Hi Necroware, i've seen many motherboards with this socket, but I haven't had the chance to see a single module before you create one from scratch. Thank you for all your amazing work 😀
Reflow on a hot plate is so satisfying, and produces such incredibly professional looking results, especially when you use a stencil to put down the right amount of solder paste. 😊
I was looking through my Socket 7 mobos recently, and found that they are all later ATX models with onboard auto-sensing VRM's for MMX. I guess it's good that I don't have to deal with this socketed VRM stuff... But on the other hand, I don't get a chance to play around with it either. I think the DOS build I have together right now uses an "Intel AN430TX (Anchorage)" that you can find on the retro web. Fairly sure it came out of a Gateway2000 P5-200 Mid-Tower. It has all the fancy gubbins that make earlier P1 mobos look bad... Onboard ATI Rage video, and SoundBlaster audio... USB 1.1 ports... etc.
you don't add flux to pcb when soldering with solder paste, solder paste is already flux mixed in with powdered solder in perfect ratio to just work, otherwise it flow too much
@@necro_ware well yes i did , talking about keyboard warior overacting i use solder paste to bridge solder jumpers on laptop motherboards because soldering iron just suck solder off the pads, bit it is magic with paste it self align and bridge hair thick pads like mind controlled liquid metaland if you have low flow hot air you don't need hot plate it work just as well
Awesome. I actually have no idea which of my boards support VRMs, but I have about seven S7 or SS7 boards so now I'm really excited to find out! Thank you for all the effort you've put into this.
I like to use this module design on my Socket 7 board with split voltage but a linear regulator that doesn't support lower voltages and more power-hungry CPUs. It doesn't have the VRM socket of course, but I was going to hardwire the module set to 2.2 V into the board to allow an upgrade to an AMD K6-2 at 400 MHz. Good info on the sense pin, I'll see how I can implement this without the socket. Huge thanks for designing and sharing this great project!
Yeah, that is also definitely possible. You can just hard wire the Vcore to the Sense pin directly on the module and call it a day. If you properly solder the connections, that should be fine.
I would have gone absolutely crazy for this back in the day, I had a PC shop right around the time the k-6 II came out and loved building and overclocking systems.
Good job and by the way it was a surprise to see the PB because I have a D130 model that only supports single voltage processors up to 200mhz, but it came with 133 mmx and works with overclocking up to 166, but the mobo does not support external VRM.
I'm afraid 133 MMX doesn't exist. MMX starts with 166 MHz and is a dual voltage CPU. If you have it in a single voltage board, then it is over volted. Some CPUs are robust enough to handle that with good cooling, but in the long term it will kill the CPU. As a rule of thumb MMX means dual voltage, non-MMX means single voltage.
@@necro_ware If I'm not mistaken, mmx was written on the back of the processor, I have a 166 mmx here and they are different, as far as I know up to 133 are ceramic, but as I restored this PB some time ago I could be wrong.
@@exitustec From 166 onward there were two versions of the CPU, either in ceramic housing, or plastic with metal shield. You can't tell by the housing if it's an MMX or not. All you can tell is that if it's plastic version, it's at least 166MHz. There the inscriptions are on the back and you can read MMX at 2.8V, or no MMX with 3.3V. The 133 MHz one always came in ceramic housing and is a single voltage non-MMX CPU.
Amazing work. However, I always would prefer a super 7 board with integrated dual voltage support because they got higher bus speeds and sdram. But for older boards, it's a good upgrade feature.
Absolutely, that's why I want to do a big run between those older Socket 5/7 and Super Socket 7 boards. Later will be faster of course, but if we talk about retro hardware it's not so important right? If we have a 386SX we see how fast we can get that, despite that we easily could take a 386DX with much higher throughput. Furthermore, that's why I mentioned those nice OEM machines, which have proprietary hardware and a VRM option.
Since you are concerned about possible automated assembly in the future, you way want to consider adding optional pads for such SMD components on your next layout revision. Then you keep both sets of pads, and can still use through hole components as you prefer, while offering the alternative.
Board can be easily designed to suport both SMD and THT parts (caps and inductor). You could then use (possibly lower profile) polymer caps and pre-made inductors.
why are people still afraid of smaller smd packages like 0603 or 0402 ? You and other members of the retro community actually inspired me with your projects, so a while ago I decided to do something myself and steal AC97 codec from a dead motherboard, along with its related components, to make a CNR audiocard (actually two :D) out of it. And transplanting few dozens of small SMDs, mostly 0402, wasn't as hard as I imagined. I'd say it's even easier to solder smaller packages because pads are so close together so tip of soldering iron could touch and heat them both simultaneously, and you don't need to move the tip a lot to re-align or desolder the component. Also I made a mistake and used 0603 footprint for 0402 details, but it was a good mistake - in many cases after tinning there was already enough solder on the pads to make a good and even beautiful joints with the component, whilst on proper sized pads it always needed a bit of additional solder after aligning the component to achieve joints that resembles ones on factory-made boards. Yes sometimes SMD capatitors may crack when you trying to desolder them, or if you accidentally smd component on the floor its most likely gone forever, but if you do have a strip of new components it shouldn't be a problem.
In the retro community we have a lot of members who are not quite as young anymore. I'm very happy about it, but some eyes and fingers are not the best after all the years and it's easier to solder slightly bigger parts. That's why I also try usually to use through hole parts in my projects. Not only does it better fit the retro look, they are also easier to solder for many people. Another question of course why then I use SMD parts at all and, well, today it's getting hard to find required TH parts and of course they usually need a lot more space. So, bigger SMD parts are a compromise.
I'm a fan of 0805. It's the smallest size where (e.g.) resistors still regularly have labels, so you can see what they are, but it's enough smaller than 1206 to be meaningful. I'll often use 0603 too, as it can really help when packing components around dense pin-pitch packages, and it's not THAT much smaller, so it's not hard to work with. My eyes haven't gotten any younger, though. Compared to just before The Great Plague, I've had to rely more on magnification. Even still, SMD work isn't really that hard. I recommend ALWAYS buying the stencil -- which is usually only a couple bucks on top of a PCB order anyway. Spreading paste over a stencil is much faster, and yields much more predictable results. With a hot air station (with or without heating plate), stencils, and some good SMD tweezers, populating a board full of SMD parts is faster and easier than dealing with through-hole parts, and I've come to prefer it for both the convenience and availability of parts.
looks great. i wish i had the space and resources to keep vintage motherboards around, but alas, i'll have to settle for watching your fantastic videos
Some suggestions for a future Version: A OLED Display presenting vital Parameters as Wattage or Amperage. Also RGB LED Support would be nice as you already know RGB makes your PC at least 10% faster. For controlling RGB you would need an Adapter to be set between a RAM Slot and the module to be able to access the System Management Bus. . . . . . Just joking!
