Hi there! I hope you like the 30-pin SIMM 16MB memory module. Would you be interested to get a set of 2, 4, or even 8 modules? I want to understand what the demand is for such modules. Hit the like button of this comment and reply with how many modules you would be interested in. This is not an order, I just want to know if I should make a few and utilize the memory modules you have seen in this video! Thanks everyone for watching, commenting, and liking! Cheers
This is awesome. I could use 8 such modules for my Macintosh SE/30 to max out the memory :) Btw, the Sound Blaster AWE32 also supports 16MB SIMMs, but the usable memory is limited to 28MB due to the 4MB ROM sharing the address space with the RAM.
I'd absolutely be interested in a set of 4. Being able to switch between FPM and EDO would be invaluable for testing old hardware. Also can I please come visit your scrapyard one day? That place looks like heaven! :)
The max amount of memory I had in my 486 SX2 in the 90s was 20 MB. It had come with 8 MB across 8 slots and I had pulled 4 of those and changed them to 4 MB modules. The CPU I swapped with an Evergreen upgrade.
The project is very nice 👍 Really great, especially since it is very difficult to get really large modules. I also have a 386DX 40 MHz with 64 MB, but I bought these modules ready-made many years ago.
I was not a PC guy, more an Atari, ut at work the 386 that was assigned to me had 8 MiB and it was more than adequate. With DOS and a little bit of Windows 3 there was no point in much more memory. Only at its life end would a little it more memory be necessary when I started to do some AutoCAD. AutoCAD 9 was the maxi I could use on that 386. Acad 9 was limited in its Lisp implementation to 1 segment (64K) and as I did a lot of lisp I would have needed an upgrade to Acad 10 but the 8 MiB were too small. Upgrade of the whole computer was the better choice. I got a Pentium 90 PC with a 2nd CPU socket (but initially only 1 CPU) with Neptune chipset and 6 SIMM72 slots for FPM modules. At the end of its life I managed to upgrade the machine to 2 Pentium MMX-233 (with adpater sockets) and 80 MiB of memory. I even had used SIMM30 to SIMM72 adpaters in it. That insane machine also had 1 IDE disk only for ooting and 4 16 bit SCSI harddisks, 1 SCSI CD-ROM, 1 SCSI burner and an external Zip-drive and SCSI HP-Scanner. The motherboard was oveclocked from 33MHz to 35.2 MHz (70Mhz oscilator to replace a 66 Mhz), not much but more was difficult for the EISA slots (yeah I forgot, it had EISA and PCI slots but no floppy/IDE on board). I should try to revive it and film for yt.
I wonder how many of those SD-like modules with EDO memory chips exist. I have never seen them before. Just once, when I visited the scrapyard, I found all those modules and was like 'eh, let's take them. Maybe they will be useful in the future'... And here we go! I hope you can make good use of the PCBs!
Nice! I thought I was special with my 4MB modules :p Have you tried to tighten the memory timings? Those timings are probably the most important feature of EDO memory :D
I haven't done much with those modules yet, but I'm planning on doubling the cache on the board and also exploring what's possible with that amount of memory.
I had a 8088 computer for way to long. I was young and had no money, so I saved all the money I got to upgrade my XT. The day I got enough, I bought a motherboard with an Intel 386SX, and 1Mb memory. The motherboard had 4 SIMM slots, and I wanted a single 1Mb module, but on the store they told me that I had to fill at least a bank (2 modules). My options were 4x256Kb, or 2x1Mb. No 512Kb modules. Having waited more than a year to be able to update my computer, I caved and got 4x256Kb, my options to upgrade gone. Then I had a 386sx (I think 25Mhz, don't remember), 1Mb of RAM, an Hercules graphics card, a green-phosphor monitor, a single 5,25" 360Kb floppy disk, and an MFM 20Mb hard disk. It wasn't much, but Windows 3.1 worked well and the CGA software emulation worked SOOOO much better.
When you started up that 386 for the first time with 64MB I had a huge smile on my face as I'm sure you did. What a really great achievement and just shows you that "official" and "supported" specifications may be two different things! Thanks for sharing.
I am working on an updated version that hopefully follows the specifications on the datasheet. I do it mostly to learn more. The current modules seem to work well and I got plenty of comments supporting both perspectives. There will be a follow up on this one very soon!
The counting algorithm was also the same, but CPU performance grew, so on a 286, it'd slowly count up to 1MB with a tick...tick...tick, a fast 486 does its 32MB in a "brrrrrrrrrrp" in a few seconds, and some late Athlon counts 2GB in a brief high-pitched "zeet". I guess if contemporary ones still did the same, it'd just be in the ultrasound spectrum and thus inaudible to mere humans.
@@Armand79th Maybe you have the fast boot option enabled in your BIOS. It cuts down startup times by a substantial amount, by skipping most of the usual tests you usually have at the start of the computer.
Congratulations on such a successful project! Now people can finally stop bugging me to make 16MB SIMMs! Of course I'm kidding, I only got a few messages asking about them since I published that project so many years ago. But I'm glad someone finally took up the mantle and did what I never got around to doing. I had every intention of getting back to this project and making my own 16MB modules, at one point even tracking down a source for the requisite chips, but they were very expensive since they were NOS, so I never got around to ordering them. And then my career took the front seat and I've had neither the spare time or energy for such a project since then. But this is an even better design, since it's using chips that are basically scrap. The fact that you can use EDO chips really opens the door to potentially using other chips with a relatively simple modification to the design. So congratulations once again, you've really done some impressive work!
Thank you so much! Without you paving the way, I would have never picked up on PCB design. Your Gerber files and schematics allowed me to learn about those modules. Thank you for your work!
your PCBWay ads are the only ads on YT that i actually watch. not even just letting the ad roll while i do other stuff, you have me paying attention and engrossed in the product. can we please make all advertising like this? relevant, informative, confident.
Oh well, thank you 😊. I usually get the opposite view about ads in general. But without PCBWay, I wouldn't be able to do all this. Not only do I get access to their services and products, but they also help a lot with their financial support. It wouldn't be feasible without sponsorships to run a channel like this. Thanks for your support! 🙏
I agree - 64MB is not really useful for a 386. Most of the memory is not usable by the software that runs on such a platform. As you said, the driver behind this project was to see what is possible. I learned a lot by going through the documentation and make sense of it. And finally also to understand, why we can only use those high capacity modules in memory bank 0.
@@bitsundbolts RAMDisk would be very useful. Large smartdrv cache. But back in the day, this kind of system would have fetched a very high price. You are right, RAM was expensive!
@@bitsundbolts Back in the day I was super happy when I got my P90 system to 8mb. I put 4mb to a ram drive and had the autoexec.bat copy my current game there. 64mb would have been heaven. 😵
In 2000, our home PC only had 32MB RAM, 3GB HHD and 300MHZ AMD K6-2 processor (586?). The fact you were able to get twice the ram on a system at least two generations older is impressive.
We had a k6 too, classic. I still have the chip as a memento. I also remember my cousin had a 486 amd I think, the motherboard had the option of 30pin or 72pin, if I'm not mistaken, because I definetly remember those 30pin RAM sticks and I always found them slick and clean compared to the bigger ones.
@@bitsundbolts Yeah that's one part of the 'cutdown' part of a low-price 386 alternative to what then got called the 386DX, only a 24-bit address bus was exposed on the 386SX. Internally the CPU could act like it had a full 32 bits of address space but physically impossible to talk directly to more than 16MB. As was invariably the case from SIPP to SIMM to DIMM to SODIMM, machines may be designed to take a bigger stick capacity, but 2x the RAM would cost so much more than double the cost of the smaller sticks. They were at first only for hardcore uses, then as the line aged they became a bit more affordable, and then technology had moved on to another speed/bus/etc arrangement so they never became commodity parts because the next high and mid range systems used the commodity new thing instead.
Years ago I ordered a number of 128MB 5v EDO DIMMs and the supplier shipped 3.3v modules instead. When I told them about the mistake they told me they'll send the correct replacement and to just keep the 3.3v sticks. Never had any use for them until now it seems. My chips are Micron parts but they appear to be 16Mx4 as well and are sharing the same pinout. Guess I should sacrifice a stick out of my pile to create a slew of 16MB 30pin SIMMs. It's a parity module too so two additional chips for an extra 30pin module... Awesome job!
