 MHz? Not in the slightest. Am I surprised an AMD DX4 labeled 100 _would_ run at 120 MHz? Not in the slightest. AMD used this as a secret weapon for _years,_ almost every builder knew that as long as your board could handle a 40 MHz bus, it didn't matter what speed your AMD 486 was marked. It was going to be happy with a 40 MHz FSB, maybe needing an extra 0.1V thrown at it. This was even true of their later 386/33 chips. They all happily ran at 40 as long as the board was good there. I also noticed some pins on IDE port 2 that are bent enough that they might be touching. That might be worth checking.
I don't follow your math about the 1 MB of cache vs RAM size, so I must be missing something fundamental here... Isn't it more about the application you are running, and how much memory that app accesses, than the amount of physical RAM you have on the system? I mean, if your DOS app consume 1 MB of RAM, it will fit all in the cache and be super fast, but it has nothing to do whether you had 32 MB of RAM on the system or 512 MB. I'm confused... To me, cache has always been about optimizing the frequently used memory, however much memory you have behind that cache.
That is actually a complicated topic, which I wanted to handle one day in a dedicated video. The memory on an x86 is not linear, as we imagine, but is organized in pages, which can be cached. The system has to know what is cached and what not, it's not enough just to put memory pages into the cache. To resolve the memory address to the cached pages there are lookup tables which can cover only a limited amount of physical address space. It's hard to explain in a short comment, because there are many details like WB/WT cache type, bank interleaving and more. This all adds more complexity, but basically more cache doesn't automatically mean better performance.
@@necro_ware Yes. I was oversimplifying by ignoring the TAG, which means you may have too much RAM to effectively cache, but that's why I was being specific about a DOS app (ie no multi-process, so no vastly different memory locations being pulled). I agree in Windows, with not enough cache and an address space too large, it becomes pointless trying to cache if as soon as you get a context switch your entire cache becomes invalid. Still, bigger cache (bigger TAG) it always better here. Just not linear I agree. Also, WB > WT speed-wise as far as I know. And not affected by RAM amount, afaik. Bank inter-leaving, is where I draw a blank. I do not know what that is. So perhaps that's the source of my confusion. As an embedded developer, I'm mostly familiar with non-MMU system caches, so maybe that plays into it as well. Either way, yes cache would make an interesting video if you ever go there ;)
i think you should have used a different battery holder and not modify the power supply connector for safety and in the interest of keeping it as close to original within reason.
@@necro_ware I wonder if they were actually never intending to fit a socket and instead they had tabbed cells in mind which might have the same pin distance as a socket. But that has risk of collision too. Anyway it seems a little silly to trim the power connector instead of the more readily replaceable cr2032 holder :D Hypothetically a CR2032 holder with trimmed sides could exist which wouldn't collide, but i severely doubt one existed at the time. Modern slimline ones are SMD and made of LCP, which was much too expensive for this type of use back then. I do lean towards them only ever actually intending to fit NiCD and only drawing in the CR2032 footprint for no solid reason whatsoever.
Seeing this old equipment calls to mind the time a colleague told me about working for a semiconductor manufacturer in the 70s. He ended up inheriting someone else’s logic diagnostic routine that would get stuck in an endless loop until a manual break command was issued. Anyway, the micro code (and silicon) was so primitive that it didn’t have an exception built into the code in case of a divide overflow, so of course every time the routine would run it would get stuck in an endless cycle trying to divide by zero, catastrophically overheat in the span of a few milliseconds, and destroy itself (and the ancient machine it was running in) spectacularly by blowing the lid right off the chip leaving a little charred cavity presumably where the ALU was. Took a while to fix that bug but he figured it out and everyone went on to live full and somewhat meaningful lives.
the fix probably was : check if the divisor is not 0.
go live your life, but you wouldn't believe how hard is to make programmers check things (they'll excuse "performance", as if a single check would matter).
Good hint with memory stick during soldering, thank you! 🙏🏻
Paying attention to classes at school pays off 😂. Simple and elegant explanation on the voltage regulation
Basically, anything above 3.7V is out of range of lx8383a because dropout 1.3V, MB's pdf says 4.0V for cyrix, it is possible only for low-dropout regulators with dropout 1.0V
Yeah, that's true. The dropout is extremely hard on the edge, for LX8383A it is typically at 1V, but I think, even for Cyrix I would use it rather at 3.6V. It usually works just fine if you don't overclock it.
