I upgraded an IBM XT to a 486 via a 286 upgrade card. I then upgraded the 286 to a 486 slc chip. Was nowhere near the speed of even a true 486 SX system, but a LOT faster than the original 8088.
I’m glad you’re back! Maybe you can swap the crystal on the motherboard while running the the 386 upgrade? I suppose you’ll run into the clock limit on those old chipsets. Of course there are 20mhz 286 chips too. Perhaps a fastest 286 ever test?? 60ns RAM, 0ws and an over clock!
yeah, would be a try worth to change the crystal. Hah, yeah, maybe I should make a follow up video to my 486 record and make the fastest 286 board. 😇. Actually I am working right now on another 286 video. Its a unique board at the end which deserves to get called the most beautiful 286 board. 😉.
@@wishusknight3009 Someone lent me one of those MakeIt486 boards for a video but I never made the video I wanted to. (And I gave it back to the friend.) The thing was a PLCC board just like the 386 board in this video -- but it had cache and a 486SLC: www.cpushack.com/2014/08/30/improve-technologies-make-it-486-286-upgrade/
@@dalecomer5951 The biggest issue is the chipset -- unless it was one of the very latest that was designed for faster 386SX clock speeds, most earlier boards (At least all the ones I have) top out at 12mhz or 16mhz.
He will still need a software tool to activate the cache. But CacheCheck may actually see it installed. Betting pool anyone? lol I wonder also if an A20# handler would activate it?
@@wishusknight3009 All I wanted to know was if it would see the cache... or for that matter, post with a leaning tower of stacked socket upgrades - I'm a man of simple desires, lol. My guess is that the cache controller is mapped to HMA, but I'd suspect some additional code would be required by an A20 handler to kickstart the cache controller to do it's magic of intercepting the address lines and transparently caching data. That said, I'm just guessing - though I'm dying to know what the controller is and how it's all implemented. I've never heard of a 286 solution that involved a cache controller (not the least of which is because, while I think that 386sx might somewhat benefit from cache, as it maintained things like the dedicated ALU bus of the 386dx, along with 32bit wide paths, the 286's relatively simple internal architecture seemed pretty well tuned to its native external 16bit bus).
VERY nice video! I have always had a soft spot for these 286s and upgrades ever since I worked with/on them in Berlin in the early 1990s. Your channel is impressive.
Great to see you back. Indeed, my first thoughts on the cache upgrade were also that it must have had some software. The SX/Now needed a tool to enable its own cache, but it is doubtful that it would work on your upgrade board. Suppose it wouldn't hurt to try it if you could find it. Seems the Tandon and Toshiba were not the exception, then, meaning the 386 is for sure slower than the 286 at the same clock. Good to see this confirmed.
I like your style "Get the proper tool", proceeds to use anything in hand to do the job. Very old school and informative. You are cool. Hope someone replies about how to fix the cache thing tho 😅
In the early 90's I used similar parts from Kingston in order to upgrade IBM PS/2-30 ( 286@16MHz ) into ( 386SX @ 25MHz or 33MHz - I don't recall the speed exactly ). The Kingston daughter-board even had a socket for a 387SX math co-processor and the kit came with a replacement bios-chip.
4:43 Aah the old xt power supplies, common ground is superbound, common red is dead. I will never forget this sentence. Great upgrade options, i think those are quite rare nowadays.
There may be a way to run Win95 using that 386 upgrade chip! If you can manage to track down an "AST Advantage 202051" ISA card, it can add an additional 3MB of memory! I used to have one when I was a kid (30 years ago). My dad's work retired one, and he managed to bring it home for me to play with on our old 286. It was amazing to watch my 286 slowly count all the way up to 4 megs of memory at boot-up. Wasn't good for much on a 286 though, as most programs that could make use of the extra memory also required a 386 processor or better. I had bought a copy of Mortal Kombat for PC, but it refused to run because of the 286 processor. I had always wondered if getting one of those 386 (or the Make-It-486) upgrade chips would've made a difference.
So glad you’re back, my friend! This was a fantastic video and I’m super interested in the 386 upgrade. Why would any user ever purchase something that only enabled 386 protected mode? How strange! It’s a shame about that cache module not working, but I hope you can figure it out; if anyone can, I know it’s you!
Cool I used to have a 26 when I was younger I looked at the BIOS you had I think I had the same functions that was somewhat 30 years ago and it was still interesting to go back and look at this older machine I still have my 286 motherboard I must see I miss the old days of the 286 386 CPUs this is where technology was starting to open up this was the beginning of things going much faster I have seen this technology evolve and I'm still amazed at the calculations some of these machines can still do if you would have to do all this calculation VI pen and paper it would take us so much longer however thank you for the video it brings back memories
I remember my dad replaced our 286 with a DIY 386 back then (granted, I messed up the 286's harddrive with a speaker magnet) that he upgraded all the way up to it being a Pentium... Overdrive probably. This was the coolest thing for me when I was young, that said it's sad your chip sandwich does not work as intended.
In the late 1980s there was a DIY project in the German c't magazine for exactly that purpose: 386SX onto a 286 mainboard. Project was called PAK SX (PAK = Prozessor Austausch Karte = CPU exchange/swap board)
As always, fascinating. It may be the chipset just not knowing what to do with the 386. I feel that there is a solution to be had and Mr Black make some brilliant observations. :-)
The 386 SX 16 upgrade PCB you showed came on my old Wang desktop I had as a kid. I am not sure if it was made for those but it definitely had that installed.
I love the idea of running advanced tech alongside older tech. I am currently building a fast 486 build just so I can mix PCI, VLB, ISA, POD83 and 5x86s alongside each other in different configurations.
I'm glad someone found a use for all those trays of smt mount i386's. If I remember correctly they were still heavily used on some industrial embedded stuff (i presume medical related stuff they didnt want to rewrite from scratch) hence their really late manufacturings. Pretty cool!
yeah, the only upgrade I did of CPU the last 15 years was from a 5820K to a 6990X. Nothing fancy like stacking components on top of another, what a time that was.
Can you think of any tech, or even any consumer product previously that people turned out to be willing to throw away every three years or less for a new model? THAT was strange at the time. Upgrading little by little made more sense, if it weren't for Moore's Law.