Oh, wow. I always have to send stuff back 😅. Good for you - I guess you found a use case for those modules now! I wonder what you possibly could have used those EDO modules for otherwise (I guess you have the 168-pin version). I would rather use SD-Ram since it is probably a lot faster.
When the 386 counted up to 64MB the first time, I had to laugh like you did. Insane idea and great implementation! Another question: I am an earlytime member and from all the videos I have watched I might think you are German. There are some hidden words and things you did that make me think that. Am I wrong?
Yes, I am German - you are right. Listening to the memory test counting up is just great and I am glad those PCBs worked! Although, after making the 4MB SIMMs, it wasn't as challenging anymore.
Back in 1990s I had a borrowed Am386DX-40 machine with 4MB of RAM (and without FPU). It run Win3.11 For Workgroups on MS-DOS 6.22. There was Corel 5 and Corel 7 installed. Corel 5 was working fine, Corel 7 had trouble fitting in RAM and was swapping a lot. It was fun seeing Corel Draw 5 actually drawing the business cards when a file was opened.
Oh, I remember Corel. I think I had version 4. That was for Windows 3.1 if I'm not mistaken. It was slow if I remember correctly. Those vector graphics just took some time to be drawn on screen.
They are no longer that common. I have many and not really any use for them. I prefer larger capacities. However, if you want to build a period correct system, those 1MB modules are probably the sweet spot.
@@bitsundbolts I had never even seen one and wasn't aware they existed, I was going to make it one of my retro machines and max it out with what I could find. Once I fix the board though.
@@bitsundbolts It's weird to see "period correct." I'd say who cares. But the there's automotive, where original, oem, has the most value. There it makes sense since most can't build better than the manufacturer, and no one wants to buy junk used. I think we had 1mb in our 386 back in 92
Please note that there have been reports on similar chips dying in 5V systems. On the other hand, I have a few SIMM72 5V modules made with 3.3V-only chips but indended for 5V boards. They don't even have the regulators, just two diodes in series to drop the voltage to 3.5-ish. Also there's another way of making 16M SIMM30 sticks. It involves 8 4x4 chips and a PAL chip for reorganizing the memory. Works kinda like SimmConn. I do have the PCBs drawn and firmware written for PAL but just don't have the time to build it 😢 Great video as usual!
Hey! Good to hear from you again! Thanks for pointing out the issue with the 5v issue. I'm going to research the topic and try to address the issue somehow. I have seen other SIMMs that also just have the voltage regulator and a few capacitors. I'm confused to be honest. But I'm researching level shifters at the moment.
Wait, 20 years ago I had 96 MB FPM in a 486 system. I was moving about a lot and at that time, got myself a P4 or whatever and stored that 486 back at my parents'... so my mom promptly recycled it. WOMEN! I still can't forgive her, for this and also my pimped out 386 router with a Hercules console :( Anyways, 32 MB FPM sticks not only existed, not only I had them, but they were scrap when I got them . The world already moved from EDO SIMMs to DIMMs, I was simply too poor for any of that and had to actually research what was actually fast, and my resulting 486 pwned!
Fantastic work! One potential problem, having to fit 3.3v regulators for 3.3v RAM, the RAM may have a shorter life as any clamping diodes are driven frequently by 5v logic levels of the chipset. It's also possible the chipset could die when interfacing 3.3v and 5v. It's possible the RAM you are using is 5v tolerant? I see you scoped the address and data levels there, and they look right - but probably because of the clamping. It's worth measuring that same way with 5v RAM fitted - see what levels you get. Don't be put off by this btw - the chances are it will work reliably even if the clamping is an issue. It's just one of these things that I am sure will get pointed out.
The ram I'm using is not 5V tolerant. In the data sheet it says absolute maximum voltage is 4.75 or something. I am wondering and will do some additional tests before starting a mass order of those PCBs. In case there are 5v signals, would a simple voltage driver be enough?
@@bitsundbolts What you might be able to do - since it says the RAM can go up to 4.75v, try measuring the data and address bus levels with a 5v SIMM, lets say the levels are around 4v - that would mean you could perhaps try and regulate the RAM VCC to 4v ish, and in theory the RAM would work safely as its under its 4.75V limit, and clamping would be (I think) less likely as the logic levels are within the VCC level.
I wondered same issue, chip lifespan will be reduced with 5v operation. I am not sure how tight timings on address bus but maybe level shifters are fast enough to use on this modules?
I'll try to understand this more. I already got details regarding level shifters. More things for me to learn and definitely a thing to look at before I order larger quantities of those modules. Thank you for all the great hints!
64MB of RAM on a 386 is quite amazing, nice job. One remark/question: the memory chips are designed at 3.3V, for which you added a Vreg on the power rail. But the address- and data-lines are still driven by the 5V chipset. No risk in damaging the chips on the longer term?
I tested the data lines with an oscilloscope and they don't seem to go above 3.4 volts. However, I wasn't able to check the lines that are facing the board (due to limited space). The chips have a rating of up to 4.75 volts as "absolute maximum" - so, there is some margin. Maybe someone with more knowledge than I have will be able to confirm the long term implications if such chips would be exposed to higher voltages (e.g. on a different motherboard/device).
Early PowerMacs and IIRC some workstations used 168-pin 5V FPM DIMMs, so the chips on those would not have any long-term issues on a pre-socket 7 motherboard.
SoundBlaster AWE32 (and AWE64 IIRC) supported 2x 16M modules, but could only use 24MB of RAM for soundfonts. Back in 1990s this was The Ultimate MIDI Setup. SB-Live! / Audigy soundfonts will work with it too. (yes, yes, didn't forget, there was GUS, and Holy Wars wars between some GUS and AWE users, I know ;)
The SoundBlaster AWE32 was released in 1994. But later soundcards where much better MIDI setups. I had a Terratec EWS64XL in around 1997. This was still in the 90s and the EWS64XL was much better than the earlier SB AWE32 and still an ISA card for best compatibility to DOS games.
@@OpenGL4everIn fairness, with full ram you can get very nice midi quality out of the AWE32, so long as you chosen DOS game will run under Win9x. Depending on the model they can also have real OPL3 FM, which is a nice perk as well.
My old Packard Bell 486 started out with 4mb, but by the end of its life had 32mb. Since it was all low-capacity SIMMs, I had to use several of them per slot via expansion boards. This was def more than enough for Windows 95, but then I made the mistake of installing Win98 on it…
Ohhhh, I remember when I installed windows 95 on my 486. It was such a great experience. I loved it, my PC felt like new. I only used Windows 98 with a Pentium II 350. I never attempted to install it on my AMD 486-DX4-100.
@@bitsundbolts By the very end I did upgrade it from DX2-66 to DX4-100, but even this didn't make and difference for W98 performance. Should've just downgraded it back to W95. Back then couldn't figure out how to upgrade the video card - would've been fun to see how far I could push it: )
Related to these 30-pin SIMM memory modules, I am very curious on this Socket 7 Motherboard model by GCT: GCT-8IV. According to The RetroWeb, this motherboard has the ability to utilize 3-4 different types of RAM (the manual debates this) and can run a k6-III and Pentium MMX 200-233 mhz processor. Can you do a video involving this Socket 7 motherboard? On paper, it sounds like a rough diamond but is it? 🤷♂🤷♂
I had a 486 DX2-66 machine with a weird motherboard that had 3 72 pin SIMM slots. I ended up populating it with 3x16MB modules for a total of 48MB of capacity. It was a fun machine to put slackware linux on.
I thought it was common knowledge that EDO memory supported FPM. I remember back in the day using EDO SIMMs in my Acorn RiscPC which only supported FPM and EDO was much cheaper at the time!
This makes me wonder what kind of memory was in larger x86-based systems like 80386-based Sequent Symmetry and PPro ASCI Red. 64MB on a 386 enables some interesting benchmark comparisons vs contemporaneous systems with MIPS or SPARC CPUs running at the same clock rate.
