How to use EDO as FPM memory - and make it work on a 386 motherboard
Vložit
- čas přidán 2. 06. 2024
- My AMD 386 DX-40 is sad. It only has access to 8MB of memory. The motherboard this CPU is soldered to supports up to 32MB utilizing all 8 memory slots. Join me in today's episode where I will create 4 SIMM modules with 4MB each. At the end of this project, the AMD CPU will have access to 16MB, but there is a catch:
I only have EDO memory chips...
Sponsored by PCBWay: pcbway.com/g/lF253I
PCBWay provides full feature custom PCB prototyping services, 3D printing, and much more. Check them out using my referral link above and get a 5 USD welcome bonus when you sign up as a new customer!
Order the 3dfx Voodoo Memory Mod (Revision 1.1) from PCBWay directly:
www.pcbway.com/project/sharep...
Repository with Gerber files and other useful items for Voodoo Memory Mod (Revision 1.1):
github.com/BitsUndBolts/3Dfx-...
UpLateGeek Memory Module Design/Gerber:
github.com/UpLateGeek/Let-s-M...
Soldering iron (FNIRSI HS-01):
s.click.aliexpress.com/e/_oBq...
You can support me on Patreon:
/ bitsundbolts
00:00 Intro
01:15 Requirements
03:21 PCBWay
03:56 Assembly
05:17 Why EDO may work
06:48 First test
07:00 Debugging
07:59 Drops of knowledge
11:00 Mod EDO to FPM
12:25 Retest
13:47 Conclusion and future projects - Zábava
You are very welcome! 🙂As far as I can tell, this /OE trick was first mentioned in February of 2005 (no link because CZcams may block this comment). Despite being mentioned again here and there, it never gained the attention it deserves, at least until now. It also works for bigger 16-bit ram chips, used on late SIMM-72 sticks.
You can get fascinating results by manipulating /WE and /OE pins (by means of programmable logic aka PAL/GAL), and that is exactly what SimmConn project has done to let us use SIMM-72 on AWE64 cards. One of my ongoing projects is making 16MB SIMM-30 FPM sticks with 8 4x4 EDO chips, for use primarily with AWE32 sound cards. If I ever finish it, I'll let the world know 😅
I am really surprised that this mod is not so well known. I tried to find information on how to enable FPM timings on EDO chips, but nothing usable showed up. So, thanks again for this information! I wish you all the best and good luck for your projects!
In order for it to get attention, it should be placed in the Wikipedia EDO RAM article. However, it is sometimes difficult to do this because on Wikipedia moderators and long-term users often interfere and do not allow you to incorporate such information into the article.
It has to look encyclopedia-worthy for them and these guys just don't have a clue to judge that. Therefore, any book should still be cited as a source in order for it to be accepted. (Tip: nobody reads the book and nobody checks whether it is really in the book)
Unfortunately, these guys are not about sharing knowledge and making life easier for humanity.
@@bitsundbolts You should change your video title from "386 with EDO memory" to "How to use EDO RAM as FPM memory". This way, also 486 users will find that tip.
I will do this for the next video. To be honest, the hack was a plot twist! I already had part of the video done (until the non-working modules). Then Arti9m came to the rescue! When I get the modified PCBs (hopefully they work), I will make another dedicated video for the mod.
До смешного, я узнал об этом хаке только месяц-полтора назад на стриме Максима Крюкова @Fagear. Почти в конце стрима зашла речь о том, что поздние 486 материнские платы, построенные на чипах SiS 85C496/85C497 в теории могут работать с EDO памятью, НО не все такие платы работают с EDO из-за недоделанного BIOS по этой части (предположительно). И сам Максим об этом тоже не знал до этого момента. А вообще получается, что сам по себе чипсет уже сделан так, что может синхронно переключать логические уровни с /OE и /CAS для совместимости с EDO, но если BIOS об этом не знает, то и задействовать такой режим чипсета не получится. К примеру, его плата Soyo так делать не умеет. У меня же есть плата PC Mate 4IP-SIS-AIO, построенная на том же наборе чипов, но пока у меня нет процессора, чтобы проверить, как она работает с EDO.
А вообще, я неделю купил около 25 модулей 72-pin SIMM разных. Они по большей части EDO, но есть и FPM, и старые, и более новые. Они были изъяты из игровых автоматов (внезапно), в тот же день я научился отличать их объём "на глаз", прочитав даташиты на микросхемы и статью на ixbt, а ещё понял, что существуют 80-pin SIMM, которые вовсе не являются оперативной памятью, а флеш-памятью, которые, оказывается, могли применяться в коммутаторах Cisco для хранения их прошивки. Один такой я тоже купил. Не знаю, зачем, но пусть будет, он на чипах AMD. Тем не менее, при всём желании он не будет работать в старой материнской плате. Они тоже стояли в игровых автоматах, и на них была записана, видимо, их прошивка с игрой, причём, прошивка разделена на два таких модуля. А в качестве оперативной памяти стоял всего один модуль 72-pin SIMM, причём абсолютно случайные модули, из купленных сложно выбрать парные, почти все они на разных чипах, местами очень похожих, но парных нет.
This is great! I have tons of worthless EDO and a bunch of 486 that require FPM. Thanks a MILLION!!
Happy it excites you! Keep me updated!
I've got the same problem - lots more fast EDO than FPM. Just been checking out a 4MB 72-pin SIMM with 42-pin SoJ chips and I don't think the !OE > !CAS trick can work on those, since there are two CAS lines per chip and only one !OE.
It does work with the smaller 20-pin SoJ chips shown in this video - I've just successfully converted a 4MB EDO 72-pin SIMM with MT4C4007JDJ chips to work with FPM by pulling up the !OE pin on each chip and soldering it to !CAS
@Voodle Haha, cool!! Congratulations!