Thank you very much for the feedback. Would be nice if you could make a ticket on the github project page with some detail about the board, so I can make a proper entry in the list of tested boards. May be a link to the retro web with the related board and some info about BIOS and if you had to use a resistor mod for K6-III+
Just one of an infinite number of features, sockets, and jumpers that were provided by an engineer and almost never populated, documented, and thus ever used by anyone.
These heater standoffs do not fill me with confidence! I would have an urge to give it longer ones and then put some heavy steel sheet underneath to make it stable, and rubber feet bottom; also would run a third wire to earth the plate surface and the whole external metal basically. I know it should be packaged well inside so it should be "safe" but i'd rather be doubly safe and not just single-safe. By all reason it doesn't need temperature control since the heater is PTC, so it should balance out around the target temperature. So other than the construction sketchiness, it should be perfectly good.
Yes, please, please, please add a third earth wire to ground the plate surface! Do not rely on there always being adequate electrical isolation between mains heating elements and the plate.
The IBM Aptiva 2176 has a VRM socket. I had a hard time tracking down one of the original PC55 VRM modules to upgrade the machine to an Pentium 200 MMX CPU. Annoyingly the BIOS does not recognize any MMX CPUs and complains at every boot up, however skipping the error then just allows the machine to run without issue! With a modified BIOS and your module, I imagine the 2176 will happily run faster than 233Mhz, the main issue with going any faster might be the lack of necessary jumpers to set the desired speed.
I'm an electronics engineer and have worked with computers and software that I "don't" have a clue about my entire career, how ironic. Now I'm retired and started tinkering with electronics again as a hobby just like when I was a teenager and I have as much to learn as you, electronics these days are not the same as they were then. Like for example, I was looking for a regular 100mA NPN transistor at Mouser and found that I'd better find another one because the one I was looking at was 1mm. (0.039 in.) on the greatest length. I shake my hands too much so I have chosen that my lower limit for now is 0805.
That's why I make my projects using either through hole parts or bigger 1206 SMD with pads optimized for hand soldering. I want, that even people with minor handicap are able to build those projects.
3:40 - If I could make one small suggestion... With the use of the Sense pin, my only worry would be if that loses connection somehow, (ie. if somebody is moving the module around while it's powered, or if header pin on the motherboard was dirty / corroded.) The Sense signal (and FB signal) on the module could default to (close to) 0V. The switching reg would then think the output voltage is too low, then output too high of a voltage, and possibly damage the CPU. (R3, and the DIPSW resistors will also be acting as pull-downs on the FB pin.) To test the theory, you could remove R4 temporarily, but beware of the higher output voltage. I would try adding a weak pull-up resistor, between the Feedback pin and VCORE, just in case. Maybe something like 47K ? Not sure if that would be too high?
Actually, I'm not sure how to solve that, with just a pull-up resistor to VCORE. lol Because the 47K resistor would be as if you used that in place of R4 (which is currently 5K). R3 (plus the DIPSW resistors) would be still acting as the voltage divider, for the FB pin, but giving a higher output voltage. (due to the larger 47K pull-up between Sense and VCORE). (47K was just an arbitrary value I chose.) It's probably fine as-is, tbh, but maybe a future improvement could be some type of overvoltage protection. That would need to measure the voltage difference between VCORE and the Sense pin. If that difference is too high, it would trigger the Disable transistor.
Maybe a simple diode between VCORE and SENSE could work? That way, if the voltage difference between those two pins went above say 0.7V or so, the diode would still give some feedback to the FB pin, in the event the Sense pin on the mobo had a problem. You would need to measure the average expected voltage difference between VCORE and SENSE, on say a power-hungry CPU. If the maximum expected difference is less than about 0.6V, then a normal diode could do the trick. (Anode hooked up to VCORE, Cathode on SENSE.) If the max voltage difference is higher, maybe a Zener diode (connected in reverse to the above).
So glad to see someone who actually knows the steps in through-hole soldering: set, trim, solder. Trimming after solder weakens the joint, which can go cold under thermal variance in use!
If your ripple is ever causing stability due to load-stepping, you should likely look into a 'feed forward' capacitor, which is basically just a tiny capacitor that bypasses the feedback resistor (your 'R4') for faster transient events. Looking at the datasheet for that part and the project, I can see there is a compensation network, but that is for loop-stability and cannot functionally work as a stand-in for a feed-forward capacitor. Because what a feed forward capacitor does, is make sure the regulator gets a more dramatic start on adjusting to a transient events, as an impulsed load forces the sense line to behave more like a transmission line. (suddenly the slow ramp to adjust to, is turned into a square wave) That becomes more important as your 'voltage sense' path starts getting more complex and especially parasitic capacitances after any substantial resistances. This is what your "R4 - 5K" is! So essentially the copper on the "FB" node that goes to the switch bank, are subject to creating a parasitic RC network.