In the early 90s, I do remember coming across a few IBM PS/2s ... can't remember the model, but they were one of the ones that originally came with a socketed 286. It had a 386sx daughterboard in place of the 286. It looked very similar to that first one you showed.
Maybe not time to give up on the cache module just yet. Perhaps those 100ns drams are being run at 0ws anyway (83ns) As I was able to use 100ns Ram in a little project I had going that required 80ns back in the day. Would be interesting to try on 20Mhz 286. A speed where the 100ns DRAMS will definitely not go. If it still doesn't gain performance, then I agree the 6k cache is dead. As for the 386sx being slower, I remember my 1st upgrade was from a 286-16Mhz to a 16Mhz 386, and it was a massive disappointment. If anything it seemed a little slower. HOWEVER, whereas my 286-16Mhz was unstable overclocked at 20Mhz. My lovely 386sx cpu would over clock to 25Mhz. At which point it was a big improvement after all :) Plus it ran windows3 properly :)
Great video man, we've been missing you. Very strange that the upgrades had no effect. Could it be that maybe the CPUs were made for specific motherboards that support them officialy? Nice ending with Hexagon there, I played a lot of that game haha. Thanks again and have a great weekend!
In short, no. Intel "Aboveboard" 286 and 386 ISA cards were the favored corporate upgrade, and the best strategy for clone vendors was to sell entire motherboards as upgrades, or entirely new computers. CPUs were quite pricey compared to other components. This niche product's only survival strategy was to appeal to as wide a variety of clone owners as possible. They mostly failed as a commercial product anyways. During this time it was unclear how planned obsolescence might work in the PC marketplace. Theoretically, maybe, some company might have tried a one-board upgrade, maybe for IBM or Compaq motherboards, but it would have been a remarkably foolish business decision, and most likely they would have been branded quite clearly, if that were the case. Seems highly unlikely.
alright :) now lets do this the otherway around ;) squeeze a 286 cpu into a 386 board :) i totally expected that 386 being used through the cache improved adaptor. glad you are back :)
Very cool parts! Would be really interesting to see that cache thing running. But nice to see, that my guess about the 386sx yesterday in the pre-announcement wasn't wrong ;) This were similar results, as what I got in the past. 286 is always slightly faster at the same frequency, then a 386sx. However, with 386sx you get huge benefit in Win3.11, since 386 has better support for protected mode and even more important VM86, which allowed to run more than one real mode program in windows in parallel. With 286, windows had to be closed during the execution of a real mode applications and was restarted again, when this application was finished. Thank you very much. I'll ask around, if someone knows anything about that cached 286 thing.....
maybe those cypress memories are not a cache, but, just some faster direct mapped memory, probably used by special sw that put critical code and data that needed to run as fast as possible.
I skipped from an 8088 to a 386, so I never encountered anything like this. Wow! That's so interesting.. i was cringing when you put the AMD chip in knowing how they have to be carefully removed (i guess i broke a socket like that in the past)..
As i suspected, 386SX brings nothing performance-wise. As for that cache board - if RAM runs at 12Mhz with 0 wait states, than cache can not possibly do anything more. The only reason why cache is installed is that it runs at higher speed and lower latency than system RAM. If RAM already runs at CPU speed with zero latency, there is nothing you could do to speed the transfer up with caching.
If that was the case here though you'd expect performance would actually slow down a bit with the cache due to cache miss overheads. Pretty likely the cache is just not activated at all, although I suppose it is possible the cache controller tests ram throughput on boot and deactivates itself on purpose if the RAM is fast enough already?
They really sold some strange parts back in the days. I started with a 286 at 12mhz, then I asked for an upgrade, and they told me I need a different mainboard for the 386, but they have a cheap one that works perfectly for me (I was a student back then) and they take the other parts in commission for me. I trusted them and was not disappointed. After upgrading to an 386 with mindblowingly 33mhz, it ran like warp speed for me. And the best thing: The mainboard was somehow compatible to the 486 which I put in 2 years later (1995) and sold the 386 again. Now I had a 486DX2 80mhz and again, 2 years later I upgraded to a DX4 from AMD with 120mhz and overclocked it to 133mhz, so my 3DFX Voodoo card had something to do :D Such good times.
I have the feeling that the answer is yes. The 486SLC CPUs could be paired with 387SX and 287 FPUs (depending on the motherboard) and their buses had the same widths as those of the 386SX.
Very nice ! For the cache adapter, maybe it's possible to retro-engineer it using software in some way ? Don't quote me on that but maybe you can use a software to "sniff" the ports on the bus and therefore find those used by the adapter board ? (then you'll have to try many things to figure it out but anyways).
That is a really nice motherboard. Plus cache if king too. I remember purchasing a p60 on clear out and loading up the board with either 1 meg of catch or 512. Anyway what a little speed demon. In some area it out performed a p75 and the horrible cache on a stick.
I have a similar little 386 board installed into the 286 socket on my IBM 5170. It's called the "Intel SnapIn 386" but mine has a 2.5x clock multiplier and 16kb cache right on the upgrade board. The multiplier is always active but the cache gets enabled by software - and even with the 20MHz clock speed, most benchmarks show slightly less performance than I got with the original 8MHz 286 if the cache is not enabled! Really goes to show how badly the 386 CPU is hampered by the slow memory pathway on this board (main RAM tests at at roughly 2MB/sec whereas the 16kb cache does about 10MB/sec). I wish I could mod it to have, like, 1MB of cache onboard!!
At first glance I thought that 386SX upgrade might have been a Kingston SX Now, but after a quick google it looks much simpler, which is probably why it was no benefit. One of my IBM Model 50s came with a 386SX-20 upgrade boardr, which is also a lot more complex, including a few larger support chips and an 8K SRAM. Although it is a single 8-bit SRAM chip, so it would take a minimum of two clock cycles to retrieve any data, since the 386SX and 286 uses a 16-bit data bus. Unless it's doing some weird clock multiplying, but I feel like that would be way too complex for such a small upgrade board.