Still have a soft spot for late 486 motherboards - while all of my friends at LAN parties could afford pentiums, a very similar motherboard served me well enough for years with a Cyrix 5x86-100 overclocked at 120MHz.
I had 486DX2/90 when friends bought new and shiny P60 machines. With my S3 Vision864 VLB and bus speed @45MHz it run Duke Nukem 3D better that friends P60 😆 Of course Quake was totally different story...
@@djpirtu2 There was an official DX2/90? 80 and 100, yes -- the DX2/100 was short-lived because very few VLB cards were happy at a 50 MHz bus speed, but the DX2/80 did fine because almost all cards that were specced for 33 MHz also ran at 40, and it was only the VLB card(s) you had to worry about (assuming the ISA clock divider was changed from CLK/4 to CLK/5).
@@mal2ksc No not official. If I remember right the CPU was Am486DX2/66 and I just overclocked it to 90MHz by swapping oscillator. Never had any problems with VLB cards at 50MHz. Now I have one Am5x86 system @150MHz so 3x50MHz, and everything in BIOS is set to tightest, Cirrus 5434 VLB flies on it. 🔥
Thank you for telling and showing everything in such detail. It was interesting to see how the voltage regulator works in detail.
I have really hard time to calculate the right values of those smd resistor.. Never been good in math.
For those who may be confused by why the resistor values did not fit: Some regulators have the 1.25 V reference voltage between output and ADJ pin, like the LX8383A, which follows the classic LM317 school. Other like the SHARP regulator (it seems to be a Japanese thing) went to 1.25 V between ground and ADJ pin, which also means they have an explicit ground connection at the IC. In any case trying to swap one for the other will result in the roles of R1 and R2 effectively being reversed, so the voltage output ends up wrong as observed. This is why swapping two resistors worked for the 3.45 V setting.
Nice to see it live again! 🎉
This was great, esp the part about the voltage divider! Thanks!
It's really true what a scholar once said, if you can't handle the fatigue of learning, you'll have to endure the pain of ignorance later. Like me, if I hadn't wasted time in school, maybe now I'd be making cool content like you.
Fantastic tinkering and nice board, looking forward to more!
Motherboards with soyo brings me back memories from my teens…
That and a Cyrix processor 486 dlc
Ah, Soyo. Good memories. This was a preferred manufacturer when I had a pc parts vendor I worked with in the ‘90s.
Hi, great video as usual. Just small note: that voltage regulator with higher current capability will be heated to same temperature as that other, because same amount of heat needs to be dissipated. 🙂👍
True, that was a misleading remark indeed.
@@necro_wareGreat video anyway! You inspired me to buy and successfully repair a few retro motherboards.
Hello, clueless person here, so in order for that to occur you would need to increase the surface of the component or use more than one component so the heat is distributed? 😅
@@RetroTinkerer Hi, yes, it depends on temperature resistance from chip to package, from package to ambient air... so yes, using cooler, or other package can enlarge that surface (that is, why heatsinks are ribbed, to make bigger surface with contact to ambient air) and make it not so warm. 🙂
Also, it is exceptionally rare that a part won't work _because_ you put a heatsink on it. It's not impossible, but hardly worth worrying about. I think I'd try to mount it (and its wraparound heatsink as shown in the website photos) in a vertical position rather than parallel to the board and screwed to it, if it's really getting that hot, or maybe still place it parallel but as high above the board as its little legs will allow so it gets some cooling from the back. The ground plane it was bolted to does distribute heat, but then it hangs onto it because it wasn't designed for radiating heat, it was designed to be a ground plane. (It's one of the two inside layers of a typical four-layer board, the other is generally power.)
In essence, you've taken a "parts board" for someone else, and made it into a "working board" for you! It's amazing to take something where most people back in the day would just want to rip the guts out of this motherboard to fix a barely broken motherboard, and instead actually fix it!
One generation after this, the Pentium 1, ignited my love for PCs, back when I was just a elementary school kid, so nice to see such old hardware being well cared and kept alive
Very cool tests done with the motherboard. Congratulations on the video. Hugs from Brazil.
another one saved from the landfill! Thank you for sharing!