There's actually a comprehensive discussion of how to approach compensation in the ISL6545 datasheet. I find it hard to believe that among C1, C2 and C3 referenced there, none would be tackling parasitic capacitance at the FB pin along the way. Now the crux is that any such Cp would effectively be in parallel with gain-setting resistor Ro, so it'll be very tricky to effectively nuke this pole across the entire range. It may have been a good idea not to be choosing Rs (= R4) at the very top of the recommended range of 1-5 kOhms. Ro is in the 1-3 kOhm range as-is, which doesn't exactly strike me as super low. It should still be moving that pole out to >=10 MHz when faced with Cp = 10 pF, and my understanding is that even Fp2
@@PileOfEmptyTapes I had to think on this a little bit, but according to the datasheet "The goal of the compensation network is to provide a closed loop transfer function with high 0dB crossing frequency (F0; typically 0.1 to 0.3 of FSW) and adequate phase margin (better than 45°)" I still have to go with the opinion of it can't fully compensate for information it doesn't have. The information it doesn't have, is what gets lost to those parasitic capacitances due to the 5k RC. I think a step response is precisely what slips past their definition of "better than 45°" because until they can definitively close that loop delay to 0° a step can still create a ripple. The point of a feed forward capacitor is to help close the phase-delay gap specifically for step response. Though also I might note that just putting a cap across the 5k might not be the best approach, you'd likely want to also run those both into a series 0-500 ohm and adjust the 5k to compensate for that ratio adjustment(a 5k or lower potentiometer might be a clever way to do this without needing anymore parts swapping), that way the feed-forward cap doesn't dominate the feedback network. I've never been amazing at pole filters etc. But I don't think a compensation network can create compensation for a step response it never saw coming, if that information is lost before it enters the comp network. And anything behind the 5k resistor has now already lost part of the transient impulse that has occurred. That's how it makes sense in my head at least.
I have collected 2 hp vectra pentium powered computers. Both of them have Vrm socket support for pentium mmx processors. I'm interested in seeing what the hp bios will support. They are 430hx and 430fx chipset. Hp also had a sis chipset in socket 7 in the vectra. How can I get my hands on an assembled module? My skills are not ready to solder like this yet.
No, I don't, but the old ones were either linear, or very limited switching regulators, which only provided one voltage 2.8V. So, even if I'd have one of the old exemplars, there wouldn't be much to compare.
Осталось только выпустить более последнюю ревизию с 4pin 12В разъёмом питания CPU, для совместимости с блоками питания стандарта ATX12V 2.X, может даже ATX12VO!
I hope that when I say this it doesn't sound rude but I was also hoping that one of your announcements would have been that these can be ordered already made. I would love to be able to get one of these for an older motherboard that I have that has one of these VRM socket upgrades but my issue is is that I don't know where to buy all the little chip parts and I have no soldering skills. I need to find someone who can actually sell one of these especially latest version already pre-made so I can buy it. Again I hope I don't sound rude in any way. I just need help.
I would like to sell those things, but German laws are unfortunately complicated and are made to kill every inventor asap. I'm currently trying to organize some kind of official tiny online shop, but so far it looks like it would be easier and faster to buy it from someone else.
@@clintthompson4100 I don't know personally, but I already saw couple of time my projects been sold on ebay. Also the last revision of this VRM. Most probably the latest revision will also appear soon.
Why not use some IC with integrated power stage? Like AP62600? Small, high frequency (smaller inductor, smaller capacitors, SMD only), 2 channels will not be problem.
Because it's too weak. The revision 0.1 also used an integrated solution. It was not AP62600, but the point is, that it can deliver only 6A. That is at least half as much as needed.
@@necro_ware Running a higher frequency controller would be helpful though. The one criticism I have about this module is that it's awfully chonky. You could use much smaller SMD parts with a faster controller. Also, I haven't looked at your schematics, but please tell me you have a weak resistor between the Vout rail and FB pins, and not relying solely on the path through the CPU for Vsense? If anything should ever happen to any link in that chain, you really want to have a fallback path to keep it from going open-loop.
@@nickwallette6201 You are right, it can be improved further. Like everything can be always improved. And after it is improved, it can be improved even more. In my career as a software engineer where improvements can be done even easier than in hardware, I had to learn to tell myself when something is good enough. This thing now is good enough for me. If someone wants to improve it even further, it's open source. In regards of sense resistor, I know what you mean and that would be not good. Luckily it is not possible to connect Sense directly to FB. It has to go through a voltage divider, hence there is always 5K in between. And if someone messes with Sense outside, it's really not a fault of the adapter. Eventually doesn't matter how much effort you put in your stupid proof system, universe will come back with an improved stupid :) Don't remember who's quote it was, but it's unfortunately still valid.
@@necro_ware But would you still lose the other half of the divider? E.g., consider what would happen if your Vsense pin is damaged somehow -- corrosion, bad fork in the socket so it doesn't make contact, solder failure, etc. What happens to the output voltage in that case? A good remote sense circuit will always follow the local rail through a resistor that is overridden if the remote sense is present, but provides a correct (albeit uncompensated) output even if the remote sense is entirely missing. That way, you mitigate the risk of ending up with 5V on your Vcore. Just food for thought. :-) And yes, I know perfect is the enemy of good with any product. That's totally fair. But seeing the height of this module immediately makes me wonder whether you'll run into clearance issues with, e.g., AWE32, or other cards that are long enough to hang over the CPU area of a motherboard. Probably not an issue for a good 80% of users. Just worth considering, if you ever decide to hack at this design again. Although, layout gets increasingly important at higher speeds. Everything is a compromise.
I would personally recommend just getting your boards assembled at one of those known PCB places. A board like this would probably be around 15 euro delivered. Probably even less if you optimize the design. In that case I would also make all parts SMD. (Except for the jumpers probably)
If it's already tested and released part, which you can order in a package of 5 or more, probably it's worth it to order pre-built modules. However, during a development of a prototype where you want just one piece to test and improve it's a lot cheaper to do it manually.
@@necro_ware I do it on a regular basis. For 40-70 bucks delivered, I can't be bothered even as a hobby. Especially when you consider that you have to order the parts etc Anyway, it was just a suggestion 👍🏻😎
Yeah I tried that, but it was too expensive for prototyping. Especially I found it disappointing, that the selection of parts they have for cheap assembly is very limited and if you need something what they don't have in the standard set, they have to order it for you off site and then the price explodes to over 50€ per module. Quite a lot of money to get one and realize, that you made a mistake and throw it away. Of course for the revision 0.6 which I opened people can easily order pre-assambled modules. They should be fine.
@@necro_ware if you use jlcpcb. You can order anything from lcsc. What I often do, is hybrid assembly. Meaning, getting all the passives and connectors done and doing the critical parts myself. I think this board can also be optimized. I might have a look, just for fun, but I am a little busy atm. Far too many projects myself 😁😜 Still really enjoy what you're doing btw 👍🏻😎
I have a tip, I noticed that having a Solder Fume Extractor no matter the brand as long as it sucks out the fumes it actually advances the soldering process, I don't know if you noticed that, having more air flow/oxygen over the area being solder it gets the job done faster. PLUS the benefits of not getting fumes in your lungs.