Yeah - at 6:34 your PLCC socket is cracked. The sockets weren't designed for many insertions/extractions - I cracked the sockets on a couple of PS/2 Model 50Z planars testing the Kingston "386 Now!" upgrades.
I'm guessing at 12MHz the ram is running at system speed with no wait state? Hence the cache would have no effect and might have been working all along.
Thats a very good point. I need to check that. but i guess the RAM is running on a waitstate. on another 10 Mhz 286 setup i have i get better results with the benchmarks. there i know the jumper setting to set the waitstate to 0. the board in this video here does not have a jumper for that. 🤔
@@CPUGalaxy The IBM AT runs at one waitstate. If your 286 at 12MHz performs "like an 18MHz AT" (quoting Landmark), this shows quite clearly that your board uses less waitstates than the AT, which means 0WS. The 286-10 that is faster likely was faster in a video-heavy benchmark like the 3D bench you showed. I expect the 10MHz 286 board to just overclock the ISA bus to 10MHz, whereas 12MHz 286 boards often used additional waitstates or dynamic clock spreading to limit ISA speed for compatibility. Furthermore, that Trident 8800 graphics card is not exactly known to be on the fast side of ISA cards. Trident got good performance with the 8900CL, peaking at the 8900D (which is claimed to max out the ISA bus like an ET4000).
I feel like the 386 adaptor was made for upgrading some machine tool equipment. Maybe some upgrade offered by a company to run their newer software with just a plug and play hardware upgrade. I don't know honestly. You just get these feelings when working on those old types of machinery. Such as getting a 486 based press brake running in 2021 and seeing the mixture of upgrades and repairs that have happened over its life.
Pretty sure someboby wrote it already in the comments, but oh well, combine the 386 upgrade modul with the cach module and run your software to test if this cache is even recognized.
After seeing the video, think you missed some jumpers to increase the clock accordingly to the processor. Also, the big justification to use 386 instead of 286, is the 386 can run workgroups and is network compatible who is not with the 286. I still use a 386DX40 with a stepper isa board to control an OZO PCB prototyping router/etcher/driller/ for manufacturing Motherboards in 1987. I have a 286 Zenith working but not network. The 386 is a slave of a win7 to transfer the Gcode to control the OZO (jot) unit.
Maybe try popping the 386 upgrade on that cache board. And isn't there some flags you have to change for other cache enabled cpus sometimes. Kinda like what they cyrix utilities do for those chips?
Very cool video! It's not surprising that the 386sx performs similar/worse than the 286, the bus drivers would slow it's ability to interact with the bus slightly. Unlike other steps in CPU generation the 386 actually has a similar IPC to the 286, with some instructions being faster on the 286 and vice versa. For 16bit software in general 386 and 286 systems of the same clock speed are generally roughly equivalent. Even proper 32bit 386 machines are similar in IPC, the main speed difference being they can fetch instructions faster. What differentiates most systems of this period is actually the mainboard, memory and graphic card. A chipset with cache, faster RAM, or a faster graphic card makes a far greater performance difference than the CPU given the same CPU clock speed. The only way to make the best use of these upgrades is to have the software for the cache, or slightly increase the clock speed on the board for the 386 (it can do 16Mhz).
Both of these adapters are things I've been trying to buy for a long time! Super super rare stuff. The cache is probably working fine, but is mostly there to compensate for slow 286 bus access to very slow DRAMs at probably 150ns. Might work better if not for using faster modern memory. I'd expect a tiny boost despite that, though, so maybe there is some software involved. The 386 will have worse performance per clock because of it's strangled bus, which means to do the same amount of work it must also take twice as many memory fetch cycles. Disappointed you didn't combine the 386SX with the cache shim! It has a higher chance of actually having a benefit.
Can you find any name or number on that cache board. The number that can be seen is not enlightening, seems common to many PCBs And secondly, the 286 was always rumoured to be a better DOS CPU than the 386sx - seems to be true Can you slow a "real" 386sx system to 12MHz and get the same result, just to prove it's not the adapter?
I guess the only reason would be some Netware or UNIX that needed a 386 for memory management. Either way, intresting video. Would also like to see the "386 on the cache stack" and/or a crystal replacent to get it up to 16Mhz.
Microsoft had been toying with Xenix, or I should say they actually put out Xenix for 286 as a commercial product, but changed course and decided to put their efforts into the Windows GUI. In no small part thanks to the success of the Mac. MS could avoid competing with established Unix vendors, while encouraging the clone makers to stick with them, and weaken IBM's position in their partnership. It was "Game of Thrones" level intrigue!
From what I remember, in practical terms the 386SX was nothing more than a 286 that could run 386 code and its main advantage was being able to run at higher clock speeds (than 286). For most hobbyists and enthusiasts, it was actually cheaper to just replace the motherboard/CPU. Those CPU upgrades only made (some) sense for proprietary/business systems.
The 16/32 thing on the SX was, allegedly, so that OEM's could use as much of their existing PCB layouts as possible. So it's kind of weird these little bolt-on modules weren't a more common sight even if just soldered straight to the mobo from the factory.
The socket sandwich is pretty ridiculous to see, too bad the cache upgrade did not engage. I'll hope you'll get more info on these thing, it's pretty intriguing stuff.
Just started watching. I'm *thinking* the 386SX plugged into a 286 board won't really be any faster on 16-bit code; except that a 386SX can clock higher than a 286 if both are OC'd. I look forward to seeing the results :)
I am no expert on this at all, but just really feels you need some extra drivers to get the extra performance? Glad you're back btw! Hopefully you're all doing well! :)
IIRC, the cache in these upgrade chips sometimes required a BIOS update to enable the cache. Which, as it turned out, was not always available for specific motherboards. There were compatibility issues. The only real advantage to the 286-to-386 thing was it gave you 386 enhanced mode for Windows, which was not available for 286-class machines. As for the memory limitation, there were peripheral cards that allowed more addressable memory. Four- and eight-meg cards were easy enough to find (but expen$ive!) So, in many cases, it was often more cost-effective (long-term) to spring for a new machine than to 'upgrade' what you had.
It was known that 286 was clock for clock faster than 386sx. That's why 286 with Harris 20 or even 25 Mhz PC still sold long after introduction of 386.