@@voodler On a 42-pin chip you could try lifting /OE and connecting it to the nearest /CAS, which should be /CASH or "/CAS High". People have reported it worked on those chips as well.
Was about to comment something similar. Got lots of random useless single 72pin EDO sticks, and almost no 4MB 30pins, with new/repro ones from the US costing a fortune to import to euroland. If I can get this to work, this might finally solve a ton of RAM shortage issues I've been having
Sounds like something that can be added to the PCB with three pads. Bridge Pad 1-2 for EDO, 2-3 for FPM.
A tiny dip switch would be excellent for this too so you can change modes on the fly without an iron
My next question would be, does the FPM module require the !OE pin to be grounded, or would it operate with the mod too?
It shouldn't even need that, BuB's discovery here suggests that it should just be a new revision. The !OE signal isn't really used on FPM/EDO memory since tying it to ground makes it always on and the 30 pin SIMM pinout doesn't include an !OE signal. Having an FPM module's output enable signal tied to CAS should I think still allow it to operate normally still, while allowing EDO chips to operate as FPM.
Generally speaking, there are less applications for 30-pin EDO memory than FPM, so a new revision that simply ties !OE to !CAS should cover both FPM and EDO memory chips to make it easier to source chips for newly made 30-pin FPM SIMMs.
The main application for a switchable FPM/EDO mode would be on 72-pin SIMMs. Either a solder bridge or jumper pins would be a good fit - a set of small solder bridges would probably be best since a 72-pin SIMM has 4 CAS lines while a 30-pin SIMM only has one. Installing 4 jumpers per SIMM would be a hassle.
@@voodler In theory that should work, but in practice there will be outliers. This video is a good example. In theory the EDO chip should have just worked as FPM but it didn't without changing !OE. There's also very little downside to having a selectable pad option other than maybe an extra minute of assembly time. You already have the soldering iron and everything out to solder on the chips. It's not that much extra work to bridge a pad have a solution that will cover 100% of all outliers and use-cases.
Someone should inform the guy of that github gerber project. I can't, i don't have a github account.
THIS INFORMATION FROM rt9m IS GOLD, because it allows to better recycle the more abundant and high capability EDO memory chips. Kudos to him and the github guy who made the blueprints for simm memory modules.
The github guy could change the blueprints for the memory modules and add a way to bridge the right connection EDO or FPM through soldering or a pin header.
Fortunately, the PCB design is open source. Anybody can change it. I used the last two days to do exactly that. A modified prototype is already underway. You can see a post with the modified PCB in the community tab on my channel.
@@bitsundbolts That sounds great. I hope they work and there are no timing issues.
That's an awesome hack! I never heard of it, but I'm glad you tried it out and it worked. I'll have to give it a try myself, and possibly think about updating my project. But that might have to wait until I get some time off work, most likely around Christmas. Although considering my queue is literally over 20 projects deep (and counting!), that might be more like Christmas 2024!
Well, that would be awesome! Thanks for all your efforts! If you don't mind, I may try to give it a shot. I may also try to maximize the free area per PCB sheet. And I am happy that it's not just me who is amazed by this little hack - it actually feels like it's the majority!
Thanks again UpLateGeek for your work on the PCB design and layout. Without your work, this video wouldn't have been possible!
@@bitsundbolts Yeah, no worries mate! I think it would be pretty cool if you can modify my project with this hack. As long as you publish it as open source (if you're going to release the files) and mention where you got the files for the mod.
My project is published under creative commons - attribution, non-commercial, share alike license, so you don't even need my permission as long as you publish your project under the same agreement. (Unlike _some_ people who were clearly "inspired" by my design, but are selling their own modified version for profit!)
I will probably take some time to work on it. Once I am done and tested, you may just pull the changes into your repo if you approve the changes.
Yes, actually, you only need to modify the wiring of the boards a little in order to add to each chip a platform for soldering an SMD jumper, with which you can either connect the signal /OE of the chip to the ground, or connect to the signal /CAS. These will be universal boards for EDO/FPM.
Nice, you should update the git pcbs to include ic adaptors and then you can add a switch on each module to swap between the modes. Maybe increase the hight of Simms to account for raised profile. what a brilliant hack
This is pretty awesome! It is also interesting to learn more about the difference between EDO and FPM RAM as it relates to how the memory is accesses by the system.
As always a great video :)
As for easy mods:
- the classic pin hack for socket 370 and slotkets that allow a tualatin CPU to run on coppermine boards.
- AMD athlons/durons from slot A up to (at least) socket AM2 have a whole host of pin mods and conductive paint mods that allow changing/unlocking multipliers, cache, FSB speeds, vCore and more. IIRC early socket A (Thunderbird core) Durons and Athlons could have their multipliers unlocked using only a pencil.
- On socket 775, the BSEL 0 and BSEL 1 pads can be modded to alter the default FSB of the chip.
- On my old Asus socket A board (A7N8X) back in the day, there was a fairly easy VCore mod that only needed a pot and a single solder point. Had some nice OC adventures using that mod, including my first foray into dry ice cooling...
Yes, I briefly looked at AMD CPUs, especially the Thunderbirds. There are those L-bridges that can be modified to unlock a lot of cool stuff. Some day I will look into those. Socket 775 sounds interesting too! Thanks for sharing!
