My 486 Build

bestjunky

New member
Joined
Oct 7, 2020
Posts
22
Country
Canada
Region
BC
My first 486 from a few years ago. Thanks to parts found on AmiBay I'm going to give that machine an upgrade, to bring it closer to Pentium speed.



Have one overdrive and one DX5 133MHz chips in the mail that can potentially run with the motherboard, but won't know until they get here. Want to benchmark them against the current DX2-66.

Back in the 90's didn't have a 486, so I didn't know much about their motherboards. I did accumulate a few CPU's. Then when I got a 5V socket 3 I tried some CPU's and probably fried my 3.3V AMD's in the process. I did not know that DX4 series had a different voltage. Potential casualties are 2x 100MHz and 1x 120MHz AMD chips. Though there was no abnormal behavior. No chip overheating. No smoke. So maybe the CPU's or the board have protection against incompatible hardware. Either that, or they fried so silently, that I didn't even notice.

Does anyone know if inserting 3.3-3.45V CPU into a 5V socket kills it or if power is delivered via different pins, offering some kind of idiot protection. My board behaves as if it recognizes the wrong CPU type and refuses to power on entirely. Maybe it's a good sign. On DX2-66 and slower chips it runs fine.
 
Last edited:
I already bought 2 from him. That's the whole reason I ended up on this forum. I hope I get to come across original Gainbery some day, or maybe replace regulator on the current one and then a miracle... I don't know.
I spotted few solder bridges on the CPU. Thought they were in error. Tried fixing them and fried the regulator in the process. I desoldered it and although I'd expect the input to read 5V, it's closer to 3.3V, so I assume there's more damage than meets the eye. Could've easily taken the motherboard with it, but tests indicate that everything else is running fine.

And as an insult to injury, I found high res photos of somebody else's Gainbery and they had 2 solder bridges where I had mine. They weren't in error. So I was fixing what wasn't broken to begin with. Though then conclusion is that it was a bad chip from the moment I got it. Losing Mototola J2955, which I assume is what drops the core voltage - the one with the largest heatsink, that's problematic. When I measure voltage on an empty pad I get ground where I expect +5V and +3.3V on the ground pin. So either J2955 is not what I think it is, or something else got fused. To fix it for sure I'd have to get an SMD DX5 chip, that exist on ebay on sawed off pieces of motherboard. Transplant that and get a 2955 somewhere. I really don't want to give up on that Gainbery, since I've only seen a single one show up on ebay in 2 years.

Closest thing to that original chip is Evergreen 486 upgrade. That one may even have more cache, but they're rare and expensive. A complete DX2-66 system costs less than a single Evergreen.

Mine is alright on 66, but just a tad slow running games that I remember playing on P100. Since I don't have much of a choice when it comes to videocards, I have one S3 and one Trident that went into the build with slightly better benchmarks. I can't upgrade video any further. Ram is maxed out at 64MB, cache at 256, replaced with 15ns timing chips instead of 17ns originals. I even got a new clock chip that can run the bus at 50MHz, but I doubt the rest of components would keep up. VLB only runs 1 port at 50MHz bus, so that's a no go for me.

With a socket adapter I'll run through my DX4 chips and find out once and for all if they're dead or not, and will stick with whatever ends up being fastest.
 
Last edited:
the 486s 'era' was perhaps one of the most 'complicated' times in building PC systems, for a number of reasons, not the least of which is that they probably went through more 'revisions' of technology involved with PCs during it's life-cycle than any other generation.

486 boards saw 3 different generations of sockets, Socket 1 (PGA-169, 5v, 16-33mhz fsb), Socket 2 (PGA-238, 5v, 25-50mhz fsb), and Socket 3 (PPGA-237, 3.3 and 5v, 25-50mhz fsb)

afaik, each will take a PGA-168 (effectively all 486 'class' cpus), but only the Socket 3 boards support the later DX4 and AMD 5x86 chips (and also the Cyrix Cx5x86) without a voltage regulator. (and some (most?) earlier boards may not work with them at all)

in addition to the variants of CPUs that came out to fill them, SX, DX, SX2, DX2, SLC, DLC, DX4, Cx5x86, 486 Overdrive, Am5x86... (also Pentium Overdrive, which required the additional pins of Socket 2 and 3)

they also saw 3 separate generations of expansion ports, starting out with only ISA ports, adding VLB slots (from 1992, superseded by PCI by 1993), and PCI prior to the end of it's life-cycle. (of course, not even getting into MCA and EISA...)

