• Please review our updated Terms and Rules here

PDP-8/a parts needed

BitWiz

Experienced Member
Joined
Sep 7, 2021
Messages
419
Location
Glen Ellyn, Iil
I have recently acquired a PDP-8/A to go with my PDP-8/E.

Unfortunately it came with the industrial front panel and not the programmers panel.

Does any one have an extra PDP-8/A programmers panel they can spare?

Also, does any one have any PDP-8/A memory they can spare (or any Omnibus memory).

Or any other omnibus boards they have spare?

Thanks,
Mike
 
Hi Mike, I have some bare boards of Roland's 32KW memory board. This is just the memory, not the version with RAM plus boot loader.
 
Hi Mike, I have some bare boards of Roland's 32KW memory board. This is just the memory, not the version with RAM plus boot loader.
I have been using Roland's 32KW memory board for a while and it works 100% reliable all the time. It would work perfectly in a 8/a and we can make as many as we like.
 
I have one of Roland's boards in my PDP-8/E and it's works perfectly.

The problem is getting RAMs that work. While talking Vince about this he says that some 62256 RAMs just don't work right on this board.
 
Addressing the original posters question. 8/a boards are pretty common although they are becoming less so. There is a 16k core on EBay right now. I see CPU's, Option 1 and Option 2 boards sold frequently there. I have a lot of those cards that are in mostly working condition. What I don't have is any spare front panels. I think there is one of those on EBay but they want far more money than I am willing to pay.

As was mentioned, Vince did an 8/a front panel. I am sure he will say something about it. I saw a prototype when I was visiting a few months ago.

It has been reported that the 32k memory board does not work correctly with the KK8A CPU in the 8/a. This is the M8315 board found in many (most) 8/a units. The other option was the KK8E CPU which was installed when an EAE was desired. I have that arrangement in one of my 8/a machines. The differences are the KK8E CPU takes 5 slots (7 with the EAE) and the KK8A CPU takes one slot. The KK8E CPU is slightly faster then the KK8A (around 14%).

There are a number of design errors with the Lafferty design (this is mentioned on the original web page) which don't cause problems with the Cypress branded chips which are no longer available. The most notable issue is that for simplicity a tristate bus transceiver was used to interface to the Omnibus. Everything that drives the Omnibus is supposed to be open collector. There is a possibility that the memory board can drive the Omnibus to 5 volts while the CPU (or other bus master) is pulling the bus to ground causing a shoot thru current. While I don't think this ever happens with the KK8E CPU I think this is the problem with the KK8A CPU. What I do know is that there is shoot thru on the memory board itself when the memory chips are being read and the transceivers are in write mode. This is only a few nanoseconds long but it causes large currents in both the transceiver and the memory chips which can be seen by looking at the power supply pins on those devices. Another mistake is not buffering the address lines. DEC on all their semiconductor memory cards buffers the address lines with a Schmitt trigger gate. The reason it mostly works without it is that there are only the two CMOS inputs on the address lines and almost nobody has a maximum length omnibus. The other thing DEC always did on their semiconductor boards was latch the data coming out of the memory chips and then drive the bus with open collector devices.

In our testing of the Alliance memory chips we found that they do work about 99.9% of the time. Of course this is not nearly enough as there is the possibility of a misread a couple of times a second. We (me and Vince) were able to get it to run for several minutes of diagnostics before it would fail.
 