@@bitsundbolts Instead of overclocking a CPU, you can simply buy a faster one from that era. This is especially beneficial when that faster CPU is a generation or two ahead and fits on the same socket. After all, it's not just the clock that often changes between generations, but also the available cache and instruction set. The latter becomes particularly important when the first CPUs with SSE appeared. The NX bit and the PAE mode are important for modern operating systems.
ya it should be shared, to as many as possible, one thing i've been trying to share with people is a sdram mod i've done on my casio fiva to get it to work with sdram sticks larger then 64mb, if you jump pins 62 and 68 it can even work with some sticks of 512mb. honestly an amazing upgrade for that win98 pda, but i think it might work on other devices.
Great! Thanks for sharing! I will have a look.
I remember EDO RAM.... Got to love teh days of SIMMs
Back in the old days I was successful using EDO Modules in incompatible SYSTEM when pairing them with FPM Modules.
Worked like a treat :D
Easy mods? The taulatin to Coppermine pinhack comes to mind , you only need to mod two pins on a taulatin CPU to make it work on a Coppermine compatible motherboard , tested it on several via boards and it's stable and works great.
Are you going to modify the SIMM layout to have this hack built into the layout?
I will give it a shot.
I had intended on releasing an updated version (fully open-sauce!) a while back, but I ran into major issues testing. I don't trust my ADHD, overworked and sleep deprived brain not to have made a stupid mistake on the design, so I was waiting until I could test my memory chips separately to confirm it's not my fault before releasing the new design. Now it looks like I'll have to go back to the drawing board anys ways!
Thanks! Looking forward to trying this!
I'm just fascinated by what can be done. It's wonderful when comuitity can help solve a problem. Thank you for the many exciting videos, I'm looking forward to the next ones.
Thanks for the interesting content. Very neat job.
Cooles Projekt und egal wie es ausgeht, man hat immer etwas gelernt. Nice Job.
Thanks!
Great video, Thanks for sharing
🎉wow! thats a substantial upgrade!
very informative as usual! Thank you!
How do you have time watching my videos? Aren't you playing with your new Playstation 4? 😉
@@bitsundbolts working things are boring. I'm already tinkering with more broken stuff :)
AMAZING, 386 with 32MB RAM. This much memory used to cost ~1000 USD or even more back in the mid-90s.
a jumper added on that pin would allow it to be used in EDO or FPM mode!
Guess that would be some work for UpLateGeek 😃. Maybe he watches this video and reads this comment. Would be cool to switch between the types!
Unbelievably valuable tip! Thanks so much to you and to that guy explained about /OE and /CAS. The mod seems to be stupidly simple and cheap.
Very interesting video. I have 486-vip-io2 which accept only FPM memory but I have a lot of EDO modules so I have to try this.
Good job @Bits und Bolts! I liked this video and I updated my SIMM project based on these findings. I also left a note about your video on my project's page.
Keep up the good work!
Hey there! Thanks for letting me know and linking the video! I have visited your page "4Mx9 FPM/EDO 30-pin SIMM" in the past - I came across your nice project when I was researching the standard thickness of those PCBs :)
@@bitsundbolts Glad it's been of use to you! I initially built them for my DIY 80386DX singleboard microcomputer and didn't thought of them being very useful. But now, I see that a lot of people actually build new SIMMs. Which is good! This way, we help keeping those old PCs alive.
Great news! I might give a try and mod one of my 32MB 72PIN EDO to work as FPM as my Terratec EWS 64 refuses to detect its full capacity and the compatibility is broadest when FPM is used!
This is nuts! The first computer I ever built myself was using a baby AT mobo using this exact CPU. Tough little chip!
100nf between +5v and Gnd is a decoupling cap. Basically is acts as a UPS for the chip cleaning out any ripples in the power source and potentially power spikes.
This opens up a whole new world of possibilities on old Macintoshes which need and/or prefer FPM!
This is *exactly* the sort of thing I was looking for - I have a 486 laptop with 4 MB of FPM on a module, would love to upgrade it to its max supported 16 MB, however, I cannot get hold of any FPM chips. Thank you so much, I'll try to source some EDO chips and possibly make a module of my own!
Good luck! Hope it will work for you too!
The sound of memory counting brought me some good memories. My first PC was a Pentium 100MHz with 16MB of EDO Memory.
Also, it's incredible how we evolved since then. The EDO memories gave you a bandwidth of 20MB/s while nowadays a DDR4 3600MHz can give you a bandwidth near 60.000MB/s.
Meanwhile I'm here just waiting for a 1THz graphene processor...
Great mod, I maybe knewed this because it sound kinda familiar (lift a pin and make a bridge to use incompatible memory in 486), but maybe I am wrong.
About having "tons" of memory on old 286/386 it does not make any sense for running programs in DOS (I do not remember a dos program complaining about not having enough EMS/XMS memory with 8/16M of RAM), unless you use a ramdrive. I used on my 286 and later on 386 to have a folder (most used DOS commands, NC, archivers, etc) zipped and at boot, it was copyied and uncompressed on a ramdrive and the speed of the machine was increased. Setting the comspec and temp to the ramdrive also help increasing the speed, especially on old I/O controllers and motherboards that only support HDD PIO access. On my 486 I already had UDMA33 HDD access (VLB I/O controller) and no need to do that, the disk access speed was good enough to not feel a difference.
5:00 It's less "stabilizing", and more about shorting out high-frequency AC emitted by each IC when switching.
In the context of AC, capacitors can be viewed as frequency-dependent resistors. If gives high-frequency components on the Vcc line a low-impedance path to ground, so they don't travel along the Vcc line to other components.
Thanks for clarifying!
@@bitsundbolts It's important to visualize AC as something traveling along conducting paths. This also gives the explanation why non-terminated transmission lines show reflections. The AC component is traveling from one end to the other, there's nothing to consume it, so it travels back along the transmission line and shows up as reflection. A capacitor is a low-impedance path, so AC gets shorted out, i.e. shorted out. Or in the context of an oscilloscope, you might simply place a 50 Ohm resistor.