TBH I planed to do a fume extractor since a long time, even bought a fan and carbon filters, but it's in the TODO box since forever. I always solder with an FFP mask telling myself, that it helps. Probably it does a little bit, but having a fume extractor would be even better.
Hello. So this is the thing, that goes into PCChips M520 next to the socket? I never knew, for what is that thing, but I saw some jumpers in it. I was told, I should not touch it, if I dont understand it. So it is needed in that board only for Pentium 133 and higher? So without it, I can use only Pentium 120 and lower?
It is needed for dual voltage CPUs in the first place. This would be Intel Pentium MMX, AMD K6-2 etc. Any non-MMX Pentium is a single voltage CPU and doesn't need such a VRM. You can theoretically use even Pentium 200 (non-MMX!) CPU in your board, but at that speed the linear voltage regulator on your board would get extremely toasty. Using this VRM would not only allow to use a lot faster dual voltage CPUs, but keep your system much more efficient and cooler even with single voltage models.
@@necro_ware Thank you. I put it to notes. But what about year 1996? Did they sold these voltage regulator modules separately? Or computer shops used them, when there was Pentium 166 MMX built inside computer? Because, I've never seen some prehistoric ones, like from 1996 or 1997. Anyway, board was released in 1996, Pentium MMX came in january 1997. So seems, PCChips knew about it, that it will come later next year or what.
@@warrax111 They did exists, yes, but they were not nearly like this one. Usually they've been sold for 2.8V MMX only with no other voltage option. Some had couple of jumpers to select a narrow voltage range between 2.8V and 3.2V, but all of those were not powerful enough and had no option for 2.2V or lower, which K6-2 needed.
What is the Current draw of the CPUs ? Asking since you are using a very old and slow chip and there is a good chance you can make it all SMD with a newer controller. maybe for a future Socket 8 VRM version? Anyway great job and if you want electrical help just ask.
Thank you very much for the offer. And yes, I could use a faster controller, but it would be probably an overkill. The current draw doesn't go above ~12A at
@@necro_ware Well i am coming from the electronics side and i can tell you this: - synchronous is like the normal for high current for obvious reasons.... - a lot of chips these days can do 2MHz+ best i saw was 3.4MHz if i remember well -i didnt say you should but here is my 2 cent : if you increase the switching frequency you can decrease the size of the magnetic s and caps , i sort of felt strange when i saw so many electrostatics out there , i am not saying that it is incorrect but odd to me -you have a single buck phase* there if you look at modern motherboards you can se 8-12 phases , powering even a slow* weak modern CPU or MPU is a bit harder these says but progress has a cost Anyway if you need help ask , you also inspired me to think of a ho small it can be made with modern chips. Also keep up the great work! PS ironically i work in electronics and looking at programming and computers as a hobby LOL
Yeah, it actually doesn't matter. 1.0 sounds like it's well tested and feature complete. With this module we have the feature completeness I would say, but it needs to be tested by a wider audience. I just don't have the capacity to test it with so many boards and CPUs. Once I got some reliable feedback I can release it as 1.0
LOL I remember when the Linux kernel was 0.97, 0.98, 0.99 then they just kept adding a patch level number to the end. People wondered if it would ever reach 1.0 😅
Necroware i need your help with my GPU. what can be potential causes of a GPU that randomly stops working and still looks new? The GPU is an AMD RX 6600
(Digital: venturis FX) pc has VRM socketed. that would be nice to see in full action cause it can take 192mb ram and has onboard s3 graphics (2mb upgradeable)
Possibly my favorite thing about the retro community is how often a hardware design is released as open-source.
It is one of my favorite things about this channel. I don’t know a channel that is more open source than this one. Quite a few are not open source.
Opentendo ftw
"I'm not an electronic engineer" * rolls his own inductor *
:)
He made the coolest vrm addons for vintage motherboards.
He also : "I have no clue about electronics".
Me : "WUT?"
This is what happens when ignorance and bravery come together :)
@@necro_wareThe up side of Dunning-Krueger 🤔
Every great project starts with, "how hard could it be?" 😂
He builds his own coils as well ... 16:37
There are two *"oh no, I haven't studied for the exam!"* types of people:
- one, failes the exam.
- the other, gets "only" 80% score.
Amazing job! I like the “I’m going to kill you” heating plate 😂
Thank you. May be you remember, I wrote about it in the comment on your channel not so long ago. Here it is in all it's "beauty" :)
Hi Necroware, i've seen many motherboards with this socket, but I haven't had the chance to see a single module before you create one from scratch. Thank you for all your amazing work 😀
6:55 nice to recognize that shop and congrats on the UtterCrap 3000 😂
You're my favourite retro pc hardware hero
You had me at socket 7 😊
For someone with "limited" understanding of CPU Voltage Regulation, your design and thought processes for the VRM are sure on point!
Reflow on a hot plate is so satisfying, and produces such incredibly professional looking results, especially when you use a stencil to put down the right amount of solder paste. 😊
This has just given me an idea for a Socket 5 to Socket 7 interposer that gives selectable I/O and core voltages.
I was looking through my Socket 7 mobos recently, and found that they are all later ATX models with onboard auto-sensing VRM's for MMX.
I guess it's good that I don't have to deal with this socketed VRM stuff... But on the other hand, I don't get a chance to play around with it either.
I think the DOS build I have together right now uses an "Intel AN430TX (Anchorage)" that you can find on the retro web.
Fairly sure it came out of a Gateway2000 P5-200 Mid-Tower.
It has all the fancy gubbins that make earlier P1 mobos look bad... Onboard ATI Rage video, and SoundBlaster audio... USB 1.1 ports... etc.
you don't add flux to pcb when soldering with solder paste, solder paste is already flux mixed in with powdered solder in perfect ratio to just work, otherwise it flow too much
Yes, this is what I also thought after the first try. Later you can see, that I don't add it anymore. Lesson learned I'd say ;)
@@necro_ware well yes i did , talking about keyboard warior overacting
i use solder paste to bridge solder jumpers on laptop motherboards because soldering iron just suck solder off the pads, bit it is magic with paste it self align and bridge hair thick pads like mind controlled liquid metaland if you have
low flow hot air you don't need hot plate it work just as well
Awesome Job!!! I love it! Never seen a socket 7 board with this socket.