Addressing the original posters question. 8/a boards are pretty common although they are becoming less so. There is a 16k core on EBay right now. I see CPU's, Option 1 and Option 2 boards sold frequently there. I have a lot of those cards that are in mostly working condition. What I don't have is any spare front panels. I think there is one of those on EBay but they want far more money than I am willing to pay.
One is $250 (kinda reasonable) and one is $2500 (totally unreasonable)
As was mentioned, Vince did an 8/a front panel. I am sure he will say something about it. I saw a prototype when I was visiting a few months ago.
I will look into this.
It has been reported that the 32k memory board does not work correctly with the KK8A CPU in the 8/a. This is the M8315 board found in many (most) 8/a units. The other option was the KK8E CPU which was installed when an EAE was desired. I have that arrangement in one of my 8/a machines. The differences are the KK8E CPU takes 5 slots (7 with the EAE) and the KK8A CPU takes one slot. The KK8E CPU is slightly faster then the KK8A (around 14%).
I bought the PDP-8/a because it has a matched set of KK8E and EAE boards. My EAE does not work with my KK8E in my PDP-8/E
There are a number of design errors with the Lafferty design (this is mentioned on the original web page) which don't cause problems with the Cypress branded chips which are no longer available. The most notable issue is that for simplicity a tristate bus transceiver was used to interface to the Omnibus. Everything that drives the Omnibus is supposed to be open collector. There is a possibility that the memory board can drive the Omnibus to 5 volts while the CPU (or other bus master) is pulling the bus to ground causing a shoot thru current. While I don't think this ever happens with the KK8E CPU I think this is the problem with the KK8A CPU. What I do know is that there is shoot thru on the memory board itself when the memory chips are being read and the transceivers are in write mode. This is only a few nanoseconds long but it causes large currents in both the transceiver and the memory chips which can be seen by looking at the power supply pins on those devices. Another mistake is not buffering the address lines. DEC on all their semiconductor memory cards buffers the address lines with a Schmitt trigger gate. The reason it mostly works without it is that there are only the two CMOS inputs on the address lines and almost nobody has a maximum length omnibus. The other thing DEC always did on their semiconductor boards was latch the data coming out of the memory chips and then drive the bus with open collector devices.
I was honestly hoping to find a semiconductor quad or hex board for the 8/a. I have a hex wide 16k x 12 (H219B) core plane board but I do not have the driver board it mounts to. If I can't find something I will have some of Vince's memory/bootstrap boards made. One question, can the boot strap board and the M8317 Option board? If not I will put my existing 32K card in the 8A and put one of the new memory/bootloader boards in my 8/E.
In our testing of the Alliance memory chips we found that they do work about 99.9% of the time. Of course this is not nearly enough as there is the possibility of a misread a couple of times a second. We (me and Vince) were able to get it to run for several minutes of diagnostics before it would fail.
Can we re-design the board to better match DECs design (like using open collector buffers, etc)?

I will front the costs of new boards if you want to do this. If we do, there is a mod I want to make to the bootloader circuit as well.
 
I was honestly hoping to find a semiconductor quad or hex board for the 8/a. I have a hex wide 16k x 12 (H219B) core plane board but I do not have the driver board it mounts to. If I can't find something I will have some of Vince's memory/bootstrap boards made. One question, can the boot strap board and the M8317 Option board? If not I will put my existing 32K card in the 8A and put one of the new memory/bootloader boards in my 8/E.
I've got three of the semiconductor boards and none of them work. The DRAM's of the day don't seem to have done all that well with the passage of time. And since the boards are multi layer they are more difficult to safely repair.

I believe I have four of the 16k core board sets and they all work. But maybe I am just lucky with the core or unlucky with the semi.

It looks like you can disable the boot loader by setting S1-6 S1-7 and S1-8 to the off position on the M8317 board. That should let you use Roland's board.
 
Doug,

Thank you.

I have 4 partial core boards.

1 Data RAM 16K x 12 core plane (missing the driver board)
1 DEC 16K x 12 Hex core plane (missing the driver board)
1 DEC 4K x 12 core plane missing the sense/inhibit and X/Y driver boards
1 DEC 8K x 12 core plane missing the sense/inhibit board.

I also have 1 functional 4K set and 1 of the Vince/Roland 62256 RAM boards.

More memory is always welcome.

Are any of your 16K board sets spare?
 
Are any of your 16K board sets spare?
Not really. What I want to do is put the ClassIC (Classroom Instructional Computer) back together. The ClassIC is an 8/a with an RX01 and a terminal. I don't remember which terminal (VT102 maybe). That machine has 32k and the FPP-8 in it. There is a desk with the CPU on the left and the terminal on the right. I even have the original chair that DEC shipped with it. The desk had seen a lot better days and is currently sitting in primer grey waiting for me to get motivated and do the semi gloss black top coat. The desk itself was a wood grained laminate that was badly damaged. I managed to get a close match to the original laminate at Lowes and a new desktop is also in the works. The chair is a fiberglass molding with a coarse cloth over polyester padding. The cloth has degraded and needs to be replaced.

The other 8/a is the accounting machine used by Harold's photography in Sioux Falls, SD. They did a lot of film processing but also had a retail store where they sold cameras and did professional photo shoots. That machine was still in service at least as late as 1992 as that is the last time the filters were changed in the RL01 drives. That one had the KK8E CPU with the EAE. At the moment those boards are not working and I have an M8315 in the chassis. It had 16k of memory but I was able to find another 16k for it. The RX01 is currently connected to that machine. It is in a short DEC rack but I am missing the side panels for it. I have one other 8/a backplane connected to a modern Vicor power supply that can be carried around almost reasonably. It needs the wood case completed. The plan for that is to try to get an 8/e front panel instead of the 8/a keypad 7 segment display one. I am hoping I can repair my DEC 128k (one of the M8417 variants) with the M8416. I have one of the 4 port serial cards and it would be fun to put 4 terminals on it and run TSS/8 using the 128k as the swap.

If you can snag that 16k board on EBay for $250 that would not be unreasonable. I looked at the photos and the only thing I noticed from what can be seen was the outside edge of the A connector is damaged. I think it would still make contact but that is something to consider. The top side (AA1) is not used and the bottom side (AA2) is one of the +5V pins. BA2 and CA2 are also 5V so it is probably still usable assuming no other issues.