Oh right, I think I watched a video once where this was explained. I feel like those are all basics, but hopefully I get the time to learn more and more.
This is fantastic news, you and arti9m are Retro Heroes. Thank you for sharing this info. Will you eventually try to make a Parity Ram? is that possible?
Yes, I will continue to work on this project and get the parity ram chips. I also hope we get a new version from UpLateGeek to control EDO/FPM mode via jumpers on the board. I may give it a shot myself if I get the time.
@@bitsundbolts Excellent to hear. This is really exciting stuff, you are breaking some really important ground with this.
@@bitsundbolts Note however. While parity RAM gives you error detection or even correction, which is beneficial for meaningful long-term operation, the downside, unfortunately, is that it also results in poorer performance.
Absolutely beautiful hack holy crap
Yeah, it is pretty cool :)
Kind of reminds me of the various hacks between the 700 series sockets that I saw years ago back when the hardware was just new enough to still be used in ultra budget builds.
The maximum capacity of a 30-pin SIMM is 16MB. The POST memory count is six digits long, meaning it should be able to address 128MB of RAM (16MB x 8 SIMMs). I'd love to see a 386 running an absurd amount of memory.
The decisive factor for the size of the maximum RAM expansion is not the POST memory count, but the number of address and data lines that can be routed from the CPU to the RAM modules. The manual of the motherboard usually provides information about the maximum amount of RAM that can be installed.
I had a motherboard for the ALi M1429/1431 chipset, release of 1994, with 8 SIMM slots and 128MB of memory was indicated in the passport for it - there was a slot filling table.
That was really cool.
Nice video, thanks :)
Brilliant job! I used to run W95 on a 386DX40, it was painfully slow. It was on 16mb ram and a mechanical drive so maybe with 32mb and an SSD it will be a bit better, but I'm still not sure you'll end up with a system that is practical to use. You might actually have more luck with 98, but, I'd say OS/2 Warp would be a good match. Keep up the good work!
Thanks! Windows 95 was slow, but usable to an extend. 32MB and SD-to-IDE did for sure help.
@bitsundbolts ahh yes I just found and watched your video on it. That looks a lot better than I remember it being! Perhaps I will try again at some point.
Some people have had success overclocking a 386dx40 to 50mhz, they seem happy to do it with a small heatsink. I always had a soft spot for the 386dx40, they were fantastic for their time 😊
Vital information. Share it with anyone willing to hear it!
Amazing!
Freaky awesome!
Hm, a 386 DX-40 isn't bad! My 386 SX-40 would just, just run Doom (which is the whole point of having a PC in the early '90s). It needed Doom's lo-res mode and a couple of brackets to make the screen smaller, but with that you could reasonably play a slightly-jerky Doom. Of course getting my 486 DX4-100 was a huge thrill. Full-screen hi-res Doom 2! But yeah that DX will be just that bit faster.
EDO is a fine replacement for FPM. A motherboard will just ignore the extended data period and access it normally, so you won't get the extra 10ns but there should be no problem. Ah, EDO! Everyone thought it was a miracle, 60ns RAM instead of 70ns! Now RAM access times are down below 1ns!
I had Win95 OSR2 running on 386DX-40 with 8MB of RAM. It was extremely slow, but I needed it to run Office'97
I'll try that some time. I hope it's not too slow. There is a big difference though, I'll not use an actual hard disk, but an IDE-to-SD adapter. So, I guess my "HDD" will be quite fast compared to a spinning HDD back then.
@@bitsundbolts A paging file can quickly kill the SD card.
My fist pc was 386 dx40... That sound at the begining made me happy and sad at the same time.I wish I still have that computer. Thank you for the sweet nostagia.🥲
Good tutorial..
The "Left, Right, Center" part reminded me of the sound test on some dos games, not sure if that was intentional haha.
Well, seems YT deleted my prev. post coz a contained link, fok... I just want to thank you Arti9m and BuB for sharing this forgotten wisdom. A few months ago I designed a PCB for 72-pin parity replica SIMM module esp. for Alpha64 MB that requires FPM and parity and needs at least 4 same modules due to 128bit mem.width. I was hunting for non-parity FPM modules to canibalize them for DRAM chips and reuse on my own modules. If I would know the trick I could design the PCB already for EDO chips that I have plenty of them on EDO modules. So now I just took one bare PCB, mask OE# with kapton tape, soldered 8 EDO DRAM chips Fujitsu 8117405A-60 from a common EDO module, blobed OE# to CAS# and put it to a 486 MB for quick test. The module booted and now running Goldmemory test (Memtest 4 is kinda crashy on this MB, I found GM much reliable on 386 and 486, give a try) at fastest timing set in BIOS (0WS) and till now passed ~50% without error. I would just note one small drawback, that CAS# driver of memory controller sees then 2-times more capacitive load because of 2 connected inputs at one chip instead of one. But should not be issue for 8-chips module because it would be same as with 16-chip (double-sided) FPM module...
Happy to hear that it seems like it is going to pass! Let us know when you have 100% confirmation of the entire memory working.
@@bitsundbolts Yes, test passed with 0 errors. Later I will solder on the parity chips from another SIMM and try in AlphaPC64. When I will send another oder of PCBs batch I will send also SIMM design modded for EDO chips...
Great to hear! I got my modified EDO-modded PCBs today. Will test them this week and probably make my a video (if everything works since I had to reroute the entire schematic).
ALREADY SUBSCRIBED! Where else do I get to find out about swapping EDO into FPM memory?
This is crazy! Just the other day I was asking my good friend, GPT-4 (lol), about this topic. So awesome!