I once had a Socket 3 system I maxed out with 128 MB of memory. It had a Shuttle HOT-433 motherboard. I also had an AMD 5x86 overclocked to 160 MHz in it. Why did I ever let that computer go? 😢
Amazing!! :D That Cyrix FasMath FPU is gorgeous. I also find those COB SIMMs you showed at the start of the video pretty interesting! It's been a long time since I last played with my 386s, but I remember 4MB being the most I could use... So I think i don't have any module larger than 1MB (and there's also the possibility that I achieved said 4 megabytes on a board with 8 sockets and 512KB per module being my maximum!).
Considering my first PC (was my mum's back in the late 80's/early 90's) was part of the transition period of 5.25" FDD to 3.5" (and everything was installed to 3.5" floppies instead of the 80MB HDD) - I'd have been lucky to have 2MB of RAM. Not sure about the exact amount as a rather technologically ignorant family member killed the PC by accident when moving in the early 2000's, so it has been gone for a very long time (excluding the still functioning HDD + dual gameport card).
After watching this video I realize that I have 2x EDO memory sticks (probably each 16Mb), that I storage a few decades ago. I can give it for free if someone needs it. In the chips are written CW424410-7 9528 and 9519.
Have you ever done any system design for S-100 bus computers? I would love to add every modern feature that didn't exist in 1977.. SIMMs, M.2, i2c, USB, WiFi, BT, etc Make it run in 2 different modes... and make it backward compatible with original pinout make it run to take full advantage of all the bus pins.. example the data bus had 8 input and 8 output data lines. Would it be possible to make it 16 bit bi-directional?
I had a Compaq ProLinea 4/25S. Had 4 72-pin slots. 56MB total… Config; First slot could only be 4MB as there was 4MB soldered to the board, the other three slots had my 16MB sticks. Also had a Kingston Turbo Chip 5x86 2 133MHz installed.
Although I am not a Linux user, it is interesting that this amount of RAM would allow to run software that would be difficult to use without. I may give it a try. I remember trying out some SuSe distributions long time ago, but I don't remember what version (maybe 7?).
@@bitsundbolts I just remember that Linux kernel drop 386 support in 3.8 version so last working kernel is 3.7.10 in theory all distro before 2013 should work. Also I personally want that you run Windows 98 SE and Windows 2000. I know that Windows 2000 work with 486 and 64mb ram but don't know about 386.
@@bitsundbolts I used Linux Slackware 7.0 on my 486DX with 8 MiB. Starting with Slackware 8.0 with kernel 2.4, I had to bake an extra kernel so that I could boot to the command line with the 8 MiB. I couldn't install Debian or Suse, the installer required to much RAM.
Seriously I am so sad that I gave away my 486 when switching to my CyrixP5 in 1996. But at work I used a DX4 100 with 48 MB EDO RAM up until 2001 (running NT 4.0)
Neat. I managed to find a 486DX2 66MHz Compaq a while ago and piece together 100MB of RAM for it, maxing it out. I haven't booted it up in quite some time, and I don't know if I have a monitor with a VGA port, or enough converters I can plug end to end. 😅 I inspected it not that long ago and all the caps look ok, and it's been kept dry and clean so it should work. 🤷♂️😂
I had a 386sx25 back in early 90's. I had an ISA card that had 16 slots for 30 pin memory modules and I salvaged a pile of 256kb modules. It gave +4 MB of total memory in addition of those 2MB already on mainboard. Interesting video though.
I think the max I had in 386 or 486 systems was 4mb. I don't remember if the "5x86" and similar chips from Cyrix and the like that I had counted as 486s or not, but those I had total of 16. I don't remember if it was late Socket 7 or early Super 7, but I had a board that could take either 2 sdram or 4 72p simms, don't remember the processor, but I had 4x16 for 64mb total for that, the highest I've ever had in simms.
I wonder, would it be possible to build an adapter that converts 72-pin EDO to 30-pin FPM without desoldering? I'm imagining two possibilities, one where you utilize only 1/4 capacity, another where it splits into 4.
So, here comes the question that bugs me! I honestly never heard of level shifters. But a quick search on Google brought some light into that mystery. My board seems to access those chips using safe voltages, but other devices may not! So, I am interested to know more about it. So far, the memory chips seem to work great, but I'd like to improve the modules if it can be done. What should I look for? Level shifters on any data and signal line like address, CAS, RAS, WE, OE, etc?
@@bitsundbolts I am not an EEE but one thing that I have learned is when connecting 3.3v devices to a 5v system you should use level shifters in-between to prevent damage over time to both sides. For example old 5v video game systems the NES, some people were building flash carts using 3.3v flash modules directly connected to the 5v address lines of the NES and would say it is OK because it is with-in the operating tolerances of the flash modules in the data sheet. I believe that the issue is is pulls down the voltage on the address lines. Voultar on RUclips has had videos about this.
I will research that topic and figure out if I can improve something. I wasn't aware of this issue. I want to fix that issue before making more modules that may damage old hardware over time. Thanks again for pointing it out.
What I'd really like to see for retro machines is an all-in-one storage adapter: SATA connectors and maybe a network port on one side, multiple GB of RAM for caching in the middle, and legacy connectivity on the other side (PATA/Floppy and/or ISA or PCI). Board runs Linux, you connect storage on the modern side, it reads disk images off that storage and presents those images as floppy or ATA disks on the legacy side. There are already SATA/PATA adapters, which allow a modern disk to be presented to an old system, but there are issues with old BIOSes choking on large disks. Using images would allow for keeping presented disk sizes to something BIOS can handle while using the capacity of the underlying modern disk to provide a multiboot environment with multiple legacy OSes without having to squeeze all of them into the size of whatever number of small disks will fit in the chassis.
You should consider upgrading the L2-Cache on the motherboard to 256kb. Not every motherboard will accept this, but on my motherboard it worked fine after a lot of testing with 32Mb Ram.
Very cool design. If you want to expand further you may be able to hotwire a secondary memory controller into a piggy-back board right on the CPU. This would also allow you to use even faster memory such as SRAM. High density (by late 80s standard) SRAM is now dirt cheap as it is used in some microcontroller use. You can easily get 4MBit modules for relatively cheap.
If I were you, I would use a FPGA or maybe a high performance MCU, to acquire all signal and translate it to DRAM/DDR. So you can get as many RAM as you want.
i remember saving up from my grocery store job to pay over $100 each for 2 sets of 2MB modules for my 386sx. ended up with right at 5MB with the onboard RAM.
There's a problem with connecting the 3V chips to a 5V bus. The DQ[3:0] pins are bidirectional, so they have both an input and output circuit. While the input is a high impedance circuit and can probably tolerate over 5V, the output has a clamping diode pointing to the 3.3V supply. Your particular chipset is only driving the bus to 3.4V, which is not enough to make the diode conduct, but there are worst-case scenarios that can cause high currents in the diode. The 3.3V supply could be as low as 3.0V, while the 5V supply could be 5.5V, which might drive 3.9V (3.4+0.5) to 5.0 V on to the bus and exceed 3.0+0.7=3.7V on the 3.3V output, causing the diode to become forward biased. The 3.3V supply doesn't turn on instantly, so it's possible that it may not be ready before the chipset starts driving signals onto the bus. One mitigation is a series diode between the outputs to reduce the current. If we assume a worst-case difference of 5.5V the resistance may be quite large, which together with the input capacitance will filter out the edges and violate the rise time. If you can guarantee that the 3.3V supply comes up before the signals are applied, then this will reduce the voltage and you can use a smaller resistor. Bidirectional level translators are available, they usually block until the power comes up on both sides and will convert the voltages to the appropriate levels. They have around 5ns propagation delay so you will have to verify the timing, as well as figure out a way to drive the direction input. An 8-channel device should cost fifty cents in quantity. The input-only pins are more tolerant because they are only connected to a high-impedance MOSFET gate, so you may only need one 8-channel translator per module (4 data pins per chip x 2 chips).