Awesome. I actually have no idea which of my boards support VRMs, but I have about seven S7 or SS7 boards so now I'm really excited to find out! Thank you for all the effort you've put into this.
I like to use this module design on my Socket 7 board with split voltage but a linear regulator that doesn't support lower voltages and more power-hungry CPUs. It doesn't have the VRM socket of course, but I was going to hardwire the module set to 2.2 V into the board to allow an upgrade to an AMD K6-2 at 400 MHz. Good info on the sense pin, I'll see how I can implement this without the socket. Huge thanks for designing and sharing this great project!
Yeah, that is also definitely possible. You can just hard wire the Vcore to the Sense pin directly on the module and call it a day. If you properly solder the connections, that should be fine.
What a great channel! "Unmissable" video, just another one of yours! Thanks & cheers!
"Utter Crap 3000", now that made me laugh and now for some reason I want it :D
I would have gone absolutely crazy for this back in the day, I had a PC shop right around the time the k-6 II came out and loved building and overclocking systems.
Wonderful work as always!
Excelent job, sir!
Amazing project. Thanks for sharing ***EVERYTHING****. Much appreciated !!!
Nice job
Awesome! I’ll be building one soon too! Thanks!
thank you for doing the work of the retro-gods. bless you.
Good job and by the way it was a surprise to see the PB because I have a D130 model that only supports single voltage processors up to 200mhz, but it came with 133 mmx and works with overclocking up to 166, but the mobo does not support external VRM.
I'm afraid 133 MMX doesn't exist. MMX starts with 166 MHz and is a dual voltage CPU. If you have it in a single voltage board, then it is over volted. Some CPUs are robust enough to handle that with good cooling, but in the long term it will kill the CPU. As a rule of thumb MMX means dual voltage, non-MMX means single voltage.
@@necro_ware If I'm not mistaken, mmx was written on the back of the processor, I have a 166 mmx here and they are different, as far as I know up to 133 are ceramic, but as I restored this PB some time ago I could be wrong.
@@exitustec From 166 onward there were two versions of the CPU, either in ceramic housing, or plastic with metal shield. You can't tell by the housing if it's an MMX or not. All you can tell is that if it's plastic version, it's at least 166MHz. There the inscriptions are on the back and you can read MMX at 2.8V, or no MMX with 3.3V. The 133 MHz one always came in ceramic housing and is a single voltage non-MMX CPU.
amazing video and amazing work!
Cool video!
Amazing work. However, I always would prefer a super 7 board with integrated dual voltage support because they got higher bus speeds and sdram. But for older boards, it's a good upgrade feature.
Absolutely, that's why I want to do a big run between those older Socket 5/7 and Super Socket 7 boards. Later will be faster of course, but if we talk about retro hardware it's not so important right? If we have a 386SX we see how fast we can get that, despite that we easily could take a 386DX with much higher throughput. Furthermore, that's why I mentioned those nice OEM machines, which have proprietary hardware and a VRM option.
Since you are concerned about possible automated assembly in the future, you way want to consider adding optional pads for such SMD components on your next layout revision. Then you keep both sets of pads, and can still use through hole components as you prefer, while offering the alternative.
Very nice.
Nice, like the part on the soldering 🙂 This also works with hotair, although your approach might be better for even heat distribution.
Board can be easily designed to suport both SMD and THT parts (caps and inductor). You could then use (possibly lower profile) polymer caps and pre-made inductors.
why are people still afraid of smaller smd packages like 0603 or 0402 ? You and other members of the retro community actually inspired me with your projects, so a while ago I decided to do something myself and steal AC97 codec from a dead motherboard, along with its related components, to make a CNR audiocard (actually two :D) out of it. And transplanting few dozens of small SMDs, mostly 0402, wasn't as hard as I imagined. I'd say it's even easier to solder smaller packages because pads are so close together so tip of soldering iron could touch and heat them both simultaneously, and you don't need to move the tip a lot to re-align or desolder the component. Also I made a mistake and used 0603 footprint for 0402 details, but it was a good mistake - in many cases after tinning there was already enough solder on the pads to make a good and even beautiful joints with the component, whilst on proper sized pads it always needed a bit of additional solder after aligning the component to achieve joints that resembles ones on factory-made boards.
Yes sometimes SMD capatitors may crack when you trying to desolder them, or if you accidentally smd component on the floor its most likely gone forever, but if you do have a strip of new components it shouldn't be a problem.
In the retro community we have a lot of members who are not quite as young anymore. I'm very happy about it, but some eyes and fingers are not the best after all the years and it's easier to solder slightly bigger parts. That's why I also try usually to use through hole parts in my projects. Not only does it better fit the retro look, they are also easier to solder for many people. Another question of course why then I use SMD parts at all and, well, today it's getting hard to find required TH parts and of course they usually need a lot more space. So, bigger SMD parts are a compromise.
I'm a fan of 0805. It's the smallest size where (e.g.) resistors still regularly have labels, so you can see what they are, but it's enough smaller than 1206 to be meaningful. I'll often use 0603 too, as it can really help when packing components around dense pin-pitch packages, and it's not THAT much smaller, so it's not hard to work with.
My eyes haven't gotten any younger, though. Compared to just before The Great Plague, I've had to rely more on magnification. Even still, SMD work isn't really that hard.
I recommend ALWAYS buying the stencil -- which is usually only a couple bucks on top of a PCB order anyway. Spreading paste over a stencil is much faster, and yields much more predictable results. With a hot air station (with or without heating plate), stencils, and some good SMD tweezers, populating a board full of SMD parts is faster and easier than dealing with through-hole parts, and I've come to prefer it for both the convenience and availability of parts.
looks great. i wish i had the space and resources to keep vintage motherboards around, but alas, i'll have to settle for watching your fantastic videos
Some suggestions for a future Version: A OLED Display presenting vital Parameters as Wattage or Amperage. Also RGB LED Support would be nice as you already know RGB makes your PC at least 10% faster. For controlling RGB you would need an Adapter to be set between a RAM Slot and the module to be able to access the System Management Bus. . . . . . Just joking!
Sounds like a viable proposition. 😂
I have got the 0.4 revision in my IBM Aptiva running with a K6 III+ at 400 MHz with 1.6 volts very stable since 0.4 was released.