Oh man, thank you for the videos! I’m so burned out at work and this is the best refreshing for me 🙏
Omg you are hardware genius, God. So much advanced stuff people don't know. It is pleasure to watch this old pc parts get restored and works improved. Just simply amazing
This was really satisfying to watch and a nice rebuild, a lot of effort put in but really well explained too - great detail! Makes me think I should resurrect some of my junk pile.
New video, I watch to enjoy
What a first Saturday night to come, thanks for the upload
This is the first video of yours I've seen! I really enjoyed how you explained the changes you were making! It was very clear and enjoyable to follow along with 😃
Love your work my friend keep it up!
loved the long video and the repair, awesome job you did there, and calculations
As always good work. I have 2 soyo's to bring back to life. Apparently it would just be changing the swollen capacitors, but no. Socket 370, which was not booting, now heats up the processor and stops reading memory. The other socket 7 that sometimes booted and crashed, after I changed the capacitors the processor heats up and does not leave FFFF.
Man, you never fail to impress! Well done on this board.
oh, WOW! I had exactly such a motherboard from SOYO! And I loved it for overclocking it with some AMD 5x86DX133 and clock of 50 MHz BUS (no, not 25 MHz, but real 50 MHz!) and multiplier of 3... This beast outruled every Pentium up to Pentium 90 including! 😀
i like scrap computing approach with his vertical battery blaster for small battery spots like yours. The bios tweaking you showed is useful, thanks
Very nice work.
You can use BAT85 diode for battery. It has lower voltage drop and it will hold bios settings for longer.
Yes. Specially, it's a Schottky diode, and will have a significantly lower forward voltage drop than a regular silicon junction diode. Current and reverse voltage capabilities are fine for this application.
Great work, thanks
Awesome video as always!
Nice job you did. I like your content and look up to the next video. Greetings from Steven from the Netherlands
A note about the regulator: A linear regulator is basically a variable resistor that turns all the excess voltage into heat. This means how hot a linear regulator gets is never related to how powerful it is, but rather directly to how much voltage it's dropping. if it's dropping 5V down to 3.45V, it will be dropping 1.55V, which means that at its limit of 7.5A, it will have to dissipate 11.6W of power, and at 3A any regulator is going to dissipate 4.65W of power regardless of their rated maximum (give or take a little bit depending on efficiency). This is different in switching regulators, which turn the input on and off for different periods (PWM) and then smooth out the output, meaning there's very little excess power at the cost of having a noisy output that needs very good filtering.
(Wrote this comment because I didn't see anything related to it, and then just after sending, I see that just a little further down than I scrolled, someone already mentioned it)
Yeah, thank you though. My statement about that was of course misleading.
8:16 all linear regulators emit the same exact amount of heat at a given input and output voltage! The output current rating is only the rating for load transient maximum current. Real current through the regulator is the current consumed by the load (use average value), and the dropped voltage is the voltage difference between input and output voltage. So if you have a 1A load, a 3.3V regulator running off 5V will always drop 1.7V and will always consume and convert to heat 1.7W of power. Higher current rating generally implies lower effective dropout voltage at high current, meaning the minimum necessary input voltage would be lower, but no advantage when you can't change the input voltage.
There is one other thing, how well it removes heat. the 3A regulator has 3°C/W thermal resistance to tab, while your 7.5A one has 2.7°C/W. So it does run maybe a little cooler but not by much! You can afford to dissipate up to 10% more power on it before it hits the thermal limit, which means up to 5% more current, under ideal conditions. But to be kept in mind that the total thermal impedance also contains thermal impedance from tab to board, from board to air, etc, so in reality this small difference is even smaller yet, realistically it is zero. I don't suggest skipping the heatsink if you want to remain compatible with high current loads, and you should use high temperature rated inseparable thermal paste, not CPU paste. I keep HY410 for such use. But soldering the tab to PCB should also be pretty good.
Putting a heatsink on top of plastic casing of a regulator is nearly pointless.
Very interresting content, especially about CPU voltage regulator. I like also informations about 486 dx/2 80MHz CPU. Most people use dx/2 66 or dx4/100 but dx/2 80 is forgotten. This CPU with 40MHz FSB in my oppinion is more interresting than dx4/100 with slow 33MHz FSB.