This is a great video. I have an old soyo 486 motherboard. It has 8 30pin simm slots and sis chipset. A long time ago, I saw in the manual that it could be installed up to 128MB, so I knew that. At the time, 16M modules were very expensive and uncommon, so I never had a chance to try them. Recently I was able to get four 16M modules for cheap. So I tried installing these and they didn't work properly. I look forward to seeing you work on 486 motherboards as well.
Never needed more than 4x1MB in my 386, it was enough to run doom and command and conquer poorly, but they actually ran so I was happy. C&C should not have worked at all on a 4MB 386 but somehow it did. By the time 8MB games were common I had a 486 with 8x1MB.
Make a long video installing windows 98SE on that 386! like ASMR style, with a very noisy and clicky MFM hdd with the mic right next to it and record the monitor in good quality. That would be orgasmic xD
Interesting video as usual :) Got some ideas to make 4mb modules out of old 8mb EDO modules in the same way since I got plenty of them laying around here. Not sure if its possible though. In the end of the life of my 386DX 33 it had 20mb of ram, enough to run Windows 95 a bit better then how it ran on 8mb that I had before. I had that computer until the end of 1998 when I got a chance to upgrade to a Pentium 133,
I've seen those oddball modules before. The ones I had were in an IBM RISC server I bought back in 2003. It had almost an entire gigabyte of RAM in it. I have no idea what it was used for.
*Blood from the eyes* *My teacher of special technology would have grabbed the cordial pills for a long time.* The sequence of soldering is not observed in the most cynical way. The correct sequence of assembly of the scheme: 1 soldering connectors and switches 2 resistors are soldered 3 capacitors are soldered 4 transistors and other semiconductor devices mounted directly on the PCB are soldered (after each soldering to cool) 5, lastly, IC Chips are soldered, so that there is time to cool, so as not to damage the wire connection of the crystal output inside. 5.1 IC Chips that are afraid of statics are soldered by the very last with all precautions - with heat sink, if possible, and necessarily with a grounding bracelet. (of cause see p.5 about cooling) Here I see non -compliance with technology. =)
My Atari had 1MB of memory and it was a lot! I remember reading in a magazine that althought the CPU was able to address 4GB of memory that this will never happen and extrapolated the power consumption from the 1MB and concluded that 4GB of memory would require about 200 amps during refresh cycle 😆
with that much ram i'd probably just install win95a or setup a dual boot between win3 and 95. the few 386's i've seen running 95 just had 8mb of ram, but wtih 64mb it'd probably speed it up abit.
A great video as always! With how much data we can send with our busses on the computers now, it's pretty pathetic how little our computers can do in 2024 taking into consideration we're sending multiple GB/sec even with storage now.
I remember having 4MB and later updating it to 8MB hoping to make a few dos games run better and get a bit more use out of the system. At some point this system was replaced with a used 486 with 16MB of ram running windows 95.
A successful and worthwhile project. There were some mainboards from this time that used zipp sockets for memory, do you know of any PCBWay or public domain solutions for an adapter pcb to enable reusing dram from scrap simms to solder onto?
I'm not surprised that consumer 386 (and even likely some 486) systems that used 30-pin SIMMs had a limit of 64MB when using them. That would have been an *insane* amount of RAM back then, and would have been expensive. The industry largely went to 72-pin SIMMs for higher capacity modules. It's possible that many motherboards were designed such that they could support 16MB modules (if the chipset did), but due to cost and/or availability at the time of release, they only claimed support for 4MB modules. Sill nice to know that some 386 boards out there aren't limited to only 32MB of RAM.
64MB on a 386 ... that is just crazy, I wonder if these will work on a Mac SE 30 lol PS Was a bit worried about the logic voltage , but it seems fine , so good job :)
I once had a Soyo motherboard that came with it's own proprietary software (that ran in Windows) which allowed the customization of the Energy Star logo; but as with the software you demonstrated in this video, it wasn't easy to control the colors.
Wasnt it cheaper to use a jumper shunt instead of a switch on the modules you made. I can think that make a lot of these modules and needed switches will drive up the costs a lot. And also a high failure rate when the switch goes bad. Dual voltage module itsnt a bad idea.. If one fails, the other one will take over the task. Better then over volt the ICs. ( gives you some protection)
My experience with hardware specifications back in the days, was that they only wrote what they actually tested - they were almost always never accurate.
Yeah, EDO RAM did come in 168-pin DIMMs, I suppose so you could maximise the amount of RAM in something like a Pentium system with SIMM and DIMM sockets.
It is enough memory to run in to issues with some software/games. Would be nice to have on/off switch on the ram module to turn some of the memory off... or rather to turn these 16mb modules in to 4mb modules. Then again there are sw tricks for that kind of use case. Still these sticks came out great!
Hi there! I hope you like the 30-pin SIMM 16MB memory module. Would you be interested to get a set of 2, 4, or even 8 modules? I want to understand what the demand is for such modules. Hit the like button of this comment and reply with how many modules you would be interested in. This is not an order, I just want to know if I should make a few and utilize the memory modules you have seen in this video! Thanks everyone for watching, commenting, and liking! Cheers
This is awesome. I could use 8 such modules for my Macintosh SE/30 to max out the memory :) Btw, the Sound Blaster AWE32 also supports 16MB SIMMs, but the usable memory is limited to 28MB due to the 4MB ROM sharing the address space with the RAM.
I know how to get chips for these so you don't have to desolder them from DIMMs.
@@AttilaSVK I would love to have 2 for my Sound Blaster AWE32! :D
I'd be interested in a set of four - I'm curious if my 386 boards can go as far as yours!
I'd absolutely be interested in a set of 4. Being able to switch between FPM and EDO would be invaluable for testing old hardware. Also can I please come visit your scrapyard one day? That place looks like heaven! :)
The max amount of memory I had in my 486 SX2 in the 90s was 20 MB. It had come with 8 MB across 8 slots and I had pulled 4 of those and changed them to 4 MB modules. The CPU I swapped with an Evergreen upgrade.
The project is very nice 👍 Really great, especially since it is very difficult to get really large modules. I also have a 386DX 40 MHz with 64 MB, but I bought these modules ready-made many years ago.
I was not a PC guy, more an Atari, ut at work the 386 that was assigned to me had 8 MiB and it was more than adequate. With DOS and a little bit of Windows 3 there was no point in much more memory. Only at its life end would a little it more memory be necessary when I started to do some AutoCAD. AutoCAD 9 was the maxi I could use on that 386. Acad 9 was limited in its Lisp implementation to 1 segment (64K) and as I did a lot of lisp I would have needed an upgrade to Acad 10 but the 8 MiB were too small. Upgrade of the whole computer was the better choice.
I got a Pentium 90 PC with a 2nd CPU socket (but initially only 1 CPU) with Neptune chipset and 6 SIMM72 slots for FPM modules. At the end of its life I managed to upgrade the machine to 2 Pentium MMX-233 (with adpater sockets) and 80 MiB of memory. I even had used SIMM30 to SIMM72 adpaters in it. That insane machine also had 1 IDE disk only for ooting and 4 16 bit SCSI harddisks, 1 SCSI CD-ROM, 1 SCSI burner and an external Zip-drive and SCSI HP-Scanner. The motherboard was oveclocked from 33MHz to 35.2 MHz (70Mhz oscilator to replace a 66 Mhz), not much but more was difficult for the EISA slots (yeah I forgot, it had EISA and PCI slots but no floppy/IDE on board). I should try to revive it and film for yt.
Great job! I think I can finally make myself some - I've got a box of old DIMMs that I can now put to good use :-)
I wonder how many of those SD-like modules with EDO memory chips exist. I have never seen them before. Just once, when I visited the scrapyard, I found all those modules and was like 'eh, let's take them. Maybe they will be useful in the future'... And here we go! I hope you can make good use of the PCBs!
Nice! I thought I was special with my 4MB modules :p
Have you tried to tighten the memory timings? Those timings are probably the most important feature of EDO memory :D
I haven't done much with those modules yet, but I'm planning on doubling the cache on the board and also exploring what's possible with that amount of memory.