Thank you very much for the feedback. Would be nice if you could make a ticket on the github project page with some detail about the board, so I can make a proper entry in the list of tested boards. May be a link to the retro web with the related board and some info about BIOS and if you had to use a resistor mod for K6-III+
@@necro_ware done, no resistor mod for the K6-III+ was necessary
Probably the best Socket 7 VRM 'THAT' ever existed.
I'm not an expert in English grammar, but I think it's called relative pronoun, which can be omitted in this case.
@@necro_ware I'm not an expert either, but I just call it lazy.
@@paulmurgatroyd6372 lol, let's call it efficient ;)
@@necro_ware "Possibly the socket 7 vrm of all time"
It's all very clunky and I can't approve of it! 😂
I ran a computer shop from 1997 to 2000. I have a long list of really clunky hardware, but I never encountered a VRM module until Necroware made one
Just one of an infinite number of features, sockets, and jumpers that were provided by an engineer and almost never populated, documented, and thus ever used by anyone.
Again I've lernet a lot! Thanx!
i love this
..
These heater standoffs do not fill me with confidence! I would have an urge to give it longer ones and then put some heavy steel sheet underneath to make it stable, and rubber feet bottom; also would run a third wire to earth the plate surface and the whole external metal basically. I know it should be packaged well inside so it should be "safe" but i'd rather be doubly safe and not just single-safe.
By all reason it doesn't need temperature control since the heater is PTC, so it should balance out around the target temperature. So other than the construction sketchiness, it should be perfectly good.
Yes, please, please, please add a third earth wire to ground the plate surface! Do not rely on there always being adequate electrical isolation between mains heating elements and the plate.
amazin
I thought you were about to bring out a portable stove as a heat plate but somehow, you found sketchier 😂
The IBM Aptiva 2176 has a VRM socket. I had a hard time tracking down one of the original PC55 VRM modules to upgrade the machine to an Pentium 200 MMX CPU. Annoyingly the BIOS does not recognize any MMX CPUs and complains at every boot up, however skipping the error then just allows the machine to run without issue! With a modified BIOS and your module, I imagine the 2176 will happily run faster than 233Mhz, the main issue with going any faster might be the lack of necessary jumpers to set the desired speed.
I'm an electronics engineer and have worked with computers and software that I "don't" have a clue about my entire career, how ironic. Now I'm retired and started tinkering with electronics again as a hobby just like when I was a teenager and I have as much to learn as you, electronics these days are not the same as they were then. Like for example, I was looking for a regular 100mA NPN transistor at Mouser and found that I'd better find another one because the one I was looking at was 1mm. (0.039 in.) on the greatest length. I shake my hands too much so I have chosen that my lower limit for now is 0805.
That's why I make my projects using either through hole parts or bigger 1206 SMD with pads optimized for hand soldering. I want, that even people with minor handicap are able to build those projects.
3:40 - If I could make one small suggestion...
With the use of the Sense pin, my only worry would be if that loses connection somehow,
(ie. if somebody is moving the module around while it's powered, or if header pin on the motherboard was dirty / corroded.)
The Sense signal (and FB signal) on the module could default to (close to) 0V.
The switching reg would then think the output voltage is too low, then output too high of a voltage, and possibly damage the CPU.
(R3, and the DIPSW resistors will also be acting as pull-downs on the FB pin.)
To test the theory, you could remove R4 temporarily, but beware of the higher output voltage.
I would try adding a weak pull-up resistor, between the Feedback pin and VCORE, just in case.
Maybe something like 47K ? Not sure if that would be too high?
Actually, I'm not sure how to solve that, with just a pull-up resistor to VCORE. lol
Because the 47K resistor would be as if you used that in place of R4 (which is currently 5K).
R3 (plus the DIPSW resistors) would be still acting as the voltage divider, for the FB pin, but giving a higher output voltage.
(due to the larger 47K pull-up between Sense and VCORE).
(47K was just an arbitrary value I chose.)
It's probably fine as-is, tbh, but maybe a future improvement could be some type of overvoltage protection.
That would need to measure the voltage difference between VCORE and the Sense pin.
If that difference is too high, it would trigger the Disable transistor.
Maybe a simple diode between VCORE and SENSE could work?
That way, if the voltage difference between those two pins went above say 0.7V or so, the diode would still give some feedback to the FB pin, in the event the Sense pin on the mobo had a problem.
You would need to measure the average expected voltage difference between VCORE and SENSE, on say a power-hungry CPU.
If the maximum expected difference is less than about 0.6V, then a normal diode could do the trick.
(Anode hooked up to VCORE, Cathode on SENSE.)
If the max voltage difference is higher, maybe a Zener diode (connected in reverse to the above).
This is an interesting idea. The diode could negatively influence the voltage divider, but I will think about it.
Yea Tech Tangent Necroware Covered You For Upgrade Micron Melinia (Put VRM On Micron Melinia)
So glad to see someone who actually knows the steps in through-hole soldering: set, trim, solder. Trimming after solder weakens the joint, which can go cold under thermal variance in use!
If your ripple is ever causing stability due to load-stepping, you should likely look into a 'feed forward' capacitor, which is basically just a tiny capacitor that bypasses the feedback resistor (your 'R4') for faster transient events. Looking at the datasheet for that part and the project, I can see there is a compensation network, but that is for loop-stability and cannot functionally work as a stand-in for a feed-forward capacitor. Because what a feed forward capacitor does, is make sure the regulator gets a more dramatic start on adjusting to a transient events, as an impulsed load forces the sense line to behave more like a transmission line. (suddenly the slow ramp to adjust to, is turned into a square wave)
That becomes more important as your 'voltage sense' path starts getting more complex and especially parasitic capacitances after any substantial resistances. This is what your "R4 - 5K" is! So essentially the copper on the "FB" node that goes to the switch bank, are subject to creating a parasitic RC network.
There's actually a comprehensive discussion of how to approach compensation in the ISL6545 datasheet. I find it hard to believe that among C1, C2 and C3 referenced there, none would be tackling parasitic capacitance at the FB pin along the way.
