PDA

View Full Version : Stacked memory chips on the PC AT ...



mbbrutman
March 14th, 2004, 09:06 AM
My 1984 IBM PC AT has 512K on the motherboard. I think all of the ATs have 512K on the motherboard, but the chip setup is kind of odd - 4 banks of 128K modules. And the modules look like two memory chips stacked on top of each other with their legs soldered together.

There had to be a trick that made this work. If both memory chips in a stack were exactly the same, they would both try to respond to requests at the same time. They have to be different. My bet is that they are both 64K chips, and one is just slightly different than the other in terms of addressing. Does anybody know exactly what the addressing difference is?


Mike

Super-Slasher
March 14th, 2004, 10:48 AM
You're correct in guessing that they are 64K chips. Only 64K chips were "piggybacked" on the Type I models of the PC AT motherboard, and here's why:

The first models of the PC AT only came with 256KB of RAM, with the 4 banks of 8 single 64K chips. Soon after the debut of the PC AT (I think it was only a few months; PC AT's with 256KB of RAM are seldom to find, apparently), IBM decided to offer 512KB of RAM on the PC AT. IBM originally wanted to use the 128K chips available, but they simply just cost too much at the time; the extra costs would pass onto the consumer - they did however already have a surplus of inexpensive 64K chips, so they simply piggybacked the chips to double the RAM. When the PC AT came out with the Type II motherboard (8MHz models), you'll only find the 2 banks of 8 256K chips.

As for the addressing, I don't think RAM chips back then were too complex in design. To me, the PC AT seems to simply recognize that there is more RAM. I mean, the chips are soldered together exactly the same, so each two chips would have to logically share the same memory addresses. The PC AT probably sees the two 64K chips as a single 128K chip.

:D

mbbrutman
March 14th, 2004, 11:47 AM
I know you are an AT fanatic, but I think you are off on a few points:

According to my tech ref, type 1 boards have four banks of sockets, each of which has 9 128K chips in it. Type 2 boards have 2 banks, each with 9 256K chips in it.

Also, the minimum configuration shipped from IBM as 512K. If you wanted a machine with less memory, you bought a PC or an XT. AT was the flagship.

I have never heard of a 128Kx1 DRAM chip, and I don't think they exist, so IBM certainly was not planning on using them. Standard chip sizes are 4Kx1, 16Kx1, 64Kx1, 256Kx1 .. notice the size goes up by a power of 4 each time.In a type 2 motherboard the 256Kx1 chips would work fine, but a type 1 motherboard has to use these weird stacked modules because 128Kx1 chips don't exist.

If you look at the datasheet for a memory chip, you will see that you just can't solder two identical chips together - they'll respond to the same signals at the same time, creating a mess. The stacked chips on the type 1 motherboards *have* to be different from each other, otherwise it can't work ....

Super-Slasher
March 14th, 2004, 02:33 PM
Ah, you were right about there not being 128K chips, but I am most certain that the original PC AT's in '84 for a short period only came with 256KB of RAM, hence the 4 banks of 9 (corrected) 64K chips.

A simple way to see if both chips are the same or not is to unsolder one bank and see the numbers on the top of each chip, or if you even came across a dead Type I PC AT mobo/spare piggybacked chip set and cut them apart to see.

I volunteer your PC AT to be our test subject... *walks towards it with snips* :o

mbbrutman
March 14th, 2004, 02:49 PM
Hides ATs behind a stack of PCjrs ...

Terry Yager
March 14th, 2004, 04:09 PM
Ah, you were right about there not being 128K chips, but I am most certain that the original PC AT's in '84 for a short period only came with 256KB of RAM, hence the 4 banks of 9 (corrected) 64K chips.

A simple way to see if both chips are the same or not is to unsolder one bank and see the numbers on the top of each chip, or if you even came across a dead Type I PC AT mobo/spare piggybacked chip set and cut them apart to see.

I volunteer your PC AT to be our test subject... *walks towards it with snips* :o

I have a handful of those piggy-back chips in the top tray of my toolbox. I'd be happy to send 'em to ya if ya wanta experiment with them. BTW, I don't recall the exact details, but IBM did have to do something with the chip select pins on the stacked ram chips or they would both be selected at the same time. I just don't remember exactly how they disabled the pin...

--T

mbbrutman
March 14th, 2004, 04:19 PM
Terry,

If you don't mind, desolder or clip a stack. I think all of the pins from the two chips are connected to the corresponding pins, so it must be a difference in the chips.

I don't have spares, and I don't want to sacrifice a working system to find out ...