As others have mentioned the 186 and 286 execute instructions faster clock for clock.. The 386 advantages were in 32bit instruction set which hardly anything used at the time. A flat virtual memory addressing mode used by windows. The next intel processor with this slower next gen. The 486 is faster than the first Pentiums.. Pentiums had far higher clocked versions. Another fun fact about the early 486s is that for some applications it ran faster on un-cached motherboards because the super fast 486 burst mode 128 bit memory read. If your program read through data linearly the 486 could read effectively at 1/4 wait state. So for example if you are calculating inner products to multiply matrices you could exploit this quirk. Later higher clocked 486s always preferred a cache.
What crack are you smoking? The 486 was NOT faster than the first Pentiums. Not just in Quake. Even in Doom a 60 MHz Pentium performs better than a 486 DX4-100.
Cyrix has a utility to active Cache in the operating system. May be worth a try. You'll also need to upgrade the crystal. While you would be over clocking the ISA slots, you might be able to get away with it, since most of those slots will work at a faster speed. I've heard the a clock for clock cycle between the 286 and 386, the 286 is actually faster. You have to get a fast clock speed to gain the benefit. Hope that helps.
there were some pretty good snap on 286 upgrades (less of a gimmick) made by IO DATA for PC9800 - by the nature of the architecture, most are compatible.
It would be interresting to reverse engineer the 286->386sx adaptor... Also, if it were possible to put in an SLC2 chip, it would have clock doubled and have some cache.
You could theoretically put a 486 chip to this adapter as it exists already. Cyrix made a clip on version of the SLC2 upgrade, Cx486SRx2 in 16/33, 20/40, and 25/50MHz versions.
Even if a 386 upgrade is slow, it's still better than adding cache to a 286. They both make the system faster, but cache doesn't allow you to run 386 software, so the extra functionality makes it worth it. If you don't plan to use your system to run 386 software, like, if you have a 386 system for that already, then go with the cache.
There are some 486 upgrades that clip on top of the 386's (saw one from Cyrix some time ago, I think it disables the original chip by grounding one of the pins), maybe you could test it 👌👌
It's the '-FLOAT' pin of the 386SX CPU that disables it - which wasn't set up well initially by Intel, and usually required to 386SX stepping to be the 'B' or 'C' (as marked for the CPU in the video). Clip-over upgrades are I little more rare now - I've just picked up six TI CPUs that I will test soon.
@@BaguetesGarage: IBM made a module that mated to a 286 socket that had a 486SLC2-50 (able to run all Intel 486SX instructions, with a 16Kb L1 cache) - and although it is proprietary to not be a conventional CPU socket, they also had the fastest systems with a 16-bit data bus / 24-bit address bus CPU that was produced: the 9556/9557 PS/2 with a 486SLC3-75.
@@BaguetesGarage businesses weren't sure the public would be willing to throw away their old PCs for new ones. Windows 3 then moreso Win95 proved they would. Intel Aboveboard 286 and 386 ISA cards were expensive upgrades and most consumers decided just a little more money for an entirely new computer (or at least a new motherboard) was well worth it. A little later IBM arrogantly gambled with Micro-Channel Architecture that customers would abandon clones entirely to get the latest IBM-brand technology. "Open standards" as championed by Compaq (not Open Source) and changed not only PCs, but the entire tech industry.
I have similar 286 upgrade wirh 3pcs of CY cache chips and a controller: ALL Technologies Chargecard II. It can be used for memory management like EMS with multitasking options, isnt a cache upgrade unfortunately...
I upgraded an IBM XT to a 486 via a 286 upgrade card. I then upgraded the 286 to a 486 slc chip. Was nowhere near the speed of even a true 486 SX system, but a LOT faster than the original 8088.
I’m glad you’re back! Maybe you can swap the crystal on the motherboard while running the the 386 upgrade? I suppose you’ll run into the clock limit on those old chipsets. Of course there are 20mhz 286 chips too. Perhaps a fastest 286 ever test?? 60ns RAM, 0ws and an over clock!
yeah, would be a try worth to change the crystal. Hah, yeah, maybe I should make a follow up video to my 486 record and make the fastest 286 board. 😇. Actually I am working right now on another 286 video. Its a unique board at the end which deserves to get called the most beautiful 286 board. 😉.
Who was it that posted a video of a 486s PLCC upgrade for a 286 board that used a cyrix slc?
The Harris 286-25 would allegedly overclock to 30 or 33 mHz with no special effort other than maybe socketing the crystal.
@@wishusknight3009 Someone lent me one of those MakeIt486 boards for a video but I never made the video I wanted to. (And I gave it back to the friend.) The thing was a PLCC board just like the 386 board in this video -- but it had cache and a 486SLC: www.cpushack.com/2014/08/30/improve-technologies-make-it-486-286-upgrade/
@@dalecomer5951 The biggest issue is the chipset -- unless it was one of the very latest that was designed for faster 386SX clock speeds, most earlier boards (At least all the ones I have) top out at 12mhz or 16mhz.
Your channel brings back so many memories
I used to collect older CPUs too
the cpu looks like a cheese burger from how thick it is with all the chips stacked together like that
What happens if you stack the 386 upgrade on top of the cache upgrade and run cache check?
I just ask myself why I did not have this idea 🤦🏻♂️. I need to try that.
@@CPUGalaxy I probably shouldn't be this excited by the prospect of what the result might be.
He will still need a software tool to activate the cache. But CacheCheck may actually see it installed. Betting pool anyone? lol I wonder also if an A20# handler would activate it?
@@CPUGalaxy maybe make it one of those 60s shorts that are all the rage lately? glad to see you back, keep up the great work!
@@wishusknight3009 All I wanted to know was if it would see the cache... or for that matter, post with a leaning tower of stacked socket upgrades - I'm a man of simple desires, lol. My guess is that the cache controller is mapped to HMA, but I'd suspect some additional code would be required by an A20 handler to kickstart the cache controller to do it's magic of intercepting the address lines and transparently caching data. That said, I'm just guessing - though I'm dying to know what the controller is and how it's all implemented. I've never heard of a 286 solution that involved a cache controller (not the least of which is because, while I think that 386sx might somewhat benefit from cache, as it maintained things like the dedicated ALU bus of the 386dx, along with 32bit wide paths, the 286's relatively simple internal architecture seemed pretty well tuned to its native external 16bit bus).