Wonder what your good friend had to say :)
@@bitsundbolts Haha, lots of pretty interesting information. I was specifically asking it about the differences between FPM and EDO RAM, and the different types of memory available in the 90s (like WRAM, VRAM, SGRAM, etc).
It's all stuff I could have looked up in research materials or an encyclopedia, but ChatGPT, especially GPT-4 is really good at summarizing information in a way that I understand, for whatever reason. It hits the salient points without too much unnecessary jargon or technical data.
Be careful as it says with very confidence very false things. It gave me a fake scrolling routine for the zx spectrum😂
@@jbinary82 Oh sure, yeah. These large language models aren't intelligent in the least, they're just "monkey see, monkey do" machines with very large training sets. It's easy to get lulled into the idea that you're talking to another being, but you're not, at all; they're literally no different from search engines that happen to understand natural language queries.
I've managed to install Win95 on my 386 back in the day and it was almost unusable because of lack of RAM. It would be interesting to see if it would be usable with 32MB of RAM.
I ran Windows 95 on my 386DX 40 w/32MB back in the day, was shockingly usable.
That's interesting. I never tried to. My 486DX had only 8 MiB of RAM. So I stuck with Win 3.1 until I had a more powerful machine.
We need a 10h video of pin OE pin CAS bridging
Your next mission, should you choose to accept it, is to turn registered memory into unbuffered. The amount of old server-grade DDR3 modules is insane, and we can't use any of that stuff because it's not compatible with standard motherboards. I don't think it's actually possible (something akin to putting diesel on a petrol engine)... but I also thought it was impossible to emulate a Sound Blaster card, and SBEMU showed up.
Yeah, that would be cool, there is a lot of old and cheap server/corporate hardware that would be cool to reuse in retro builds.
@@bitsundbolts Why not use the server hardware? It comes with ECC support and 2 or more CPUs on one board. The noisy fans can be replaced by quiet ones.
It can be done in some cases but it is a very complicated mod. Many RDIMM3 modules have a separate IC that does the buffering, according to JEDEC SSTE32882. If you unsolder these chips and bridge the inputs to the outputs, fortunately the layout is symmetrical, and PCB is designed for 1:1 MUX mode, you will in theory have an unbuffered DIMM. If muxing is not 1:1 you will loose some capacity as not all chips can be used. Of course the SPD will need to be updated to correctly describe the module.
Most chips with EDO support operating backwards compatible in only Fast Page Mode. They simply don't utilize the EDO access mode when connected to an older device that doesn't support it.
However, it may be possible to enable the EDO mode by modifying the firmware on the memory controller, effectively improving the speed of the RAM.
Note I said "MOST" Said chips were branded "EDO/FPM" chips. I suspect it may require them to be in a DIMM socket as well, instead of these smaller SIMM sockets, as the configuration pin might not pass through on the SIMM socket.
Additionally there is the possibility that this particular SIMM board lacks passthrough of the data out to the appropriate pin.
WAAAS?! Das ist ja krass. People simply baking their own RAM modules 0_0 Is that allowed!? I thought you need a factory to do that :D
Gute Arbeit!
My assumption with capacitors that small is that they're decoupling capacitors that keep rapid current draws from the chip from briefly sagging the power line voltage. I think almost every chip is supposed to have one, as close to the chip as possible.
I don't know the English technical term for it, but in German these capacitors are definitely called "Abblockkondensator". They are built as close as possible to the ICs.
This was a great and interesting video. This opens the door for you to also install Windows NT 3.51 and OS/2 Warp. They both work well with around 16 MiB and more. However, no WinNT 4.0 and no Win2K for you on this machine, the first requires an 486DX and the latter a Pentium and 64 MiB.
It might be a good upgrade solution for my 486DX with 8 MiB and a rather early 486 board. It is ISA only, no VLB.
But on the other hand, I've never needed more than 8 MB on my 486, because all my DOS games that run smoothly on the 486 and don't need more than 8 MB themselves, and those that might need more run on not anymore with the 486 anyway.
I've used a RAM disk in the past, but i didn't need a large RAM disk because with the amount of data, I used to get by with a small RAM disk.
Also, my 486 has a SCSI controller. This means that the SCSI controller relieves the CPU when copying data between SCSI devices.
But I see an advantage for Windows, since an upgrade would actually make sense. With Windows 3.1 I would have more RAM available for applications and with another Windows 9x license I could even use that. However, I have newer computers for Windows 9x, such as a Pentium 2 266 MHz and an AMD Athlon Barton 3000+. And I didn't actually work under Windows 3.1 anymore.
At the moment it would only be a nice to have for me. But the question of whether you can noticeably improve the speed with faster RAM would still be interesting. I see advantages here.
Win2K works quite well on the latest 486/Am5x86. You only need enough memory (at least 64 MB) and a fast PCI IDE/SATA controller with UDMA. Or fast SCSI.
@@ILRBW Did you try that? The system requirements of W2k say that a Pentium is the minimum.
The reason why it might work on a Am5x86 is because an Am5x86 does support some instructions that where introduced with the Pentium. CPUID for example is one of them.
But this doesn't mean, that W2K works on a typical Intel 486.
And then there are other instructions that were introduced with the Pentium but are not available in the Am5x86.
CMPXCHG8B for example is one of them.
If W2K uses CMPXCHG8B somewhere, the system will freeze or crash.