BTW, all of my ATs seem to be type 1 motherboards. Is the type 2 motherboard the one used in the 8Mhz systems?


Mike

Terry Yager
March 14th, 2004, 04:42 PM
Intel, NEC and others are making "stacked ram" chips today. The difference is that now, instead of soldering two DIPPs together they stack two wafers together within the package, invisible to the naked eye. This yeilds a higher capacity ram while utilizing cheaper low-density wafers. A 128Mb chip can be produced from two 64Mb wafers for far less than the cost of producing a single 128Mb wafer. (I just googled "stacked ram" and dug up this info. Try it, there's lots of interesting stuff to be found in google-land).

--T

Terry Yager
March 14th, 2004, 06:30 PM
Terry,

If you don't mind, desolder or clip a stack. I think all of the pins from the two chips are connected to the corresponding pins, so it must be a difference in the chips.

I don't have spares, and I don't want to sacrifice a working system to find out ...

BTW, all of my ATs seem to be type 1 motherboards. Is the type 2 motherboard the one used in the 8Mhz systems?


Mike

I dunno, I've never had a TrueBlue 8MHz AT before, although I did once have a 6MHz model equipped with an aftermarket accelerator board which kicked it all the way up to a blindingly fast 10MHz (selectable with a switch mounted externally on the back).

Found on the web (from a PC magazine article):



IBM's introduction of the PC AT was a different story. Based on the Intel 80286 processor, the PC AT cost nearly $4,000 and included 256K of RAM but no hard drive or monitor. Models with a 20MB hard drive sold for $6,000.


To view the article in it's entirety:

http://www.pcmag.com/print_article/0,3048,a=22650,00.asp

--T

Terry Yager
March 14th, 2004, 06:57 PM
Ok, so I sacrificed one of my spares. The top-side of the top sipp has (besides a TI logo):

ZA1250NL
AP8510

...and the underside of the top chip simply reads:

LA

...but the underside of the bottom chip reads:

LA/02406H13
SINGAPORE

However, it's interesting to note that the top-side of the bottom chip has four little bumps on it, as if to prevent the two chips from coming into intimate contact (to allow for airflow?), so it's apparent that at least one of the packages is custom-designed for just this purpose. (I think that would be necessary to get the whole setup to work, since some internal modifications would surely be called for, as no external mods are present).

--T

mbbrutman
March 14th, 2004, 07:44 PM
It's a worthy sacrifice indeed ..

My motherboard has OKI chips on it! M37S64A, 150ns. I assume they are compatible with the TI ZA1250NL that you found. I couldn't find data sheets on either one.

I think I found the trick:

The two chips are 64K chips. One of the chips moved two pins around so that the motherboard could select on or the other chip without having them both respond at the same time. The chip itself is the same, just the DIP packaging of one of the chips is slightly modified. (The lines used may have been the RAS lines.)

The difference in the markings on the underside of the chip might have been the code to tell what kind of chip it was. Most of this stuff looks lost to history though ...

Terry - don't dissect anymore. I may need them one day. :-)

SuperSlasher - I owe you an apology. Although rare, 256K ATs did exist. The motherboard is clearly marked for it.

Terry Yager
March 14th, 2004, 08:10 PM
It's a worthy sacrifice indeed ..

My motherboard has OKI chips on it! M37S64A, 150ns. I assume they are compatible with the TI ZA1250NL that you found. I couldn't find data sheets on either one.

I think I found the trick:

The two chips are 64K chips. One of the chips moved two pins around so that the motherboard could select on or the other chip without having them both respond at the same time. The chip itself is the same, just the DIP packaging of one of the chips is slightly modified. (The lines used may have been the RAS lines.)

Exactly! (See the link below with pinouts). There are actually two different RAS lines, RAS0 and RAS1


The difference in the markings on the underside of the chip might have been the code to tell what kind of chip it was. Most of this stuff looks lost to history though ...

Terry - don't dissect anymore. I may need them one day. :-)

SuperSlasher - I owe you an apology. Although rare, 256K ATs did exist. The motherboard is clearly marked for it.

Here's another article with the pinouts:

http://www.classiccmp.org/pipermail/cctalk/2002-June/002847.html

The chip I disected was probably bad anyways...the pins were badly bent (from kicking around in my toolbox all these years).

--T

CP/M User
March 14th, 2004, 08:11 PM
"Terry Yager" wrote:

> I dunno, I've never had a TrueBlue 8MHz AT before,
> although I did once have a 6MHz model equipped
> with an aftermarket accelerator board which kicked
> it all the way up to a blindingly fast "10MHz"
> (selectable with a switch mounted externally on the
> back).