a couple of things should be noted about VLB here, one, it was reliant upon the 486s memory bus design, and as such will only work on a 486 system, OR required a VLB to PCI bridge (such as the OPTi 82C822), to work on a few early Pentium systems . (the Pentium being released about 3 months before PCI)

and secondly, 50mhz is 'out of spec' for VLB, which was intended to run between 25 and 40mhz (this was probably only an issue on 486DX50 systems) , so many cards will either not work, or wont work correctly, VLB was notoriously finicky... (also, for compatibilitys sake the ISA slots typically still operate at 7.15mhz, and up to 14mhz on some boards, though most ISA cards will not operate at that speed, I typically found that 11mhz was the absolute fastest a lot of ISA cards would go, older cards wanting the 'standard' 7.15mhz or perhaps 8mhz at best)

with the addition of PCI, most boards saw the integration of IDE ports, FDD ports, Parallel and Serial ports generally, whereas they previously varied from system to system. most ISA and VLB systems wanted a Multi-IO card, this was far less the case on PCI equipped boards. (though, certainly not unheard of)

RAM also went through 3 different generations in the 486s lifecycle, with 30-pin SIMM RAM being the standard when the 486 came out, being superseded by 72-pin SIMMs during it's time, and SDRAM coming out very late in the 486s existence. (SDR being introduced in 1992, but not really catching on in a larger way until around '97, which saw the release of the Pentium II, whereby it became the standard)

*one thing I forgot was that EDO was added to RAM quite late in the 486 period (about 1994), with a few chipsets supporting it (ie: ALi) which was about 30% faster than standard FPM RAM.

by comparison I found 386s, 286s and even 'XT' (8088/8086) systems far easier to work with (albeit more limited in what you could do)(not that 486s were all -that- complicated, as such... but sourcing parts, and getting things to work trouble free together was sometimes an issue, particularly on VLB systems, but a lot of issues were resolved by the time PCI rolled around)

if I've missed anything here, or forgotten... I am getting old... :LOL: (I'm sure I've built a few hundred 486s, I own 3 (2x DX2-66, 1x OD66) and a Samsung 486slc25 laptop) the original Pentium and prior was definitely when I've had the most fun building systems, they've slowly become more like 'appliances' since...
 
Last edited:
@SaviorX, if we could rate posts on AmiBay that would get a 10/10 from me (y)
 
@SaviorX, if we could rate posts on AmiBay that would get a 10/10 from me (y)


thanks :LOL: I'm just glad I can still remember any of these things after 25+ years.

there were a couple other things that come to mind as well... such as LBA (28-bit) being introduced in the 486 'era' (1994), going from supporting only 528MB drives as a maximum to begin with, to 2GB and 8GB drives...

also, if memory serves, OSes didn't offer 32bit hardware support until around 1992 or 1993, at least not Microsofts offerings anyways, Windows for Workgroups 3.11 and the Win32s add-on for Windows 3.x

which was well after the 80486s release in 1989, and even longer after the 80386DX in 1985... (the best I can come up with is that OS/2 2.0, released in April 1992 was first...) (though I guess that was a bit more relevant when 32-bit expansion was incorporated via VLB (1992) and PCI (1993) and on-board devices, even though MCA had been around since 1987, and was 32-bit (and backwards compatible to 16-bit), it was not very successful)

side notes:
AMD was effectively 'made' by IBM requiring Intel to have a second-source supplier for x86 chips. AMDs 8088/8086 and Am286 are fully licensed x86 processors, Intel tried to end this agreement with the 386, but AMD won in legal challenges to allow them to continue making x86 chips, this is one of the reasons Intel dumped the 80x86 designation when they went to Pentiums. (and also why AMD continued with 5x86)

Cyrix began by making x87 compatible math-processors, some of which were superior to most 'licensed' x87 MCPros, and they started offering reverse-engineered 486 (and subsequently Pentium) compatible 486SLC/DLC chips though they were not always 100% compatible. (Intel sued Cyrix for patent infringement, but it was deemed that Cyrix had changed and improved upon Intels designs that they were non-infringing, and ultimately completely independent designs, Intel also paid $12M to settle anti-trust litigation by Cyrix as a result)

as most people will probably know, a 486SX is a 486DX where it's math-processor had failed tests and was switched off, and sold as a 'budget' processor (a few may have been made this way on purpose later, when a market for them had been firmly established), but a 487, made to go into a socket on a 486SX system was actually a full 486DX, and to use it the board simply turned off and no longer used the 486SX.
 