I discovered recently that you can get SMD sample packs of the full E96 range of resistance values for relatively little money (£16). A set of 0603 and 0804 would cover most eventualities.
Well done!
Higher current specification of a regulator doesn't mean that it will be running colder. It has to dissipate the same amount of power regardless of its amp rating. Dissipated power depends on the voltage that the regulator has to drop and the current that it passes (P = V * I). That's why LDO regulators can supply quite more current (when supplied as low input voltage as possible) than non-LDO variants. And that's the same reason why switching regulators are used. For those input to output voltage difference doesn't really matter.
True.
16:00 who cares if it looks a bit ugly? It does the job, it is soldered down properly and it allowed you to get on with the repair instead of spending more money and waiting for a few days for the parts. As a bonus, new parts don't need to be shipped to you which makes this repair even more environmentally friendly ;)
It might make you feel better, but whether or not you have new parts shipped is unlikely to have any meaningful effect on the environmental impact.
It's not like they're sending a whole plane, truck, etc just for your order:
So you know, Adrian would probably call a 33 minutes video, a "short" one. Great video nonetheless. Thank you.
yeah Soyo always was a nice company back in the day.
Enjoyable.
I also have that BenQ lcd monitor :) Good to know: besides VGA It accepts 15khz rgb signals, so you can hook up Amiga and other retro platforms that use rgb scart connectors with an adaptor!
Just a tip: LDO's dissipate power only relative to their voltage regulation drop and current supplied to load.
This means no rating on the LDO will change this and every LDO put in the board will be just as hot.
True, that was a misleading thought
Would be neat if fastdoom was added as a benchmark option. While original Doom is a good representation of how it would run back in the day, fastdoom would be a better representation for people who want to do some retro PC gaming on hardware and want to eke out a bit more performance on a less-ideal PC.
Wonder if you could get Quake running acceptably (by the standards of the day of someone forced to play on a previous-gen PC :p) on that.
I like the kinda chill almost liminal background music. :)
Actually if you look at the vast majority of DRAM specifications you'll find that they all pretty much settled on a refresh period of 15.6uS. This even carried over somewhat in to SDRAM. You have to go back a loooong way to find DRAMs that need to be refreshed much more frequently than that.
But I think what you're seeing here is that DRAM refresh is already quite a small percentage of overall bus utilisation, and that a slower refresh rate is probably more of a power saving feature for battery powered devices than a performance enhancing feature.
As usual great job ! I love watching your videos. And just out of curiosity I'll ask if you have in your collection the motherboard I once had in my first PC of my own ? It was a Zida/Tomato 4DVS.
Thank you very much. I don't have that board, but I got some donations to work through and may be there. If there is one you will see it here ;)
I have 2pcs of i486DX4/100 CPUs and the multiplier can be changed between 3x or 2x. And when these SiS497 can go up to 66MHz bus speed then 2x66 gives 133MHz. Both of those CPUs run at that speed with default voltage. Motherboard is Asus PVI-486SP3. PCI-bus don't like that post-write option anymore but when it's disabled, everything works.
You can run the Intel DX4 100 at a 2x multiplyer with the jumpers. I have a board set up with a 2x multiplyer and a 50Mhz fsb.
Hello! I am referring to your experience in board repair. There is a rare 624415-230 board with A80501-60 (Intel Pentium 60 MHz), the board is very similar to DFI THE LEOPARD, which is on The Retro Web site, the board starts, makes one peak, but there is no image with any video card, perhaps after one peak it freezes. There is a Dallas on the board, can it behave like this because of a dead battery in Dallas? The code on the postcard is 0088. If you remove the RAM, the board swears about it with a series of peaks in the speaker.
What about Overdrive CPUs on this board?
What is the name of the tool you use to straighten bent pins?
I have no clue how it's called. Search for soldering tool fork or something similar.
Will 1MB of L2 cache work like 3D V-cache on AMD X3D CPUs and massively boost gaming performance? If you have a Voodoo 2 you can run Quake in this system and test 3D gaming performance 😁
The cache on this system is used to accelerate memory access. Means, that it makes sense to have a lot of this if you have a lot of memory. With 32MB of RAM the difference between 256K of cache and 1MB would be insignificant, but with 128MB of RAM it would. Voodoo 2 makes no sense on a 486, even Voodoo 1 is an overkill, since that needs a strong FPU, which 486 doesn't have. The same is valid for Quake, it also needs a good FPU and is a game made for Pentium.