I love how cheap and fast PCB production has made creating custom modules like this, especially for non-standard things like the EDO->FPM adapter.
It is really easy! And the learning curve to my first PCB wasn't as steep as it might seem.
I had a 8088 computer for way to long. I was young and had no money, so I saved all the money I got to upgrade my XT. The day I got enough, I bought a motherboard with an Intel 386SX, and 1Mb memory. The motherboard had 4 SIMM slots, and I wanted a single 1Mb module, but on the store they told me that I had to fill at least a bank (2 modules). My options were 4x256Kb, or 2x1Mb. No 512Kb modules. Having waited more than a year to be able to update my computer, I caved and got 4x256Kb, my options to upgrade gone. Then I had a 386sx (I think 25Mhz, don't remember), 1Mb of RAM, an Hercules graphics card, a green-phosphor monitor, a single 5,25" 360Kb floppy disk, and an MFM 20Mb hard disk.
It wasn't much, but Windows 3.1 worked well and the CGA software emulation worked SOOOO much better.
When you started up that 386 for the first time with 64MB I had a huge smile on my face as I'm sure you did. What a really great achievement and just shows you that "official" and "supported" specifications may be two different things! Thanks for sharing.
I am working on an updated version that hopefully follows the specifications on the datasheet. I do it mostly to learn more. The current modules seem to work well and I got plenty of comments supporting both perspectives. There will be a follow up on this one very soon!
8:55 That scrapyard is magnificent, I wish I had a place like that nearby with even a quarter of the stuff portrayed there.
The 64 MB RAM count was so funny i watched it several times. I really like the ram counting sound on old computers
The counting algorithm was also the same, but CPU performance grew, so on a 286, it'd slowly count up to 1MB with a tick...tick...tick, a fast 486 does its 32MB in a "brrrrrrrrrrp" in a few seconds, and some late Athlon counts 2GB in a brief high-pitched "zeet". I guess if contemporary ones still did the same, it'd just be in the ultrasound spectrum and thus inaudible to mere humans.
I've got 64GB in my main rig and I don't even see a RAM count. 😅
@@Armand79th Maybe you have the fast boot option enabled in your BIOS. It cuts down startup times by a substantial amount, by skipping most of the usual tests you usually have at the start of the computer.
Congratulations on such a successful project! Now people can finally stop bugging me to make 16MB SIMMs! Of course I'm kidding, I only got a few messages asking about them since I published that project so many years ago. But I'm glad someone finally took up the mantle and did what I never got around to doing.
I had every intention of getting back to this project and making my own 16MB modules, at one point even tracking down a source for the requisite chips, but they were very expensive since they were NOS, so I never got around to ordering them. And then my career took the front seat and I've had neither the spare time or energy for such a project since then.
But this is an even better design, since it's using chips that are basically scrap. The fact that you can use EDO chips really opens the door to potentially using other chips with a relatively simple modification to the design. So congratulations once again, you've really done some impressive work!
Thank you so much! Without you paving the way, I would have never picked up on PCB design. Your Gerber files and schematics allowed me to learn about those modules. Thank you for your work!
@@bitsundbolts Someone should make a 64GB DDR4 SODIMM now, purely for the heck of it.
your PCBWay ads are the only ads on YT that i actually watch. not even just letting the ad roll while i do other stuff, you have me paying attention and engrossed in the product. can we please make all advertising like this? relevant, informative, confident.
Oh well, thank you 😊. I usually get the opposite view about ads in general. But without PCBWay, I wouldn't be able to do all this. Not only do I get access to their services and products, but they also help a lot with their financial support. It wouldn't be feasible without sponsorships to run a channel like this. Thanks for your support! 🙏
Oh man, I have to agree, the black PCBs give it a really badass look!
It makes those modules look more premium. I'm looking forward to making these in larger quantities 😅
I was happy with 5MB back in the day. 64MB is INSANE for a 386, nice job, pushing it to double the stated maximum!
I agree - 64MB is not really useful for a 386. Most of the memory is not usable by the software that runs on such a platform. As you said, the driver behind this project was to see what is possible. I learned a lot by going through the documentation and make sense of it. And finally also to understand, why we can only use those high capacity modules in memory bank 0.
@@bitsundbolts RAMDisk would be very useful. Large smartdrv cache. But back in the day, this kind of system would have fetched a very high price. You are right, RAM was expensive!
@@bitsundbolts 64 MiB might be usable for compiling source code. You could benchmark it with 8, 16, 32 and 64 MiB to see if there is a difference.
@@bitsundbolts Back in the day I was super happy when I got my P90 system to 8mb. I put 4mb to a ram drive and had the autoexec.bat copy my current game there. 64mb would have been heaven. 😵
The OG Xbox has that much memory gawd damn
In 2000, our home PC only had 32MB RAM, 3GB HHD and 300MHZ AMD K6-2 processor (586?). The fact you were able to get twice the ram on a system at least two generations older is impressive.
We had a k6 too, classic. I still have the chip as a memento. I also remember my cousin had a 486 amd I think, the motherboard had the option of 30pin or 72pin, if I'm not mistaken, because I definetly remember those 30pin RAM sticks and I always found them slick and clean compared to the bigger ones.
Nice work! I never thought I'd see a 386 with 64MB RAM. I always thought they topped out at 16. This absolutely blew my mind!
386SX systems seem to have this limitation of 16 MBs. Glad to hear that you have seen something new in this video! 🙂
@@bitsundbolts Yeah that's one part of the 'cutdown' part of a low-price 386 alternative to what then got called the 386DX, only a 24-bit address bus was exposed on the 386SX. Internally the CPU could act like it had a full 32 bits of address space but physically impossible to talk directly to more than 16MB.
As was invariably the case from SIPP to SIMM to DIMM to SODIMM, machines may be designed to take a bigger stick capacity, but 2x the RAM would cost so much more than double the cost of the smaller sticks. They were at first only for hardcore uses, then as the line aged they became a bit more affordable, and then technology had moved on to another speed/bus/etc arrangement so they never became commodity parts because the next high and mid range systems used the commodity new thing instead.
@@bitsundboltsThat's to be expected since the 386sx shares the 24bit address bus of the 286. Only "real" 386s have the 32bit address bus
TAKE MY MONEY! 💰💰💰💰💰💰💰💰
😂
Easy Tony, easy... Beath slowly and deeply :)
I love this. saving chips that might have gone unused for the rest of time. this is something that needs to be done more often.
Haha, I am working hard on hopefully more projects like this in the future!
Years ago I ordered a number of 128MB 5v EDO DIMMs and the supplier shipped 3.3v modules instead. When I told them about the mistake they told me they'll send the correct replacement and to just keep the 3.3v sticks. Never had any use for them until now it seems. My chips are Micron parts but they appear to be 16Mx4 as well and are sharing the same pinout. Guess I should sacrifice a stick out of my pile to create a slew of 16MB 30pin SIMMs. It's a parity module too so two additional chips for an extra 30pin module... Awesome job!
Oh, wow. I always have to send stuff back 😅. Good for you - I guess you found a use case for those modules now! I wonder what you possibly could have used those EDO modules for otherwise (I guess you have the 168-pin version). I would rather use SD-Ram since it is probably a lot faster.
When the 386 counted up to 64MB the first time, I had to laugh like you did. Insane idea and great implementation!
Another question: I am an earlytime member and from all the videos I have watched I might think you are German. There are some hidden words and things you did that make me think that. Am I wrong?
Yes, I am German - you are right. Listening to the memory test counting up is just great and I am glad those PCBs worked! Although, after making the 4MB SIMMs, it wasn't as challenging anymore.
@@bitsundbolts Dann hab ich doch richtig gehört. Ich glaube es war in einem der P2B Videos wo Du ganz kurz auf deutsch geflucht hast.
I am currently working on a thinkpad 755cs that only has 8mb of RAM and boy does it feel cramped!
Not much room to run applications. I'm sure it would benefit from a memory upgrade!
Just awesome! I would love to see if any 386 boards support 128mb on board using 8 of these.
Thank you Adrian. I hope to find a board that supports 128MBs!
8:54 "Scrapyard to the rescue."