Last edited:
With the internet I got a chance to catch up on the 486 generation. In the 90's I went from a 286 straight to P-100, skipping everything inbetween, and Pentiums were indeed, easier in comparison. I don't miss the days when every component purchase was a dice roll and you didn't know if something was compatible or not, till you tried it. Days when it was possible to insert things upside down and sideways, unless you knew where pin 1 was and what the red stripe on a ribbon cable meant. Sometimes sockets were in upside down orientation and safety tabs had to be shaved off with a knife. Not very intuitive. VLB took me some time to figure out. I even bought 2 low profile videocards, because I wasn't sure that either of them would run issues-fee. Hard to troubleshoot when you have just one of something.

I went ahead and ordered an SMD CPU replacement and a voltage regulator for Gainbery. I hope by the time they get here I can find proper schematic explaining what solder bridges were doing on CPU pins.

https://www.vogons.org/download/file.php?id=78272&mode=view

That's not my photo, and his version is Cyrix, but if you look top right you'll see 2 bridges between the pins on the CPU, and those I assumed to be in error when I was fixing mine. The fact that he has them also indicates that they were probably intentional.

Appeal of Gainbery and Evergreen is that I heard they're stable on a 40MHz bus, so I could even run 160MHz, but I'd feel stupid overclocking anything retro. All it does is reduce lifespan. Maybe that's what the original owner did, and maybe that's why the CPU died.

I want to run this game at normal speed, which is my personal benchmark for retrogaming machine performance:
http://rememberpcdosgames.blogspot.com/2018/01/alien-odyssey-1995.html

It has DX2-66 as a system requirement, but it's practically unplayable on my machine that exceeds minimum spec. I could mess with jumpers on the videocard. Reduce VLB wait states in BIOS, or just sit tight and wait for the new CPU options to arrive and then see if it's the CPU or the graphics card. VLB card has one set of jumpers for >33MHz bus, but it doesn't specify what the ceiling is.

https://arvutimuuseum.ee/th99/v/U-Z/50814.htm

There is potential to tune that card a bit faster, but I haven't tried yet.
 
Last edited:
100MHz Overdrive chip did not work. I was hoping it would run under DX2-66 configuration, but it did not boot. I went though all possible CPU jumper settings, and there was no result. CPU was getting warm, as if running. I tried disabling half of my cache. That did nothing. I disabled all CPU cache in BIOS. That made no difference. There are other settings in BIOS that may make some difference, but my general belief is that the CPU is incompatible. It is from 1994 and my motherboard is from 1993.

Next I'll be testing AMD and Cyrix DX4 chips in about 2 weeks and then DX5 some time in December. Hopefully I can find some combination of gear that can lift this machine above 66MHz. I'm asking myself - what if the chip the board came with never worked with it to begin with? What if that CPU was fine all along and my board wasn't right for it. Guess I'll never know.
 
what if, y'know... you just got a Socket 3 board....?
 
For about $140 I could get some out of Ukraine with VLB slots in the same spot. I'll first wait for socket adapters to arrive and if that yields nothing, I'll consider swapping the motherboard for something faster.
Overdrive that I tested was old stock, factory sealed. I decided to overpay just to know for sure I wasn't getting a dead CPU. Now I know that overdrives don't work. But I know that Gainbery worked, which has an AMD cpu in it. I have a socket voltage adapter and a separate AMD DX5-133 coming that may work in its place. Or I also have a replacement AMD chip for the Gainbery itself, along with the voltage regulator that'll arrive in late december, so if all options fail, I can attempt to resolder the surface mount DX5. I know that's a 2 hours solder job right there. I can do it, but I have to find out first if I have to replicate bridges on the CPU legs, or if they were something redundant, like grounding unused data lines.

I had a scare today where I thought I fried the motherboard again, and turned out it was fine. I think I'm giving it too much love. It's not a good runner. Only 33.3 in 3D Bench. That's on a 2MB VLB card. That's just not right. Way too low. Feels like a crappy motherboard. I'd rather go for something with an Opti chipset.
 
134827.jpg143129.jpg

So far got 2 chips in the mail a lot sooner than I expected. First one the seller didn't secure it correctly and it ended up banging up all its legs, making SMD replacement on a Gainbery next to impossible. I am more likely to buy form a different source, than to take chances with this one, as it's not easy to test pins for connectivity, with traces mostly hiding beneath the chip. The The other CPU is essentially the same thing, but will require a socket adapter and a heatsink. I also found a Cyrix 100GP lying around. I heard it's as fast as the 133MHz AMD. Can't wait to test it. I also found one proper Opti chipset Socket 3 motherboard with VLB in the same position, but I'll wait on it till I'm done with Matsonic M601.
 
Last edited:
Back
Top Bottom