Now the crux is that any such Cp would effectively be in parallel with gain-setting resistor Ro, so it'll be very tricky to effectively nuke this pole across the entire range. It may have been a good idea not to be choosing Rs (= R4) at the very top of the recommended range of 1-5 kOhms. Ro is in the 1-3 kOhm range as-is, which doesn't exactly strike me as super low. It should still be moving that pole out to >=10 MHz when faced with Cp = 10 pF, and my understanding is that even Fp2
@@PileOfEmptyTapes I had to think on this a little bit, but according to the datasheet "The goal of the compensation network is to provide a closed loop transfer function with high 0dB crossing frequency (F0; typically 0.1 to 0.3 of FSW) and adequate phase margin (better than 45°)" I still have to go with the opinion of it can't fully compensate for information it doesn't have. The information it doesn't have, is what gets lost to those parasitic capacitances due to the 5k RC. I think a step response is precisely what slips past their definition of "better than 45°" because until they can definitively close that loop delay to 0° a step can still create a ripple. The point of a feed forward capacitor is to help close the phase-delay gap specifically for step response. Though also I might note that just putting a cap across the 5k might not be the best approach, you'd likely want to also run those both into a series 0-500 ohm and adjust the 5k to compensate for that ratio adjustment(a 5k or lower potentiometer might be a clever way to do this without needing anymore parts swapping), that way the feed-forward cap doesn't dominate the feedback network. I've never been amazing at pole filters etc. But I don't think a compensation network can create compensation for a step response it never saw coming, if that information is lost before it enters the comp network. And anything behind the 5k resistor has now already lost part of the transient impulse that has occurred. That's how it makes sense in my head at least.
I have collected 2 hp vectra pentium powered computers. Both of them have Vrm socket support for pentium mmx processors. I'm interested in seeing what the hp bios will support. They are 430hx and 430fx chipset. Hp also had a sis chipset in socket 7 in the vectra.
How can I get my hands on an assembled module? My skills are not ready to solder like this yet.
I'd probably mod that dodgy hotplate and add a fuse inline or something just FYI.
Yeah I used it through a plug with integrated protection. This thing is excitingly scary :)
Any thoughts or plans for possibly doing Slot1 Versions?
Currently not, but this design can be easily adapted for Socket 8 and Slot 1 mainboards.
17:53 I always wonders what that slot on the MB was for. voltage regulator module
I have been looking for the UtterCrap 3000, and I cannot find it for sale anywhere. Is it a discontinued item?
LOL :)
Do you have any old vrm boards from back in the day that you can compare to your version?
No, I don't, but the old ones were either linear, or very limited switching regulators, which only provided one voltage 2.8V. So, even if I'd have one of the old exemplars, there wouldn't be much to compare.
Ah, in that case it sounds like yours may be the best Socket 7 VRM that ever existed ;)
How would.u know what oem motherboard can use the vrm
If it has a VRM expansion slot, then it's a good candidate.
Осталось только выпустить более последнюю ревизию с 4pin 12В разъёмом питания CPU, для совместимости с блоками питания стандарта ATX12V 2.X, может даже ATX12VO!
Why not Multi Phase Controller + Phases?
Those very modern computers wilil help mother rusia to get into 20th century finally.
Awesome :) its a microscope recording at 16:18?
Thank you! Yes.
@@necro_ware wow what kind of microscope is that I don’t think I’ve ever seen one as good image quality as that 🤩🤩
Could you make one for a pentium pro?
May be one day. Shouldn't be too complicated to adapt this design.
I hope that when I say this it doesn't sound rude but I was also hoping that one of your announcements would have been that these can be ordered already made. I would love to be able to get one of these for an older motherboard that I have that has one of these VRM socket upgrades but my issue is is that I don't know where to buy all the little chip parts and I have no soldering skills. I need to find someone who can actually sell one of these especially latest version already pre-made so I can buy it. Again I hope I don't sound rude in any way. I just need help.
I would like to sell those things, but German laws are unfortunately complicated and are made to kill every inventor asap. I'm currently trying to organize some kind of official tiny online shop, but so far it looks like it would be easier and faster to buy it from someone else.
@@necro_ware Would you know anyone that sells your completed project. I am just stuck.
@@clintthompson4100 I don't know personally, but I already saw couple of time my projects been sold on ebay. Also the last revision of this VRM. Most probably the latest revision will also appear soon.
Oh no I miss the premiere!!!
Why not use some IC with integrated power stage? Like AP62600? Small, high frequency (smaller inductor, smaller capacitors, SMD only), 2 channels will not be problem.
Because it's too weak. The revision 0.1 also used an integrated solution. It was not AP62600, but the point is, that it can deliver only 6A. That is at least half as much as needed.
@@necro_ware Running a higher frequency controller would be helpful though. The one criticism I have about this module is that it's awfully chonky. You could use much smaller SMD parts with a faster controller.
Also, I haven't looked at your schematics, but please tell me you have a weak resistor between the Vout rail and FB pins, and not relying solely on the path through the CPU for Vsense? If anything should ever happen to any link in that chain, you really want to have a fallback path to keep it from going open-loop.
@@nickwallette6201 You are right, it can be improved further. Like everything can be always improved. And after it is improved, it can be improved even more. In my career as a software engineer where improvements can be done even easier than in hardware, I had to learn to tell myself when something is good enough. This thing now is good enough for me. If someone wants to improve it even further, it's open source. In regards of sense resistor, I know what you mean and that would be not good. Luckily it is not possible to connect Sense directly to FB. It has to go through a voltage divider, hence there is always 5K in between. And if someone messes with Sense outside, it's really not a fault of the adapter. Eventually doesn't matter how much effort you put in your stupid proof system, universe will come back with an improved stupid :) Don't remember who's quote it was, but it's unfortunately still valid.
@@necro_ware But would you still lose the other half of the divider? E.g., consider what would happen if your Vsense pin is damaged somehow -- corrosion, bad fork in the socket so it doesn't make contact, solder failure, etc. What happens to the output voltage in that case? A good remote sense circuit will always follow the local rail through a resistor that is overridden if the remote sense is present, but provides a correct (albeit uncompensated) output even if the remote sense is entirely missing. That way, you mitigate the risk of ending up with 5V on your Vcore.
Just food for thought. :-)
And yes, I know perfect is the enemy of good with any product. That's totally fair. But seeing the height of this module immediately makes me wonder whether you'll run into clearance issues with, e.g., AWE32, or other cards that are long enough to hang over the CPU area of a motherboard. Probably not an issue for a good 80% of users. Just worth considering, if you ever decide to hack at this design again. Although, layout gets increasingly important at higher speeds. Everything is a compromise.