VERY nice video! I have always had a soft spot for these 286s and upgrades ever since I worked with/on them in Berlin in the early 1990s. Your channel is impressive.
Thank you!
i had high hopes for the adapter thing..... great to see you back! servus aus österreich^^
Glad you are back! Thank you very much!
Absolutely the best nerd-pron on RUclips.
But specifically OG nerdom. Neckbeard-p0rn
Great to see you back. Indeed, my first thoughts on the cache upgrade were also that it must have had some software. The SX/Now needed a tool to enable its own cache, but it is doubtful that it would work on your upgrade board. Suppose it wouldn't hurt to try it if you could find it.
Seems the Tandon and Toshiba were not the exception, then, meaning the 386 is for sure slower than the 286 at the same clock. Good to see this confirmed.
I like your style "Get the proper tool", proceeds to use anything in hand to do the job. Very old school and informative. You are cool. Hope someone replies about how to fix the cache thing tho 😅
In the early 90's I used similar parts from Kingston in order to upgrade IBM PS/2-30 ( 286@16MHz ) into ( 386SX @ 25MHz or 33MHz - I don't recall the speed exactly ).
The Kingston daughter-board even had a socket for a 387SX math co-processor and the kit came with a replacement bios-chip.
Your channel is awesome. It’s great to see the old chipsets like UMC, Chips, etc.
Heya. Welcome back, Another quality video!
4:43 Aah the old xt power supplies, common ground is superbound, common red is dead. I will never forget this sentence. Great upgrade options, i think those are quite rare nowadays.
There may be a way to run Win95 using that 386 upgrade chip! If you can manage to track down an "AST Advantage 202051" ISA card, it can add an additional 3MB of memory! I used to have one when I was a kid (30 years ago). My dad's work retired one, and he managed to bring it home for me to play with on our old 286. It was amazing to watch my 286 slowly count all the way up to 4 megs of memory at boot-up.
Wasn't good for much on a 286 though, as most programs that could make use of the extra memory also required a 386 processor or better. I had bought a copy of Mortal Kombat for PC, but it refused to run because of the 286 processor. I had always wondered if getting one of those 386 (or the Make-It-486) upgrade chips would've made a difference.
So glad you’re back, my friend! This was a fantastic video and I’m super interested in the 386 upgrade. Why would any user ever purchase something that only enabled 386 protected mode? How strange! It’s a shame about that cache module not working, but I hope you can figure it out; if anyone can, I know it’s you!
Cool I used to have a 26 when I was younger I looked at the BIOS you had I think I had the same functions that was somewhat 30 years ago and it was still interesting to go back and look at this older machine I still have my 286 motherboard I must see I miss the old days of the 286 386 CPUs this is where technology was starting to open up this was the beginning of things going much faster I have seen this technology evolve and I'm still amazed at the calculations some of these machines can still do if you would have to do all this calculation VI pen and paper it would take us so much longer however thank you for the video it brings back memories
I remember my dad replaced our 286 with a DIY 386 back then (granted, I messed up the 286's harddrive with a speaker magnet) that he upgraded all the way up to it being a Pentium... Overdrive probably.
This was the coolest thing for me when I was young, that said it's sad your chip sandwich does not work as intended.
In the late 1980s there was a DIY project in the German c't magazine for exactly that purpose: 386SX onto a 286 mainboard. Project was called PAK SX (PAK = Prozessor Austausch Karte = CPU exchange/swap board)
As always, fascinating. It may be the chipset just not knowing what to do with the 386. I feel that there is a solution to be had and Mr Black make some brilliant observations. :-)
The 386 SX 16 upgrade PCB you showed came on my old Wang desktop I had as a kid. I am not sure if it was made for those but it definitely had that installed.
I love the idea of running advanced tech alongside older tech.
I am currently building a fast 486 build just so I can mix PCI, VLB, ISA, POD83 and 5x86s alongside each other in different configurations.
I'm glad someone found a use for all those trays of smt mount i386's. If I remember correctly they were still heavily used on some industrial embedded stuff (i presume medical related stuff they didnt want to rewrite from scratch) hence their really late manufacturings. Pretty cool!
Good to see, you're back 🙂
Upgrades components was very strange at the time ! Interesting video !
yeah, the only upgrade I did of CPU the last 15 years was from a 5820K to a 6990X. Nothing fancy like stacking components on top of another, what a time that was.
Thats because motherboards where very expensive back in the day.
Can you think of any tech, or even any consumer product previously that people turned out to be willing to throw away every three years or less for a new model? THAT was strange at the time. Upgrading little by little made more sense, if it weren't for Moore's Law.
Very interesting, rare stuff. Loved the video.
I miss those day when i could do 80 MHz on amd 386 DX 40 MHz with crystal change. I had a realtek vga with 256K memory :D
And it would still performs worse than a 486SX-33.
It''s quite funny seeing your AT power connector still bearing the marks from when that capacitor blew up on camera ;)
😅
I remember Evergreen making these overdrives when I was still a kid, but already passionned!
3:30 THE TOWER OF POWER!
4:27 that cpu insertion sound is so nice
In the early 90s, I do remember coming across a few IBM PS/2s ... can't remember the model, but they were one of the ones that originally came with a socketed 286. It had a 386sx daughterboard in place of the 286. It looked very similar to that first one you showed.
So nice. A serious looking "upgrades" that turn out to be downgrades. Also... i like burn marks on your psu plugs
thanks for this upgrade . perfect experience
I still see the burn mark on your at power supply connector from the exploded cap 😂
hah. yes. 😅. I can still remember this shock.