Don't know how reliable your machine was, but I
always considered the Turbo switch on my XT as
a health thread to it.

Just sounds like kicking an extra 4Mhz onto a 6Mhz
machine is a tad risky - after all, you're nearly
doubling the speed of this machine.

Cheers,
CP/M User.

Terry Yager
March 14th, 2004, 08:31 PM
"Terry Yager" wrote:

> I dunno, I've never had a TrueBlue 8MHz AT before,
> although I did once have a 6MHz model equipped
> with an aftermarket accelerator board which kicked
> it all the way up to a blindingly fast "10MHz"
> (selectable with a switch mounted externally on the
> back).

Don't know how reliable your machine was, but I
always considered the Turbo switch on my XT as
a health thread to it.

As I recall, the switch was set to 6MHz most of the time because of that issue--reliability. But it was not that the hardware was unreliable, it was the (older) software.


Just sounds like kicking an extra 4Mhz onto a 6Mhz
machine is a tad risky - after all, you're nearly
doubling the speed of this machine.

Cheers,
CP/M User.

The accelerator board had a 10MHz 286 mounted on it, and a ribbon cable with an adaptor on the end to jack-in to the 286 socket on the mobo, so the buss was still running at 6MHz, while the chip ran faster. The switch was necessary for compatability back then because many programs would just choke on the extra speed. (Some higher-end accelerator boards at the time did double the speed, all the way up to 12MHz in a 6MHz system).

--T

Super-Slasher
March 15th, 2004, 06:50 AM
SuperSlasher - I owe you an apology. Although rare, 256K ATs did exist. The motherboard is clearly marked for it.

Hehe, no apology needed! Thank you, on the other hand, for teaching me a few things about RAM chips... :D

CP/M User
March 15th, 2004, 11:27 AM
"Terry Yager" wrote:

> As I recall, the switch was set to 6MHz most of the
> time because of that issue--reliability. But it was
> not that the hardware was unreliable, it was the
> (older) software.

On my XT it certainally was a hardware issue & I
seem to believe that it might of been the fault of
the Hard Disk (either that or the User decided to
trash the thing).

Anyway, when the system needed to be fixed
the system restorer also had the thing on Turbo
& they couldn't install DOS onto it, unil they
switched the stupid thing off. I would have
thought that was a hardware issue (if the Hard
Disk didn't want to be written to).

Cheers,
CP/M User.

barryp
March 15th, 2004, 03:23 PM
My motherboard has OKI chips on it! M37S64A, 150ns.

They are pin-compatible with standard 4164s. I have a gazillion of them too.

Super-Slasher
March 15th, 2004, 03:59 PM
What type of chips were in the 256K AT's then, curiously; rather, which one of the two chips that were used in the piggybacked configuration in the Type I mobo's with 512KB were used, if at all?

mbbrutman
March 15th, 2004, 04:37 PM
The motherboard of a type 1 machine can only use the weird 128K modules, so a 256K machine just uses one of the two available banks. (Each bank is 18 chips/moodules.)

The type 2 motherboard used 256Kx1 DRAMs, which is more compact and a simpler design.

If you think about the type 1 motherboard, it has to hit each of the piggy-backed chips individually. It's kind of like having two banks of memory in one socket. This makes the memory address decoding logic a little more complex than if 256Kx1 DRAMs had been used. Which is what makes the type 1 special. :-)

mbbrutman
March 15th, 2004, 04:45 PM
People who's technical opinion I respect think the AT was grossly over engineered. In typical IBM fashion:

- The case was even sturdier than the PC and XT case.

- The power supply is a very conservative 200watts. I've seen demonstrations where people pull the cord out of the wall and plug it in really quickly (within a second or so), and the machine doesn't hiccup.

- The timings on the chips are very conservative. Many of the 6Mhz ATs were overclocked to 8Mhz with no problems.


The only thing that IBM screwed up on the AT was using CMI hard drives in the first batches, and we know what happened to that company. :-)

Super-Slasher
March 15th, 2004, 04:53 PM
Yeah, the AT case is a brute. Could probably take an attack from a sledgehammer with just a few scuffs and scratches, MS-DOS still running fine on the inside, hehe.

carlsson
March 16th, 2004, 05:14 AM
Who will invest in some birdshots (or buckshots) and see how much an IBM AT case can take? As I've posted before, similar experiements have been done with XTerms and Sun 3-series. I'm sure IBM would approve it, just to brag about how much better their solutions are compared to Sun.