@@necro_ware There are some videos by people that have paired the Voodoo 1 with an Am5x86@160Mhz just for fun. Yes, 3D games are much better with the Pentium's stronger FPU, but the point is just to do it for fun and science. SUCRA has a video with that pairing playing a few games including Quake and the results were interesting. Since this Soyo can use 1MB of L2, I thought it would be an interesting experiment. I have a Am5x86 running at 160Mhz in a motherboard with a single PCI slot, but it doesn't support 1MB of L2. I also don't own any Voodoo 1 or 2s. I just have a few Voodoo 3 models. Yes, it's silly, but it's just for fun.
12:40 -15:06 or short 2 jumpers - resistors paralel
I saw that you checked for unwanted battery charge current with the meter on the amp range, and apparently measured about 2mA. That's a lot for a lithium cell that is not intended to be charged. Maybe there's a meter problem...? You should ideally be using a milliamp or microamp range, and the charging current should be really a lot lower - even 10 microamps would be too much.
Why were the resistors so out of spec originally?
There was another voltage regulator on that board.. Every one uses different resistor values..
So it that out of spec with the regulator your are using, then (what already mentioned in the video) you have to change those resistors too.
Have you ever try to use supercapacitor in place of rechargeable battery?
Yes, they get empty very fast. Exact time depends on the capacity, but something not too big drops the voltage below 2V in a week.
Hello! Can you please share the image of your CF card?
I can't, due to copyright of the programs, but it's basically msdos, Phil's benchmarks collection and some optional tools.
@@necro_ware Thanks for the hint, I found what I was interested in: "Phil's benchmarks collection"
It should be possible with 486Dx4/100 to run @ 2x50MHz - I've had that setup with AMD 486 VT8.
could you please share, what exactly you use as "deoxidiser" in your videos. Some magic spray witch I cannot find in google
deoxit d5.. (iam using that great stuff to for all my vintage computer devices.)
I think in this video I didn't use any dioxide, but when I do, I use different, like Kontakt 60, WD40 Special and others.
@@necro_ware Haha yeah availability and price of D5 in EU is absolutely horrendous.
@@necro_ware thanks!
I just got a Pc with an AMD DX2 66 and a QDI V4P895P3/SMT + Miro Crystal 10SD. I installed DOS 6.22 and it runs fine. I have my problems with DOS because I'm a little bit too young to know the syntax etc. Do you think it is a good idea to install Windows 95? Do I need to do anything to use the CD-Drive I've built in? The bios can't detect it. There is only the option to configure a master and a slave drive which works great for autodiscovering HDDs.
CD-ROM and DVD always need a driver for dos.
Miro Crystal 10SD...haven't come across that name in 30 years
@@toseltreps1101 VLB S3 805?
Hi, as @MKRENB already mentioned, in DOS you will need CD-ROM drivers. 486 was usually not able to boot from CD-ROM or detect it. Just google for DOS mscdex and how to use it.
@@necro_ware Okay thx
honestly, i never had much luck with the SIS 496/497. it was always kinda slow... my m919 w UMC chipset is much faster (with real cache ofc). but the funniest thing is that my m912 is by far my fastest 486 board, with a good vlb card (CHIPS vlb cards are absolute gold) i get 48-49 fps in doom with an amd 5x86 at 133. i just got an opti 895 board though, curious to see how that fares.
To bad you did not put 128 MB Ram and 1 MB Cache on the board, that would be interesting. Anyway thanks for the video.
I'm curious how you disabled the charging circuit. Instead of moving the resistor, couldn't you just remove diode D2?
I could, but then the RTC would be always powered from the battery, even when the system is running.
Shouoldnt it have been much faster though? It's benchmarking only barely faster than a DX2-66 even with a DX4 in there? Did I miss something?
You probably have a super fast DX2-66 system if you think so. As I said, this is nothing to make world records with, but the performance is within expectations. Probably on a faster mainboard with VLB graphics card it could be another 10-15% faster, but for an average system of that time it was still quite ok.
Wouldn't it be just easier to add a tp4056 or something similar to the battery mount meaning you can use a lithium battery without messing about with the board?