Maybe not time to give up on the cache module just yet. Perhaps those 100ns drams are being run at 0ws anyway (83ns) As I was able to use 100ns Ram in a little project I had going that required 80ns back in the day. Would be interesting to try on 20Mhz 286. A speed where the 100ns DRAMS will definitely not go. If it still doesn't gain performance, then I agree the 6k cache is dead. As for the 386sx being slower, I remember my 1st upgrade was from a 286-16Mhz to a 16Mhz 386, and it was a massive disappointment. If anything it seemed a little slower. HOWEVER, whereas my 286-16Mhz was unstable overclocked at 20Mhz. My lovely 386sx cpu would over clock to 25Mhz. At which point it was a big improvement after all :) Plus it ran windows3 properly :)
Great video man, we've been missing you. Very strange that the upgrades had no effect. Could it be that maybe the CPUs were made for specific motherboards that support them officialy? Nice ending with Hexagon there, I played a lot of that game haha. Thanks again and have a great weekend!
In short, no. Intel "Aboveboard" 286 and 386 ISA cards were the favored corporate upgrade, and the best strategy for clone vendors was to sell entire motherboards as upgrades, or entirely new computers. CPUs were quite pricey compared to other components. This niche product's only survival strategy was to appeal to as wide a variety of clone owners as possible. They mostly failed as a commercial product anyways. During this time it was unclear how planned obsolescence might work in the PC marketplace. Theoretically, maybe, some company might have tried a one-board upgrade, maybe for IBM or Compaq motherboards, but it would have been a remarkably foolish business decision, and most likely they would have been branded quite clearly, if that were the case. Seems highly unlikely.
alright :) now lets do this the otherway around ;) squeeze a 286 cpu into a 386 board :)
i totally expected that 386 being used through the cache improved adaptor.
glad you are back :)
Unreal Redux rules!
@@PROSTO4Tabal indeed it does :)
omg that new intro is fakin wonderfull
Very cool parts! Would be really interesting to see that cache thing running. But nice to see, that my guess about the 386sx yesterday in the pre-announcement wasn't wrong ;) This were similar results, as what I got in the past. 286 is always slightly faster at the same frequency, then a 386sx. However, with 386sx you get huge benefit in Win3.11, since 386 has better support for protected mode and even more important VM86, which allowed to run more than one real mode program in windows in parallel. With 286, windows had to be closed during the execution of a real mode applications and was restarted again, when this application was finished.
Thank you very much. I'll ask around, if someone knows anything about that cached 286 thing.....
yeah, I had to lough yesterday when I saw your guess about the outcome. 😉
Yea I had customer complain about that. Oh it is slower. Well yes, but it now runs windows 3.1, 95, xp. Well you get the idea.
maybe those cypress memories are not a cache, but, just some faster direct mapped memory, probably used by special sw that put critical code and data that needed to run as fast as possible.
I skipped from an 8088 to a 386, so I never encountered anything like this. Wow! That's so interesting.. i was cringing when you put the AMD chip in knowing how they have to be carefully removed (i guess i broke a socket like that in the past)..
As i suspected, 386SX brings nothing performance-wise. As for that cache board - if RAM runs at 12Mhz with 0 wait states, than cache can not possibly do anything more. The only reason why cache is installed is that it runs at higher speed and lower latency than system RAM. If RAM already runs at CPU speed with zero latency, there is nothing you could do to speed the transfer up with caching.
If that was the case here though you'd expect performance would actually slow down a bit with the cache due to cache miss overheads. Pretty likely the cache is just not activated at all, although I suppose it is possible the cache controller tests ram throughput on boot and deactivates itself on purpose if the RAM is fast enough already?
@@JeremyLevi Yeah, good point.
They really sold some strange parts back in the days. I started with a 286 at 12mhz, then I asked for an upgrade, and they told me I need a different mainboard for the 386, but they have a cheap one that works perfectly for me (I was a student back then) and they take the other parts in commission for me. I trusted them and was not disappointed. After upgrading to an 386 with mindblowingly 33mhz, it ran like warp speed for me. And the best thing: The mainboard was somehow compatible to the 486 which I put in 2 years later (1995) and sold the 386 again. Now I had a 486DX2 80mhz and again, 2 years later I upgraded to a DX4 from AMD with 120mhz and overclocked it to 133mhz, so my 3DFX Voodoo card had something to do :D
Such good times.
Hmm, I wonder if a 486SLC would boot if it was transplanted to that 386 adapter.
this would be indeed an interesting experiment 👍🏻
@@CPUGalaxy That I would like to see! :)
@@CPUGalaxy Please try this, but I suspect it won't give any additional benefit.
I have the feeling that the answer is yes. The 486SLC CPUs could be paired with 387SX and 287 FPUs (depending on the motherboard) and their buses had the same widths as those of the 386SX.
Then put it in a 8088 to 286 adapter card for XT computers for maximum upgrade lmao
Very nice ! For the cache adapter, maybe it's possible to retro-engineer it using software in some way ? Don't quote me on that but maybe you can use a software to "sniff" the ports on the bus and therefore find those used by the adapter board ? (then you'll have to try many things to figure it out but anyways).
Welcome back.
That is a really nice motherboard. Plus cache if king too. I remember purchasing a p60 on clear out and loading up the board with either 1 meg of catch or 512. Anyway what a little speed demon. In some area it out performed a p75 and the horrible cache on a stick.
Amaizing stuff
The Intel 80286 was an amazing chip.
I have a similar little 386 board installed into the 286 socket on my IBM 5170. It's called the "Intel SnapIn 386" but mine has a 2.5x clock multiplier and 16kb cache right on the upgrade board. The multiplier is always active but the cache gets enabled by software - and even with the 20MHz clock speed, most benchmarks show slightly less performance than I got with the original 8MHz 286 if the cache is not enabled! Really goes to show how badly the 386 CPU is hampered by the slow memory pathway on this board (main RAM tests at at roughly 2MB/sec whereas the 16kb cache does about 10MB/sec). I wish I could mod it to have, like, 1MB of cache onboard!!
At first glance I thought that 386SX upgrade might have been a Kingston SX Now, but after a quick google it looks much simpler, which is probably why it was no benefit.
One of my IBM Model 50s came with a 386SX-20 upgrade boardr, which is also a lot more complex, including a few larger support chips and an 8K SRAM. Although it is a single 8-bit SRAM chip, so it would take a minimum of two clock cycles to retrieve any data, since the 386SX and 286 uses a 16-bit data bus. Unless it's doing some weird clock multiplying, but I feel like that would be way too complex for such a small upgrade board.