Holy smokes !... Your local scrapyard is so much better than mine.
How much to buy the whole lot ?
Fantastic work as always.
Thank you!
Back in 1990s I had a borrowed Am386DX-40 machine with 4MB of RAM (and without FPU). It run Win3.11 For Workgroups on MS-DOS 6.22. There was Corel 5 and Corel 7 installed. Corel 5 was working fine, Corel 7 had trouble fitting in RAM and was swapping a lot. It was fun seeing Corel Draw 5 actually drawing the business cards when a file was opened.
Oh, I remember Corel. I think I had version 4. That was for Windows 3.1 if I'm not mistaken. It was slow if I remember correctly. Those vector graphics just took some time to be drawn on screen.
0:36 I found a 16-SIMM board in the scrapyard and a friend gave me another one. Very happy to own these.
They are no longer that common. I have many and not really any use for them. I prefer larger capacities. However, if you want to build a period correct system, those 1MB modules are probably the sweet spot.
@@bitsundbolts I had never even seen one and wasn't aware they existed, I was going to make it one of my retro machines and max it out with what I could find.
Once I fix the board though.
@@bitsundbolts It's weird to see "period correct." I'd say who cares. But the there's automotive, where original, oem, has the most value. There it makes sense since most can't build better than the manufacturer, and no one wants to buy junk used. I think we had 1mb in our 386 back in 92
Please note that there have been reports on similar chips dying in 5V systems. On the other hand, I have a few SIMM72 5V modules made with 3.3V-only chips but indended for 5V boards. They don't even have the regulators, just two diodes in series to drop the voltage to 3.5-ish.
Also there's another way of making 16M SIMM30 sticks. It involves 8 4x4 chips and a PAL chip for reorganizing the memory. Works kinda like SimmConn. I do have the PCBs drawn and firmware written for PAL but just don't have the time to build it 😢
Great video as usual!
Hey! Good to hear from you again! Thanks for pointing out the issue with the 5v issue. I'm going to research the topic and try to address the issue somehow.
I have seen other SIMMs that also just have the voltage regulator and a few capacitors. I'm confused to be honest. But I'm researching level shifters at the moment.
Deja tu mujer y tendrás tiempo de sobra. Estos proyectos sí venden si los nostálgicos sabemos que existen.
Great share, it led to a neat video here.
Wait, 20 years ago I had 96 MB FPM in a 486 system. I was moving about a lot and at that time, got myself a P4 or whatever and stored that 486 back at my parents'... so my mom promptly recycled it. WOMEN! I still can't forgive her, for this and also my pimped out 386 router with a Hercules console :(
Anyways, 32 MB FPM sticks not only existed, not only I had them, but they were scrap when I got them . The world already moved from EDO SIMMs to DIMMs, I was simply too poor for any of that and had to actually research what was actually fast, and my resulting 486 pwned!
Noooooo! Ah, I am so sorry. This must have been something like "ah, why does he keep this old stuff around. Let's make some space."
So sad seeing all the 90s scrap, while it's all very usable for projects like this.
It is sad. There are so many interesting things. But also, only a handful of people who are interested in old computers 😞
Another awesome video and another win for making this hobby a little more affordable! Thank you!!!!
Absolutely! It looks like I will have to make quite a few of those modules soon!
Fantastic work! One potential problem, having to fit 3.3v regulators for 3.3v RAM, the RAM may have a shorter life as any clamping diodes are driven frequently by 5v logic levels of the chipset. It's also possible the chipset could die when interfacing 3.3v and 5v. It's possible the RAM you are using is 5v tolerant? I see you scoped the address and data levels there, and they look right - but probably because of the clamping. It's worth measuring that same way with 5v RAM fitted - see what levels you get. Don't be put off by this btw - the chances are it will work reliably even if the clamping is an issue. It's just one of these things that I am sure will get pointed out.
The ram I'm using is not 5V tolerant. In the data sheet it says absolute maximum voltage is 4.75 or something. I am wondering and will do some additional tests before starting a mass order of those PCBs. In case there are 5v signals, would a simple voltage driver be enough?
@@bitsundbolts What you might be able to do - since it says the RAM can go up to 4.75v, try measuring the data and address bus levels with a 5v SIMM, lets say the levels are around 4v - that would mean you could perhaps try and regulate the RAM VCC to 4v ish, and in theory the RAM would work safely as its under its 4.75V limit, and clamping would be (I think) less likely as the logic levels are within the VCC level.
I wondered same issue, chip lifespan will be reduced with 5v operation. I am not sure how tight timings on address bus but maybe level shifters are fast enough to use on this modules?
Would a simple voltage divider do the trick? I might have to get like 40 small resistors on the board if I understand that correctly.
I'll try to understand this more. I already got details regarding level shifters. More things for me to learn and definitely a thing to look at before I order larger quantities of those modules. Thank you for all the great hints!
64MB of RAM on a 386 is quite amazing, nice job. One remark/question: the memory chips are designed at 3.3V, for which you added a Vreg on the power rail. But the address- and data-lines are still driven by the 5V chipset. No risk in damaging the chips on the longer term?
I tested the data lines with an oscilloscope and they don't seem to go above 3.4 volts. However, I wasn't able to check the lines that are facing the board (due to limited space). The chips have a rating of up to 4.75 volts as "absolute maximum" - so, there is some margin. Maybe someone with more knowledge than I have will be able to confirm the long term implications if such chips would be exposed to higher voltages (e.g. on a different motherboard/device).
That's exactly what I thought. Maybe that particular board uses 3.4V, but some another one could easily send 5V, as it expects 5V module.
Early PowerMacs and IIRC some workstations used 168-pin 5V FPM DIMMs, so the chips on those would not have any long-term issues on a pre-socket 7 motherboard.
@@bitsundbolts 3.4 volts could be clamping diodes in the RAM. Might be worth a check without RAM installed.
SoundBlaster AWE32 (and AWE64 IIRC) supported 2x 16M modules, but could only use 24MB of RAM for soundfonts. Back in 1990s this was The Ultimate MIDI Setup. SB-Live! / Audigy soundfonts will work with it too.
(yes, yes, didn't forget, there was GUS, and Holy Wars wars between some GUS and AWE users, I know ;)
The SoundBlaster AWE32 was released in 1994. But later soundcards where much better MIDI setups. I had a Terratec EWS64XL in around 1997. This was still in the 90s and the EWS64XL was much better than the earlier SB AWE32 and still an ISA card for best compatibility to DOS games.
@@OpenGL4everIn fairness, with full ram you can get very nice midi quality out of the AWE32, so long as you chosen DOS game will run under Win9x. Depending on the model they can also have real OPL3 FM, which is a nice perk as well.
My old Packard Bell 486 started out with 4mb, but by the end of its life had 32mb. Since it was all low-capacity SIMMs, I had to use several of them per slot via expansion boards. This was def more than enough for Windows 95, but then I made the mistake of installing Win98 on it…
Ohhhh, I remember when I installed windows 95 on my 486. It was such a great experience. I loved it, my PC felt like new. I only used Windows 98 with a Pentium II 350. I never attempted to install it on my AMD 486-DX4-100.
@@bitsundbolts By the very end I did upgrade it from DX2-66 to DX4-100, but even this didn't make and difference for W98 performance. Should've just downgraded it back to W95.
Back then couldn't figure out how to upgrade the video card - would've been fun to see how far I could push it: )
Next stage: custom diy socket 7 motherboard)
Haha, yeah, that would take more than one day :)
Related to these 30-pin SIMM memory modules, I am very curious on this Socket 7 Motherboard model by GCT: GCT-8IV.
According to The RetroWeb, this motherboard has the ability to utilize 3-4 different types of RAM (the manual debates this) and can run a k6-III and Pentium MMX 200-233 mhz processor. Can you do a video involving this Socket 7 motherboard? On paper, it sounds like a rough diamond but is it? 🤷♂🤷♂
I had a 486 DX2-66 machine with a weird motherboard that had 3 72 pin SIMM slots. I ended up populating it with 3x16MB modules for a total of 48MB of capacity. It was a fun machine to put slackware linux on.