@@OpenGL4ever
the "official system requirements" written on the back of the box are more like recommendations ... yeah, sure ... the installer or bootloader will throw a warning or refuse to install or boot alltogether ... but some people look at that and go "yep, challenge accepted"
so ...just because I have not seen win NT or 2k on a 386 with my own eyes does not mean it is absolutely impossible ... maybe someone has already come up with a patched installer or some other hack that I am not aware of ... it would not be the first time this happened
the system requirements of win ME also say that a pentium is the minimum ... someone hacked around it
the system requirements of win 8 say you need a CPU with PAE, SSE2, NX .... someone hacked around it
the syetem requirements of win 11 say you need 8th gen intel or zen2 AMD ... and TPM 2.0 ... well ... guess what ?
@@KenjiUmino What you write is just nonsense. If the operating system uses opcodes that a 486 doesn't understand, it simply won't run.
A hack that replaces all Pentium opcodes is a hack and not the official version for which the system requirements were written. Therefore a hack is invalid for the question whether it officially runs on CPU XY or not.
And btw. you would also have to take this into account with all security patches and feature updates and drivers.
@@OpenGL4ever No, I myself have not tried to put W2K on the 486th. But I read messages from a respected Takedasun user from the "Полигон Призраков" forum, how he experimented with W2K on a 5x86@160 MHz SiS496/497 machine with a VIA6421A controller.
20 years ago, I came across information on how to choose the 486 core when installing Windows (W2K or W98 ?) in order to speed up the system on an old PC. Perhaps Takedasun installed the system in this way.
Google the topic "Максимально достижимая скорость работы дисковой подсистемы 486 компьютера" on page 5.
11:25 May be you could isolate OE in a SIMM connector without desoldering mem chips?
I did not verify it, but I think the PCB connects this pad to ground. There doesn't seem to be a dedicated pin for /OE on the 30-pin connectors.
Thx
Great video, maybe i should dig my 386DX40 board (I recall it had brand TOMATO board in it) that i used with only 8MB of RAM and see if the Varta battery has ruined it.
You definitely should do that - and if it's just to get that Varta battery off the board :) would be interested to hear what the condition of your board is.
11:26 You could have saved a lot of time by desoldering the pin in question and bend it upward a bit. Did that many times on expansions for my Amigas. ^^
Yes, did that as well. I think I did that for a PCB that swapped the ROM's 1.3 and 2.0 with a switch cut in the back of the case. Could be wrong, it was a long time ago. Amiga was great for hacks like that.
It might not be the best strategy in an era where the chips are no-longer produced. You could want to use them for EDO again later. The optimal solution is probably a modified board design that lets you jumper OE to CAS or GND with solder.
@@stevenclark2188 It is minimally invasive with only minimum danger to the chips. Just a little heat to the pad while pushing the pin upwards gently.
I think the second desolder and resolder session was more harmful to the chips.
Well, can confirm that my old 486 could run unmodified EDO SIMMs. Though they performed like fast page IIRC. Sometimes some Pre-pentium boards get along with some EDO, and some don't. Can't honestly tell you which works with which though. But I didn't have to modify the chips at all. I beleive I was also running an AMD 5x86-133 in it at the time, though given that the 5x86 is really just a clock quadrupled 486, plus maybe a cache enhancement, I doubt that that helped it tolerate EDO. Yeah, I held onto it late, after my family upgraded the household computer, it became my own. I did eventually drop a Pentium Overdrive in it when I found one dirt cheap on ebay. So maybe it was during that time, but doubt it since that was after I relegated it to file/web/and FTP server running linux when I dropped the overdrive in, and didn't use it as my daily driver anymore. So It had to be when I was still using it as my main PC. I wanted to be able to use the 586 architecture packages that pentium supported, but the 5x86 didn't, despite it's designation. So it ended service as a Pentium-83 running on a 486 motherboard, with 64MB of mixed FPM and EDO ram. And 120GB HDD. Yeah, the BIOS didn't support a drive that large, buuut it only needed to see the bootloader, since once linux bootstrapped in, it overrode the BIOS' drive parameters. Ran like a top till the 10MBit NIC couldn't keep up with demand. Along with some old computer stuff someone dumped on me, I had supposedly 100MBit nic, that was ISA, due to it having a mode LED for 100MBit in addition to one for 10MBit, but could never get it to work. So I retired it in favor of setting up a newer system I had retired from daily use.
Win95 on a 386DX-40 is no problem, I finetuned a motherboard for exactly this purpose for use by a friend and used AOL with it. You better did not open more then three windows in the browser though or it got a bit slow.
But I doubt this motherboard will do, the OPTI chipsets were lame AFAIK, Chips & Technology chipsets were the gold-standard when the full chipset was used which consisted of three (or even four?) chips, there was usually also place for 16 SIMMs which were able to address (at least) 16 MB of memory and when fully populated this was done 4-way interleaved - that was the biggest performance winner, this and handselected fast Cache-RAMs. Finding a fast enough TAG chip was a real hassle but it was worth it.
All the late 386 boards with soldered AMD 40MHz CPUs run at 50 MHz without issues, even faster but then only ISA cards with own clockgenerator can be used and even not all of those work.
But a OPTI board at 50MHz is still not nearly as fast as a fully stuffed out C&T board with memory interleave enabled.
Красавчег!!!
About Windows 95. I once installed W95 OSR2 on 386 SX (!!!) with only 4Mb of RAM onboard. I used 2.88 Mb Floppy install media (original W95 OSR2), I had 170 Mb Quantum FIreball HDD. It took me 5 hours to install but in the end it worked (though it was very slow as it used big amount of SWAP file to work, but I even ran Doom 2 on it (with 320x200 window) and Warcraft 1. I imagine that with 16Mb of RAM it will run very smoothly. Also small hint: in absence of sound card there's old PC speaker driver for Windows 3.11 which will work with W95 and will allow to output WAV audio from PC speaker.