Yeah - at 6:34 your PLCC socket is cracked. The sockets weren't designed for many insertions/extractions - I cracked the sockets on a couple of PS/2 Model 50Z planars testing the Kingston "386 Now!" upgrades.
happens often when you use a lever against the socket to remove the chip, the dedicated extractor usually prevents cracking them
Would love to see if you can track down one of ImproveIt Technology's "Make-It 486" chips that supposedly upgrades a lowly 286.
Seeing Win95 running on a 286 board would be pretty damn interesting. :)
I'm guessing at 12MHz the ram is running at system speed with no wait state? Hence the cache would have no effect and might have been working all along.
Exactly
Thats a very good point. I need to check that. but i guess the RAM is running on a waitstate. on another 10 Mhz 286 setup i have i get better results with the benchmarks. there i know the jumper setting to set the waitstate to 0. the board in this video here does not have a jumper for that. 🤔
I'm pretty sure the board in the video is running at zero wait
@@CPUGalaxy The IBM AT runs at one waitstate. If your 286 at 12MHz performs "like an 18MHz AT" (quoting Landmark), this shows quite clearly that your board uses less waitstates than the AT, which means 0WS.
The 286-10 that is faster likely was faster in a video-heavy benchmark like the 3D bench you showed. I expect the 10MHz 286 board to just overclock the ISA bus to 10MHz, whereas 12MHz 286 boards often used additional waitstates or dynamic clock spreading to limit ISA speed for compatibility. Furthermore, that Trident 8800 graphics card is not exactly known to be on the fast side of ISA cards. Trident got good performance with the 8900CL, peaking at the 8900D (which is claimed to max out the ISA bus like an ET4000).
What about putting the 386 to 486 clip-on on top of the 386 adaptor? It will probably not work, but hey, it's to exciting not to try!
That is one amazing chip sandwich.
Can u put the 386 board into the cache board?
Ja!
I got one of these in an NEC Mate 1 PC I saved while dumpster diving. I still haven't had the time to restore it yet.
The one that has the Intel 386SX looks like a Kingston brand. I had a similar one but wasn't the PLC socket type.
The 386 upgrade probably came from Improve Technologies or a company like them. In the 90s they made a lot of those.
I feel like the 386 adaptor was made for upgrading some machine tool equipment. Maybe some upgrade offered by a company to run their newer software with just a plug and play hardware upgrade. I don't know honestly. You just get these feelings when working on those old types of machinery. Such as getting a 486 based press brake running in 2021 and seeing the mixture of upgrades and repairs that have happened over its life.
yeah, this could be a possibility and to be honest I had the same thoughts.
On the 286 + cache adapter, there's a TSR program that's required to activate the cache.
and exactly this tsr i need.
286 to 386:I'm a cybernetic organisms. Living tissue over a metal endoskeleton 💪
Pretty sure someboby wrote it already in the comments, but oh well, combine the 386 upgrade modul with the cach module and run your software to test if this cache is even recognized.
After seeing the video, think you missed some jumpers to increase the clock accordingly to the processor. Also, the big justification to use 386 instead of 286, is the 386 can run workgroups and is network compatible who is not with the 286. I still use a 386DX40 with a stepper isa board to control an OZO PCB prototyping router/etcher/driller/ for manufacturing Motherboards in 1987. I have a 286 Zenith working but not network. The 386 is a slave of a win7 to transfer the Gcode to control the OZO (jot) unit.
I have one of these. It sits snugly in my PC9801VX. I actually plan to replace the 386sx16 CPU (Clocked at 20Mhz) to a 486sxl2 CPU at 40Mhz.
Maybe try popping the 386 upgrade on that cache board. And isn't there some flags you have to change for other cache enabled cpus sometimes. Kinda like what they cyrix utilities do for those chips?
Very cool video! It's not surprising that the 386sx performs similar/worse than the 286, the bus drivers would slow it's ability to interact with the bus slightly. Unlike other steps in CPU generation the 386 actually has a similar IPC to the 286, with some instructions being faster on the 286 and vice versa. For 16bit software in general 386 and 286 systems of the same clock speed are generally roughly equivalent. Even proper 32bit 386 machines are similar in IPC, the main speed difference being they can fetch instructions faster. What differentiates most systems of this period is actually the mainboard, memory and graphic card. A chipset with cache, faster RAM, or a faster graphic card makes a far greater performance difference than the CPU given the same CPU clock speed. The only way to make the best use of these upgrades is to have the software for the cache, or slightly increase the clock speed on the board for the 386 (it can do 16Mhz).
Both of these adapters are things I've been trying to buy for a long time! Super super rare stuff.
The cache is probably working fine, but is mostly there to compensate for slow 286 bus access to very slow DRAMs at probably 150ns. Might work better if not for using faster modern memory. I'd expect a tiny boost despite that, though, so maybe there is some software involved. The 386 will have worse performance per clock because of it's strangled bus, which means to do the same amount of work it must also take twice as many memory fetch cycles.
Disappointed you didn't combine the 386SX with the cache shim! It has a higher chance of actually having a benefit.
Can you find any name or number on that cache board. The number that can be seen is not enlightening, seems common to many PCBs
And secondly, the 286 was always rumoured to be a better DOS CPU than the 386sx - seems to be true
Can you slow a "real" 386sx system to 12MHz and get the same result, just to prove it's not the adapter?
I guess the only reason would be some Netware or UNIX that needed a 386 for memory management. Either way, intresting video. Would also like to see the "386 on the cache stack" and/or a crystal replacent to get it up to 16Mhz.
Microsoft had been toying with Xenix, or I should say they actually put out Xenix for 286 as a commercial product, but changed course and decided to put their efforts into the Windows GUI. In no small part thanks to the success of the Mac. MS could avoid competing with established Unix vendors, while encouraging the clone makers to stick with them, and weaken IBM's position in their partnership. It was "Game of Thrones" level intrigue!
From what I remember, in practical terms the 386SX was nothing more than a 286 that could run 386 code and its main advantage was being able to run at higher clock speeds (than 286). For most hobbyists and enthusiasts, it was actually cheaper to just replace the motherboard/CPU. Those CPU upgrades only made (some) sense for proprietary/business systems.