Yeah, could be another way to go.
Did the resistors around the voltage regulator need to be adjusted because it's a different regulator to the original?
Also, you put in a 3V CPU but measured 3.4V at the regulator and you were happy with that. Isn't the voltage too high?
yes
3V mark on cpu doesn't mean 3V. It means find datasheet for cpu and find exact voltage, 3.3V or similar. Crap marks.
Despite the 3V inscription, that CPU is actually specified for 3.3V-3.6V. I could theoretically install different resistors to get any voltage below 4V out that regulator, but this board has three jumpers predefined for 3.45V, 3.6V and 4V.
Uhm I don't fully understand the mod you did to use a CR2032. Doesn't the power supply power the RTC by default when the system is on regardless of any mod?
I modded several board but I only disable charging
Clearly you don't understand it. Maybe watch again
Per default PSU would power the RTC and recharge the battery. The mod which I made would prevent battery from recharging, but the PSU would still power the RTC when turned on.
I have even better variant or higher model then this MB but sadly it dead right now. I am troubleshoting it for years on and off and the last thing i will try is to change all tantalum caps.
first!
Am I surprised an Intel DX4 wouldn't run at 120 MHz? Not in the slightest.
Am I surprised an AMD DX4 labeled 100 _would_ run at 120 MHz? Not in the slightest.
AMD used this as a secret weapon for _years,_ almost every builder knew that as long as your board could handle a 40 MHz bus, it didn't matter what speed your AMD 486 was marked. It was going to be happy with a 40 MHz FSB, maybe needing an extra 0.1V thrown at it. This was even true of their later 386/33 chips. They all happily ran at 40 as long as the board was good there.
I also noticed some pins on IDE port 2 that are bent enough that they might be touching. That might be worth checking.
I don't follow your math about the 1 MB of cache vs RAM size, so I must be missing something fundamental here...
Isn't it more about the application you are running, and how much memory that app accesses, than the amount of physical RAM you have on the system? I mean, if your DOS app consume 1 MB of RAM, it will fit all in the cache and be super fast, but it has nothing to do whether you had 32 MB of RAM on the system or 512 MB. I'm confused... To me, cache has always been about optimizing the frequently used memory, however much memory you have behind that cache.
That is actually a complicated topic, which I wanted to handle one day in a dedicated video. The memory on an x86 is not linear, as we imagine, but is organized in pages, which can be cached. The system has to know what is cached and what not, it's not enough just to put memory pages into the cache. To resolve the memory address to the cached pages there are lookup tables which can cover only a limited amount of physical address space. It's hard to explain in a short comment, because there are many details like WB/WT cache type, bank interleaving and more. This all adds more complexity, but basically more cache doesn't automatically mean better performance.
@@necro_ware Yes. I was oversimplifying by ignoring the TAG, which means you may have too much RAM to effectively cache, but that's why I was being specific about a DOS app (ie no multi-process, so no vastly different memory locations being pulled). I agree in Windows, with not enough cache and an address space too large, it becomes pointless trying to cache if as soon as you get a context switch your entire cache becomes invalid. Still, bigger cache (bigger TAG) it always better here. Just not linear I agree.
Also, WB > WT speed-wise as far as I know. And not affected by RAM amount, afaik.
Bank inter-leaving, is where I draw a blank. I do not know what that is. So perhaps that's the source of my confusion.
As an embedded developer, I'm mostly familiar with non-MMU system caches, so maybe that plays into it as well.
Either way, yes cache would make an interesting video if you ever go there ;)
i think you should have used a different battery holder and not modify the power supply connector for safety and in the interest of keeping it as close to original within reason.
I couldn't find any holder which would fit there. I think it was a bit miscalculated by design.
@@necro_ware I wonder if they were actually never intending to fit a socket and instead they had tabbed cells in mind which might have the same pin distance as a socket. But that has risk of collision too. Anyway it seems a little silly to trim the power connector instead of the more readily replaceable cr2032 holder :D
Hypothetically a CR2032 holder with trimmed sides could exist which wouldn't collide, but i severely doubt one existed at the time. Modern slimline ones are SMD and made of LCP, which was much too expensive for this type of use back then. I do lean towards them only ever actually intending to fit NiCD and only drawing in the CR2032 footprint for no solid reason whatsoever.