Considering the price collusion of the RAM manufacturers, what would that much memory have cost at market from 92~95?
I thought it was common knowledge that EDO memory supported FPM. I remember back in the day using EDO SIMMs in my Acorn RiscPC which only supported FPM and EDO was much cheaper at the time!
If RAS MA11 is only required for 16 MByte modules, would that mean that 8x8 is possible?
This makes me wonder what kind of memory was in larger x86-based systems like 80386-based Sequent Symmetry and PPro ASCI Red.
64MB on a 386 enables some interesting benchmark comparisons vs contemporaneous systems with MIPS or SPARC CPUs running at the same clock rate.
My 386 did not have enough RAM to run Doom, but Dune II, Pinball Fantasies, Duke Nukem 2, and even Jazz Jackrabbit worked well iirc.
I once had a Socket 3 system I maxed out with 128 MB of memory. It had a Shuttle HOT-433 motherboard. I also had an AMD 5x86 overclocked to 160 MHz in it. Why did I ever let that computer go? 😢
Amazing!! :D
That Cyrix FasMath FPU is gorgeous. I also find those COB SIMMs you showed at the start of the video pretty interesting!
It's been a long time since I last played with my 386s, but I remember 4MB being the most I could use... So I think i don't have any module larger than 1MB (and there's also the possibility that I achieved said 4 megabytes on a board with 8 sockets and 512KB per module being my maximum!).
Most of my SIMMs are 1MB modules except for the ones I made myself. I think there must also be a few 256kb ones.
Considering my first PC (was my mum's back in the late 80's/early 90's) was part of the transition period of 5.25" FDD to 3.5" (and everything was installed to 3.5" floppies instead of the 80MB HDD) - I'd have been lucky to have 2MB of RAM. Not sure about the exact amount as a rather technologically ignorant family member killed the PC by accident when moving in the early 2000's, so it has been gone for a very long time (excluding the still functioning HDD + dual gameport card).
After watching this video I realize that I have 2x EDO memory sticks (probably each 16Mb), that I storage a few decades ago. I can give it for free if someone needs it.
In the chips are written CW424410-7 9528 and 9519.
Have you ever done any system design for S-100 bus computers?
I would love to add every modern feature that didn't exist in 1977..
SIMMs, M.2, i2c, USB, WiFi, BT, etc
Make it run in 2 different modes...
and make it backward compatible with original pinout
make it run to take full advantage of all the bus pins..
example the data bus had 8 input and 8 output data lines.
Would it be possible to make it 16 bit bi-directional?
I had a Compaq ProLinea 4/25S. Had 4 72-pin slots. 56MB total…
Config;
First slot could only be 4MB as there was 4MB soldered to the board, the other three slots had my 16MB sticks.
Also had a Kingston Turbo Chip 5x86 2 133MHz installed.
Awesome as usual! With this amount of ram you can install some Linux distro in future videos.
Although I am not a Linux user, it is interesting that this amount of RAM would allow to run software that would be difficult to use without. I may give it a try. I remember trying out some SuSe distributions long time ago, but I don't remember what version (maybe 7?).
@@bitsundbolts I just remember that Linux kernel drop 386 support in 3.8 version so last working kernel is 3.7.10 in theory all distro before 2013 should work.
Also I personally want that you run Windows 98 SE and Windows 2000. I know that Windows 2000 work with 486 and 64mb ram but don't know about 386.
@@bitsundbolts I used Linux Slackware 7.0 on my 486DX with 8 MiB. Starting with Slackware 8.0 with kernel 2.4, I had to bake an extra kernel so that I could boot to the command line with the 8 MiB. I couldn't install Debian or Suse, the installer required to much RAM.
Seriously I am so sad that I gave away my 486 when switching to my CyrixP5 in 1996. But at work I used a DX4 100 with 48 MB EDO RAM up until 2001 (running NT 4.0)
Neat. I managed to find a 486DX2 66MHz Compaq a while ago and piece together 100MB of RAM for it, maxing it out. I haven't booted it up in quite some time, and I don't know if I have a monitor with a VGA port, or enough converters I can plug end to end. 😅 I inspected it not that long ago and all the caps look ok, and it's been kept dry and clean so it should work. 🤷♂️😂
Epic and inspiring. This could actually be a business for old Samplers where RAM got super expensive. An E-MU E IV is much more fun with 128 MB.
I had a 386sx25 back in early 90's. I had an ISA card that had 16 slots for 30 pin memory modules and I salvaged a pile of 256kb modules. It gave +4 MB of total memory in addition of those 2MB already on mainboard.
Interesting video though.
I'd recommend ordering PCBs with Gold ENIG. It's worth the extra few bucks.
I think the max I had in 386 or 486 systems was 4mb. I don't remember if the "5x86" and similar chips from Cyrix and the like that I had counted as 486s or not, but those I had total of 16. I don't remember if it was late Socket 7 or early Super 7, but I had a board that could take either 2 sdram or 4 72p simms, don't remember the processor, but I had 4x16 for 64mb total for that, the highest I've ever had in simms.
I wonder, would it be possible to build an adapter that converts 72-pin EDO to 30-pin FPM without desoldering? I'm imagining two possibilities, one where you utilize only 1/4 capacity, another where it splits into 4.
Question: Since the the 8MB chips require 3.3v shouldn't you need to use level shifters on all the address lines?
So, here comes the question that bugs me! I honestly never heard of level shifters. But a quick search on Google brought some light into that mystery. My board seems to access those chips using safe voltages, but other devices may not! So, I am interested to know more about it. So far, the memory chips seem to work great, but I'd like to improve the modules if it can be done. What should I look for? Level shifters on any data and signal line like address, CAS, RAS, WE, OE, etc?
@@bitsundbolts I am not an EEE but one thing that I have learned is when connecting 3.3v devices to a 5v system you should use level shifters in-between to prevent damage over time to both sides. For example old 5v video game systems the NES, some people were building flash carts using 3.3v flash modules directly connected to the 5v address lines of the NES and would say it is OK because it is with-in the operating tolerances of the flash modules in the data sheet. I believe that the issue is is pulls down the voltage on the address lines. Voultar on RUclips has had videos about this.
I will research that topic and figure out if I can improve something. I wasn't aware of this issue. I want to fix that issue before making more modules that may damage old hardware over time. Thanks again for pointing it out.
Blimey, I thought I was King Shit for having 24MB in my 486SX. Brilliant video. 👍
What I'd really like to see for retro machines is an all-in-one storage adapter: SATA connectors and maybe a network port on one side, multiple GB of RAM for caching in the middle, and legacy connectivity on the other side (PATA/Floppy and/or ISA or PCI). Board runs Linux, you connect storage on the modern side, it reads disk images off that storage and presents those images as floppy or ATA disks on the legacy side. There are already SATA/PATA adapters, which allow a modern disk to be presented to an old system, but there are issues with old BIOSes choking on large disks. Using images would allow for keeping presented disk sizes to something BIOS can handle while using the capacity of the underlying modern disk to provide a multiboot environment with multiple legacy OSes without having to squeeze all of them into the size of whatever number of small disks will fit in the chassis.
You should consider upgrading the L2-Cache on the motherboard to 256kb. Not every motherboard will accept this, but on my motherboard it worked fine after a lot of testing with 32Mb Ram.
Next time you order black PCBs, take the ENIG finish instead of HASL, trust me you won't regret it 😁
Very cool design. If you want to expand further you may be able to hotwire a secondary memory controller into a piggy-back board right on the CPU. This would also allow you to use even faster memory such as SRAM. High density (by late 80s standard) SRAM is now dirt cheap as it is used in some microcontroller use. You can easily get 4MBit modules for relatively cheap.
If I were you, I would use a FPGA or maybe a high performance MCU, to acquire all signal and translate it to DRAM/DDR. So you can get as many RAM as you want.
i remember saving up from my grocery store job to pay over $100 each for 2 sets of 2MB modules for my 386sx. ended up with right at 5MB with the onboard RAM.