That brings back memories. You see it was year 1997 and I had IBM PC DOS installed on that PC. That was my mistake BTW because Windows 95 will not work after installing over PC-DOS, only over MS-DOS installation will work.
Intersting. I guess PC motherboards don't use FPM read-modify-write cycles with a delayed WE. In these cycles OE must be driven inactive after the read half of the cycle before write data is presented, followed by a WE pulse. This mod would not work for this kind of use case.
Genius
if i could take this back in time i would be a god
You should make a version of the memory pcb with a header that allows you to switch... Then you could use both types of chips
Just now forked the repository. I will need some time to work on this, but I guess I have no other choice :)
@@bitsundbolts wow that was a FAST reply 😄
It was just a random though that floated around
@@bitsundbolts Do you have the right equipment to measure the signals of the individual address lines at this clock rate? Please note that it is not enough to simply install a DIP switch, the signals must also arrive at the RAM modules at the same time or at least synchronously.
echt interessant
So that must be how AWE32 doesn't care whether you put FPM or EDO sticks into it! Interesting.
RAM ICs are known to be sensitive to decoupling capacitor condition, them being damaged leads to a progressive irreversible damage. I don't know what it is exactly down to though, I mean it's fairly intuitive how it would influence data retention, but how does it kill them?
I am not sure, they may just support both timings.. There is no pin of the 30-pin connectors that is dedicated to /OE. So, they cannot directly control /OE from the soundcard.
@@bitsundbolts I kind of imagine they had to come up with some conservative simple and minimal logic to drive both types of RAM though without some sort of explicit detection. AWE has really benign memory bandwidth requirements in principle, unlike a mainboard or video card.
You could be right. Maybe, if speed and timings are not important, the AWE32 doesn't care how long the data is present. Maybe it's possible to implement both types without any extra logic, you just end up with low bandwidth for that compromise.
an interesting experiment. only to be honest, I don't see the need for more than 8 mb for a 386. I also have a 386 in my collection (and some more modern computers), but I've always been interested in how they can be put to good use. all collectors use old computers only playing retro games
7:00 why didn't you run the diagnostic card with a video adapter?
The system should complain about "No VGA card found". Since it does not do that, it did not even reach the point in the boot process to check for a VGA card. There won't be any VGA output without working memory.
The first motherboard I designed is 8088 😁
11:49 You could have just cut the pin with steel pliers, used a steel wire to solder the two pins together. Would have saved you lots of time.
I would have wanted to see if you could overclock the timings on the memory.
Desoldering wasnt a bit issue. The PCBs are quite thin and small. I think I spent maybe 30 minutes to desolder and clean the memory PCBs.
The 386 motherboard is very limited in memory tuning, but I admit, I did not look into this yet. Maybe in an upcoming video when I compare real FPM modules to those modified modules, some optimization can be done.
Having 16 or 32 MB FPM 72 pins modules (with parity) will really help with my SGI Indy that currently has no memory.
A 386 can address up to 4 Gb. Why not put in higher density chips? The most I've ever had in a 386 was 128 Mb RAM (it was a combo 30/72 pin board)
Why not cut the OE leg at the board, unsolder the leg and suck the solder, then solder a wire to CAS and OE? Seems easier than removing the entire module.... 16 times.
the only simple hack i know how to easyly implement is runing every socket 370 cpu at 133 fsb regardless of the actual procssor speed which in turn let you overclock on via apollo chipsets without runing into pci clock above 44mhz problem
for some reason via chipsets don't apriciate software overclocking and don't aply fsb dividers as advertised in bios, pci is alwayc locked to fsb in the same ratio as "stock" cpu fsb is locked to pci clock thet is 2 times smaller for celerons on 66mhz fsb, 3 times smaller for 100mhz fsb and 4 times smaller for 133mhz pentiums 3 fsb
however by simply insolating both bsel pins on the cpu you can bypas board fsb detection so it always think cpu is 133mhz fsb and aply 4 times divider for the pci
depending on a cpu you can ger 200% overclock or just 133% on some cpus you would never suspect, either making 1,7ghz celeron tualatin from the latest revision or just 1150mhz celeron coppermine from 566 coppermine cd0 model
a bit unnesesery oc info i know but that was the only way i could play battlefield 2 on celeron 566 in the past
Hallo, nice video. Can yuo give me the name and/or link of the software that you use to measure the RAM speed (see your video at 13:48). Thank you.
That is SpeedSys 4.78. It can be downloaded from Phils Computer Lab in a bundle of DOS tools.
As a tech, I am becoming more and more versed in "retro" builds. I am actually one of the only techs in the county willing to even work on them.
That being said... the further back I go, the harder it is to get good information. Now, my hands are definitely not steady enough to do the micro solders, but testing and restoring old hardware is something I am known for... and I have other people at my disposal to assist me with that. Videos like this is my way of continuing education, you could say.
I have a question. I am very familiar with MemTest 86, and I use it all the time. But, you used another program too. Speedsis? Maybe I didn't here the name right? Could someone point me in the direction of this? I would love to try it out, especially on the stacks of odd RAM I have from the '90s. I can't sell them unless I know that they are good.
Thank you, in advance, to anyone that may be able to be of assistance.
Google "SpeedSys 4.78 Dos Benchmark Pack" and you will find PhilsComputerLab website. From there you can download a nice selection of tools packaged nicely into one archive.
@@bitsundbolts thank you very much! I just downloaded it and am going to add it to my utilities toolbox
You don't need 32MB of RAM to install Win95. I remember that when it came out, some guys installed it on a 386 with 4MB RAM and it worked...somewhat. With 32 Megs, you could go for Win98, ME or even XP Lite. My first PC had 8MB RAM and came with Win95, so 16MB is already more than enough to get it running comfortably. 32 MB is almost overkill.