So I hear you like capacitors
I would personally recommend just getting your boards assembled at one of those known PCB places. A board like this would probably be around 15 euro delivered. Probably even less if you optimize the design.
In that case I would also make all parts SMD. (Except for the jumpers probably)
If it's already tested and released part, which you can order in a package of 5 or more, probably it's worth it to order pre-built modules. However, during a development of a prototype where you want just one piece to test and improve it's a lot cheaper to do it manually.
@@necro_ware I do it on a regular basis. For 40-70 bucks delivered, I can't be bothered even as a hobby.
Especially when you consider that you have to order the parts etc
Anyway, it was just a suggestion 👍🏻😎
Yeah I tried that, but it was too expensive for prototyping. Especially I found it disappointing, that the selection of parts they have for cheap assembly is very limited and if you need something what they don't have in the standard set, they have to order it for you off site and then the price explodes to over 50€ per module. Quite a lot of money to get one and realize, that you made a mistake and throw it away. Of course for the revision 0.6 which I opened people can easily order pre-assambled modules. They should be fine.
@@necro_ware if you use jlcpcb. You can order anything from lcsc.
What I often do, is hybrid assembly.
Meaning, getting all the passives and connectors done and doing the critical parts myself.
I think this board can also be optimized. I might have a look, just for fun, but I am a little busy atm.
Far too many projects myself 😁😜
Still really enjoy what you're doing btw 👍🏻😎
However is your favourite word lol
Yeah, I use it far too often, however it's a nice word indeed ;)
I have a tip, I noticed that having a Solder Fume Extractor no matter the brand as long as it sucks out the fumes it actually advances the soldering process, I don't know if you noticed that, having more air flow/oxygen over the area being solder it gets the job done faster. PLUS the benefits of not getting fumes in your lungs.
TBH I planed to do a fume extractor since a long time, even bought a fan and carbon filters, but it's in the TODO box since forever. I always solder with an FFP mask telling myself, that it helps. Probably it does a little bit, but having a fume extractor would be even better.
Hello.
So this is the thing, that goes into PCChips M520 next to the socket?
I never knew, for what is that thing, but I saw some jumpers in it. I was told, I should not touch it, if I dont understand it.
So it is needed in that board only for Pentium 133 and higher? So without it, I can use only Pentium 120 and lower?
It is needed for dual voltage CPUs in the first place. This would be Intel Pentium MMX, AMD K6-2 etc. Any non-MMX Pentium is a single voltage CPU and doesn't need such a VRM. You can theoretically use even Pentium 200 (non-MMX!) CPU in your board, but at that speed the linear voltage regulator on your board would get extremely toasty. Using this VRM would not only allow to use a lot faster dual voltage CPUs, but keep your system much more efficient and cooler even with single voltage models.
@@necro_ware Thank you. I put it to notes.
But what about year 1996? Did they sold these voltage regulator modules separately? Or computer shops used them, when there was Pentium 166 MMX built inside computer? Because, I've never seen some prehistoric ones, like from 1996 or 1997. Anyway, board was released in 1996, Pentium MMX came in january 1997. So seems, PCChips knew about it, that it will come later next year or what.
@@warrax111 They did exists, yes, but they were not nearly like this one. Usually they've been sold for 2.8V MMX only with no other voltage option. Some had couple of jumpers to select a narrow voltage range between 2.8V and 3.2V, but all of those were not powerful enough and had no option for 2.2V or lower, which K6-2 needed.
What is the Current draw of the CPUs ?
Asking since you are using a very old and slow chip and there is a good chance you can make it all SMD with a newer controller.
maybe for a future Socket 8 VRM version?
Anyway great job and if you want electrical help just ask.
Thank you very much for the offer. And yes, I could use a faster controller, but it would be probably an overkill. The current draw doesn't go above ~12A at
@@necro_ware Well i am coming from the electronics side and i can tell you this:
- synchronous is like the normal for high current for obvious reasons....
- a lot of chips these days can do 2MHz+ best i saw was 3.4MHz if i remember well
-i didnt say you should but here is my 2 cent : if you increase the switching frequency you can decrease the size of the magnetic s and caps , i sort of felt strange when i saw so many electrostatics out there , i am not saying that it is incorrect but odd to me
-you have a single buck phase* there if you look at modern motherboards you can se 8-12 phases , powering even a slow* weak modern CPU or MPU is a bit harder these says but progress has a cost
Anyway if you need help ask , you also inspired me to think of a ho small it can be made with modern chips.
Also keep up the great work!
PS ironically i work in electronics and looking at programming and computers as a hobby LOL
How about making a Socket 8 VRM next? ^_^
It should be easy to adapt this design to Socket 8, we'd just need some more caps and probably slightly smaller coil.
oh god, I also bought the utter crap 300 and mine didn't even come with the europlug wired up, it's even worse than yours!!!
Surprised you didn't go full modern with the same mosfets as modern mobos/gpus
They are hard to solder. I try to keep my DIY projects easy to solder. You can build this whole module still using a simple soldering iron.
Well, watch the video now 😉
Part of me is really bothered when the first release version of a product isn't 1.0 but oh well 😅
Yeah, it actually doesn't matter. 1.0 sounds like it's well tested and feature complete. With this module we have the feature completeness I would say, but it needs to be tested by a wider audience. I just don't have the capacity to test it with so many boards and CPUs. Once I got some reliable feedback I can release it as 1.0
LOL I remember when the Linux kernel was 0.97, 0.98, 0.99 then they just kept adding a patch level number to the end. People wondered if it would ever reach 1.0 😅
Necroware i need your help with my GPU. what can be potential causes of a GPU that randomly stops working and still looks new? The GPU is an AMD RX 6600
häää????
häää????
Prost!🍺
@@16_bit31 🍻
(Digital: venturis FX) pc has VRM socketed. that would be nice to see in full action cause it can take 192mb ram and has onboard s3 graphics (2mb upgradeable)
Yeah, indeed. Just searched it and I see, that it has SiS chipset. I have not one SiS board with a VRM option.
i was checking and there are 2 different models on ebay on farely low price (normally i have seen them go more than 1000 euros each)