There's a problem with connecting the 3V chips to a 5V bus. The DQ[3:0] pins are bidirectional, so they have both an input and output circuit. While the input is a high impedance circuit and can probably tolerate over 5V, the output has a clamping diode pointing to the 3.3V supply. Your particular chipset is only driving the bus to 3.4V, which is not enough to make the diode conduct, but there are worst-case scenarios that can cause high currents in the diode.
The 3.3V supply could be as low as 3.0V, while the 5V supply could be 5.5V, which might drive 3.9V (3.4+0.5) to 5.0 V on to the bus and exceed 3.0+0.7=3.7V on the 3.3V output, causing the diode to become forward biased.
The 3.3V supply doesn't turn on instantly, so it's possible that it may not be ready before the chipset starts driving signals onto the bus.
One mitigation is a series diode between the outputs to reduce the current. If we assume a worst-case difference of 5.5V the resistance may be quite large, which together with the input capacitance will filter out the edges and violate the rise time. If you can guarantee that the 3.3V supply comes up before the signals are applied, then this will reduce the voltage and you can use a smaller resistor.
Bidirectional level translators are available, they usually block until the power comes up on both sides and will convert the voltages to the appropriate levels. They have around 5ns propagation delay so you will have to verify the timing, as well as figure out a way to drive the direction input. An 8-channel device should cost fifty cents in quantity.
The input-only pins are more tolerant because they are only connected to a high-impedance MOSFET gate, so you may only need one 8-channel translator per module (4 data pins per chip x 2 chips).
This is a great video. I have an old soyo 486 motherboard. It has 8 30pin simm slots and sis chipset. A long time ago, I saw in the manual that it could be installed up to 128MB, so I knew that. At the time, 16M modules were very expensive and uncommon, so I never had a chance to try them. Recently I was able to get four 16M modules for cheap. So I tried installing these and they didn't work properly. I look forward to seeing you work on 486 motherboards as well.
Never needed more than 4x1MB in my 386, it was enough to run doom and command and conquer poorly, but they actually ran so I was happy. C&C should not have worked at all on a 4MB 386 but somehow it did. By the time 8MB games were common I had a 486 with 8x1MB.
Why did DIMM quit requiring Termination modules for empty slots?
Thank you for this!
In theory, my Macintosh SE/30 could easily get its 128MB RAM maxed out!
Back in the day (1990) I had an Amiga 500 with 2.5 MB RAM (which was an incredibly large amount of RAM :) .
Make a long video installing windows 98SE on that 386! like ASMR style, with a very noisy and clicky MFM hdd with the mic right next to it and record the monitor in good quality. That would be orgasmic xD
Interesting video as usual :) Got some ideas to make 4mb modules out of old 8mb EDO modules in the same way since I got plenty of them laying around here. Not sure if its possible though.
In the end of the life of my 386DX 33 it had 20mb of ram, enough to run Windows 95 a bit better then how it ran on 8mb that I had before. I had that computer until the end of 1998 when I got a chance to upgrade to a Pentium 133,
I've seen those oddball modules before. The ones I had were in an IBM RISC server I bought back in 2003. It had almost an entire gigabyte of RAM in it. I have no idea what it was used for.
*Blood from the eyes*
*My teacher of special technology would have grabbed the cordial pills for a long time.*
The sequence of soldering is not observed in the most cynical way.
The correct sequence of assembly of the scheme:
1 soldering connectors and switches
2 resistors are soldered
3 capacitors are soldered
4 transistors and other semiconductor devices mounted directly on the PCB are soldered (after each soldering to cool)
5, lastly, IC Chips are soldered, so that there is time to cool, so as not to damage the wire connection of the crystal output inside.
5.1 IC Chips that are afraid of statics are soldered by the very last with all precautions - with heat sink, if possible, and necessarily with a grounding bracelet. (of cause see p.5 about cooling)
Here I see non -compliance with technology.
=)
My Atari had 1MB of memory and it was a lot! I remember reading in a magazine that althought the CPU was able to address 4GB of memory that this will never happen and extrapolated the power consumption from the 1MB and concluded that 4GB of memory would require about 200 amps during refresh cycle 😆
Haha, nice! And now we're considering 4GB unusable.
with that much ram i'd probably just install win95a or setup a dual boot between win3 and 95. the few 386's i've seen running 95 just had 8mb of ram, but wtih 64mb it'd probably speed it up abit.
My God, that scrapyard...
Did you get the super-SIMs from some sort of Unix server or workstation?
A great video as always! With how much data we can send with our busses on the computers now, it's pretty pathetic how little our computers can do in 2024 taking into consideration we're sending multiple GB/sec even with storage now.
I have a bunch of 256mb dubble sided memory from a sun computer, wonder If they are useful for anything
I remember having 4MB and later updating it to 8MB hoping to make a few dos games run better and get a bit more use out of the system. At some point this system was replaced with a used 486 with 16MB of ram running windows 95.
A successful and worthwhile project. There were some mainboards from this time that used zipp sockets for memory, do you know of any PCBWay or public domain solutions for an adapter pcb to enable reusing dram from scrap simms to solder onto?
From what I can remember I never actually had a 386 back in the day - I went from a 286 Tandy to a 486DX2-66 with 16MB.
Brilliant work. Would it be possible to use these modules on Sound Blaster cards such as the CT3600?
Incidentally the chips on my Sony 2x64 are the same, so might be useful.
I love maxing out 386 board with more ram than possible in the day, SSD, other tech to see how competitive it can be even with the CPU bottleneck!
Not sure if this helps, but when I worked as a tech. We got in an old IBM server which used those double stacked up SIMs. I think they are 32MB each?
And there we are the IBM PS/2 users with our non standard 30 pin FPM with parity modules 😅
I used to solder leg onto mine to use them as SIPP modules :) Much fun ! !... people were nicer during that time...
I'm not surprised that consumer 386 (and even likely some 486) systems that used 30-pin SIMMs had a limit of 64MB when using them. That would have been an *insane* amount of RAM back then, and would have been expensive. The industry largely went to 72-pin SIMMs for higher capacity modules.
It's possible that many motherboards were designed such that they could support 16MB modules (if the chipset did), but due to cost and/or availability at the time of release, they only claimed support for 4MB modules.
Sill nice to know that some 386 boards out there aren't limited to only 32MB of RAM.
SB Awe 32 maxed out at 28mb wonder if it could see your ram
I wonder that too. I have a SB with sockets. I should try that one day!
@@bitsundbolts You should definitely do that! There are some great soundfont files and it will be another interesting use for your amazing work!
64MB on a 386 ... that is just crazy, I wonder if these will work on a Mac SE 30 lol
PS Was a bit worried about the logic voltage , but it seems fine , so good job :)
There is only ONE way to find out if they work on a Mac SE 30 :)
likely yes but they will be too tall to get the motherboard back in the tray
I once had a Soyo motherboard that came with it's own proprietary software (that ran in Windows) which allowed the customization of the Energy Star logo; but as with the software you demonstrated in this video, it wasn't easy to control the colors.
My favorite PCB look is black substrate with clear solder mask, so you can see all the copper traces-
Ohhhh, interesting... PCBWay has that option... Maybe I should try those.
Wasnt it cheaper to use a jumper shunt instead of a switch on the modules you made.
I can think that make a lot of these modules and needed switches will drive up the costs a lot. And also a high failure rate when the switch goes bad.
Dual voltage module itsnt a bad idea.. If one fails, the other one will take over the task. Better then over volt the ICs. ( gives you some protection)
My experience with hardware specifications back in the days, was that they only wrote what they actually tested - they were almost always never accurate.
How does the memory controller decide whether pin 91 wil act as RAS1 or MA11? How is this pin connected to the ram slots?
Yeah, EDO RAM did come in 168-pin DIMMs, I suppose so you could maximise the amount of RAM in something like a Pentium system with SIMM and DIMM sockets.
Technically you can run BSD or Unix on a 386 and it should be able to utilize the 64 mb of ram.
It is enough memory to run in to issues with some software/games. Would be nice to have on/off switch on the ram module to turn some of the memory off... or rather to turn these 16mb modules in to 4mb modules. Then again there are sw tricks for that kind of use case. Still these sticks came out great!