Interestingly enough, Win9x put such an overhead on the system, that when you wanted to run mere DOS games on it, their memory requirements easily went up by 8MB if they required EMM. But even for that, 16MB was the sweet spot. No DOS game required more to start under Win.
Windows 95 uses a paging file when RAM is insufficient. So more RAM makes sense, of course. This is especially true if you only have 4 MiB. For smooth operation you should have 16 MiB at least.
Nice Video. Win95 on 80386 DX40? :O) This is hard. This is more a Machine for Win3.11 or Tinyroot Linux
I remember all the hassle to sell edo-modules to buy older FPM modules. And you only had to connect pins. Where were you 20 years ago?
Anywhere I can purchase the chips from to solder directly?
If you want to use EDO chips and apply the mod, you can lookup 32 MB EDO memory modules and look for those chips on sites like eBay or AliExpress. You can also get FPM chips mentioned in the BOM file in the repository - you don't need to apply the mod when using FPM chips.
Unfortunately, I cannot guarantee correct operation if you use chips not mentioned in the BOM file.
@@bitsundbolts Thanks, I definitely will be giving this a go. Finding SIMM memory is so difficult these days.
hmm, version 2 of the PCB could use a jumper (or a jumper solder pad) to do this. so you can switch between EDO and FPM without having to do/undo the mod
I wish I could post a picture in the comments 😉. I am currently assembling version 2. Will post in the community area in about 2 hours (probably with a link to the GitHub repository or PCBWay direct order link). Hope everything works...
The pcb could be modified to allow the ~OE to be connected to either option, without having to bridge individual IC's
I already sent a modified version with this option for manufacturing. Should hopefully be ready by next week.
if /OE was tied to ground, the chip could never have data written into it. Perhaps that chip isn't right for those SIMM boards.
Wow those FPM Chips are expensive, I have some old PS/2 SIMMs with EDO and FPM Chips on them I could have send you for free, also shipping time would be much shorter because I live in Germany.
the first thing when i got 24mb ram, was run windows on ramdisk.
found a seller for 4Mb x4
so, what does building this solution cost in total ?
and if those 4Mb sticks do work - will it also reduce % failure ?
It depends how much the shipping cost will be to your place and if you chose PCBWay or a different manufacturer. 10 PCBs cost 5 USD + shipping. Plus the cost of the memory chips.
I mainly thought this would be a good idea if you have old modules that could be salvaged, like the memory chips from that one broken EDO memory module.
I don't know if it would help reduce failures, if you're talking about memory chips, then I would assume that the quality of memory chips got better over time.
Well, did you know that DOS is capable of memory compression?
I wanted to play some game that needed 12MB of ram. And I had only 8MB. There was lot of swapping when I tried to play it.
When I learned about disk compression and disk cache I was curious which happens before.
With smartdrive and drivespace my 486 had compressed data in its disk cache and it stopped swapping to disk.
Another trick.
Windows 95 himem.sys and emm386 when used with dos 6.22 improve performance. I would just recommend moving those files to a another directory and copying windows85 versions inplace of DOS version of those files.
To be honest, dukenum3D was fast enough for me to play with my 486DX 33mhz after I tuned the boot configuration files optimally.
I did have a decent graphics card, 256kb of relatively fast L2 cache 8MB of relatively fast memory memory modules since I saw far slower 50mhz 486::s with my friends.
If game uses dos extender often best performance comes from loading only himem.sys without emm386.
With 33mhz 486 I was constantly trying to work around its limitations as a child. There was constant goal of trying to free up conventional memory so that the memory extenders had less work to do. So disk-cache was only used when needed etc... I've forgotten most of the details how I made it perform the way it did, but these are some things that made such impact that I remember them today.
Why did you have to desolder the chips again? Just desolder the pin and lift it from the PCB with a small pointy tool..
I didn't want to bend any pins. That is really all that is to why I decided to remove the chips again and isolate that one pad.
These days I think you could simulate 386 ram with fpga. Actually you could probably simulate the whole 386.
CPU L2 cache module as RAM when? :D
On old PCs, so I head, parity is a bad thing anyway. A parity fault causes the MB to assert INT 2 which pops up and tells you your RAM is buggered and halts the machine, there and then. So it's actually worse. The ideal solution I suppose might be to run a RAM test now and then on un-parity chips. In modern PCs it's different if they employ ECC RAM, it can detect an error, and actually correct it, for one bit per word (usually 64 bits). Since it knows where the error is, it's a simple matter to correct it. If it's a 0 and it's wrong, then it obviously must be a 1! ECC can also detect if more than one bit per word is faulty but can't correct that.
Look up Hamming code, it's so simple it's almost obvious. At least, once you know it, it seems obvious. Actually inventing it was a work of genius. It's what makes ECC work, needing 8 bits of "parity" to protect 64 bits of data.
Generally though RAM is pretty reliable, though as it increases in capacity it means the transistors get smaller and more susceptible to cosmic rays zapping them. Really though you're unlikely to have a problem with a single PC. Maybe with racks full it's something you need to be wary of, but again ECC takes care of that to the level that it likely won't bother you.
👏👏👍
Hey duuuude, that makes this 386 with an FPU the BEST test bench for the first true 32Bit real multitasting, not semi-preemptive, OS specifically OS/2 Warp 3 (which I still miss and think should have been what 95 ended up being in spite of 95 being pure shitty crap and not even a real OS, only 2K and XP fixed that on the broad market when OS2 could have done it years prior.. sad sad story
You have a very strong opinion. I accept that. I have never used OS/2 Warp 3, but I may check it out.