The 16/32 thing on the SX was, allegedly, so that OEM's could use as much of their existing PCB layouts as possible. So it's kind of weird these little bolt-on modules weren't a more common sight even if just soldered straight to the mobo from the factory.
The socket sandwich is pretty ridiculous to see, too bad the cache upgrade did not engage.
I'll hope you'll get more info on these thing, it's pretty intriguing stuff.
Just started watching. I'm *thinking* the 386SX plugged into a 286 board won't really be any faster on 16-bit code; except that a 386SX can clock higher than a 286 if both are OC'd. I look forward to seeing the results :)
I am no expert on this at all, but just really feels you need some extra drivers to get the extra performance?
Glad you're back btw! Hopefully you're all doing well! :)
Could you try making some DOS tests with a Transmeta Crusoe with adaptative instructions?
I really appreciate your channel!
try the 386 in de cache adaptor...
Ouch, thats a tower of power!
Adds new meaning to the term "CPU stack"
IIRC, the cache in these upgrade chips sometimes required a BIOS update to enable the cache. Which, as it turned out, was not always available for specific motherboards. There were compatibility issues.
The only real advantage to the 286-to-386 thing was it gave you 386 enhanced mode for Windows, which was not available for 286-class machines. As for the memory limitation, there were peripheral cards that allowed more addressable memory. Four- and eight-meg cards were easy enough to find (but expen$ive!) So, in many cases, it was often more cost-effective (long-term) to spring for a new machine than to 'upgrade' what you had.
It was known that 286 was clock for clock faster than 386sx. That's why 286 with Harris 20 or even 25 Mhz PC still sold long after introduction of 386.
As others have mentioned the 186 and 286 execute instructions faster clock for clock.. The 386 advantages were in 32bit instruction set which hardly anything used at the time. A flat virtual memory addressing mode used by windows. The next intel processor with this slower next gen. The 486 is faster than the first Pentiums.. Pentiums had far higher clocked versions. Another fun fact about the early 486s is that for some applications it ran faster on un-cached motherboards because the super fast 486 burst mode 128 bit memory read. If your program read through data linearly the 486 could read effectively at 1/4 wait state. So for example if you are calculating inner products to multiply matrices you could exploit this quirk. Later higher clocked 486s always preferred a cache.
What crack are you smoking? The 486 was NOT faster than the first Pentiums. Not just in Quake. Even in Doom a 60 MHz Pentium performs better than a 486 DX4-100.
I have an Atari St with an add on 286 processor. I'm hoping to find an upgrade to a 386 or 486 that fits the CPU socket.
Cyrix has a utility to active Cache in the operating system. May be worth a try. You'll also need to upgrade the crystal. While you would be over clocking the ISA slots, you might be able to get away with it, since most of those slots will work at a faster speed. I've heard the a clock for clock cycle between the 286 and 386, the 286 is actually faster. You have to get a fast clock speed to gain the benefit. Hope that helps.
there were some pretty good snap on 286 upgrades (less of a gimmick) made by IO DATA for PC9800 - by the nature of the architecture, most are compatible.
try the 386 for 286 with the cache for 286..maybe the 386 sees the cache..worth a try
I really hope to see the cash working
Lets put 386 OverDrive into 386 to 286 adapter, so we run 468 while running like 386 on 286 board!
The CPU tower or Power!
It would be interresting to reverse engineer the 286->386sx adaptor... Also, if it were possible to put in an SLC2 chip, it would have clock doubled and have some cache.
You could theoretically put a 486 chip to this adapter as it exists already. Cyrix made a clip on version of the SLC2 upgrade, Cx486SRx2 in 16/33, 20/40, and 25/50MHz versions.
I used to like playing HEXX and supaplex on an old system.
Your 286 socket cracked. On the corner closest to the camera when you were extracting the AMD 286. Near Pin 51.
is it possible to stack cache upgrades to cache upgrades?
Have you tried stacking the 286 cache pass through board, with a 386 CPU to 286 M/B adapter?
Even if a 386 upgrade is slow, it's still better than adding cache to a 286. They both make the system faster, but cache doesn't allow you to run 386 software, so the extra functionality makes it worth it. If you don't plan to use your system to run 386 software, like, if you have a 386 system for that already, then go with the cache.
Could you change the motherboard oscillator to increase the mhz to 16?
There are some 486 upgrades that clip on top of the 386's (saw one from Cyrix some time ago, I think it disables the original chip by grounding one of the pins), maybe you could test it 👌👌
It's the '-FLOAT' pin of the 386SX CPU that disables it - which wasn't set up well initially by Intel, and usually required to 386SX stepping to be the 'B' or 'C' (as marked for the CPU in the video). Clip-over upgrades are I little more rare now - I've just picked up six TI CPUs that I will test soon.
@@IBM_Museum that would be great to see, a "486" in a 286 motherboard 😃
@@BaguetesGarage: IBM made a module that mated to a 286 socket that had a 486SLC2-50 (able to run all Intel 486SX instructions, with a 16Kb L1 cache) - and although it is proprietary to not be a conventional CPU socket, they also had the fastest systems with a 16-bit data bus / 24-bit address bus CPU that was produced: the 9556/9557 PS/2 with a 486SLC3-75.
@@IBM_Museum triple-clocked hey, interesting. There was some weird stuff going on at the time 😅
@@BaguetesGarage businesses weren't sure the public would be willing to throw away their old PCs for new ones. Windows 3 then moreso Win95 proved they would. Intel Aboveboard 286 and 386 ISA cards were expensive upgrades and most consumers decided just a little more money for an entirely new computer (or at least a new motherboard) was well worth it. A little later IBM arrogantly gambled with Micro-Channel Architecture that customers would abandon clones entirely to get the latest IBM-brand technology. "Open standards" as championed by Compaq (not Open Source) and changed not only PCs, but the entire tech industry.
I have similar 286 upgrade wirh 3pcs of CY cache chips and a controller: ALL Technologies Chargecard II. It can be used for memory management like EMS with multitasking options, isnt a cache upgrade unfortunately...