Best Wishes!
 
As was mentioned, Vince did an 8/a front panel. I am sure he will say something about it. I saw a prototype when I was visiting a few months ago.
It's been a while since I looked at it, though I do have PCBs, last time I checked. Parts are hard to find these days, though, so I wouldn't count on being able to populate them unless you're sitting on a pile of vintage chips.

Vince
 
Can we re-design the board to better match DECs design (like using open collector buffers, etc)?
That's been done, but the Alliance SRAM still won't work reliably. It boots OS/8 and then fails randomly after running the diagnostic for a few minutes. Works fine with every other brand of SRAM I have.

There's virtually no other compatible DIP SRAM in stock anywhere that I can find. DIP FRAM willl also work in the new prototypes, but that is even more scarce. (Which is my excuse for why I haven't been working on the 32K board prototypes lately.)

Vince
 
What are the requirements for "compatible DIP SRAM"?
What is your hypothesis(es?) for the Alliance SRAM instability?
I don't understand why the OMNIBUS is peculiar in this regard.
Or maybe it's just a pecularity of the 8/E and/or 8/A processor interfaces to the OMNIBUS?
 
What are the requirements for "compatible DIP SRAM"?
What is your hypothesis(es?) for the Alliance SRAM instability?
I don't understand why the OMNIBUS is peculiar in this regard.
Or maybe it's just a pecularity of the 8/E and/or 8/A processor interfaces to the OMNIBUS?
The board as it is currently laid out needs two 62256 equivalents in DIP28-6 packages.

There are only two possibilities -- either we've missed something, or the Alliance SRAM really are defective and can't pass the memory checkerboard reliably. The other tests pass (with the current prototypes), and the system is actually stable as far as I can see, but the diagnostic won't pass. I think the longest time I've ever seen it run is about 20 minutes. It is also suggestive that I've tried several different brands of SRAM and even FRAM I bought off eBay, and those all run the checkerboard test without issue.

I don't know that it has anything to do with Omnibus. It seems more to do with the checkerboard test. OTOH, I don't have anything else on hand to test 62256 compatibility with.

Vince
 
Got it, thank you. In essence it's a form-factor limitation coupled with a(n apparent) supplier issue. But there are other suppliers in the third-party arena. So folks should socket and expect to possibly have to swap parts coupled with extended testing. For a new design/implementation the form-factor problem can certainly be overcome. Use a pair of AS6C4008-55PCN (DIP-32; $6.38 each at Mouser) and "move on" :->.

For current boards, why not create a small adapter since the pin assignments are compatible -- tie down pins 1/2/30/31 and shift pin 32 to socket-28 (Vcc)? There appears to be sufficient room on the full-size (but I believe also true for the mini) PCB to accommodate the overhang on that end. Or has this already been tested and found wanting?
 
"Alliance" was a pretty vague specification. I was assuming that was a reference to a 62256-equivalent. I'm boggled by your findings WRT the AS6C4008-55PCN since it's used reliably in lots of 8-bit designs. I can't help but think that the problem is marginality in the OMNIBUS interface (as you/others have pointed out elsewhere), not the SRAM.

Why do you have "little faith in their larger parts either"?
 
I'll admit that I haven't tried the 512Kx8 parts, but only their 32Kx8 parts. I have treated it as an issue with vendor quality control, though I can't think of any way to actually prove that.

I don't see anything remarkable about the Omnibus interface, other than that it is open collector (instead of tri-state). The timing isn't difficult, and the signals are nowhere near the speed limits of the chips.

Vince
 
ISSI makes 62256 parts. All in some kind of surface mount. The IS62WV25616EBLL-45BLI is a BGA. We might be able to find a socket for this. Mouser has 145 in stock @ $3.21/1 or $2.89/10.
Cypress makes the CY62256LL-70PC. Jameco has 8 in stock at $4.95/1 and $4.59/10. Arrow

There are many on EBAY.

I'm tempted to buy the 8 that Jameco has.

If we redo the memory/bootloader board I would def like to make a circuit addition to the boot loader part.

What I would really like to do is use a raspberry pi pico cpu and add memory read capabilities. This is the start of a PDP-8 Omnibus Debugger/Multi I/O board that I want to make. I just don't have the hardware chops to design it.
 
I'll admit that I haven't tried the 512Kx8 parts, but only their 32Kx8 parts. I have treated it as an issue with vendor quality control, though I can't think of any way to actually prove that.
Well, could try out a pair of 512Kx8 parts using the bent-pin technique and a small wire for the Vcc repositioning. Might need to add an intermediate piggy-back socket to ensure clearance from adjacent parts. If MT-sockets were used on the PCB then adding a piggyback socket is trivial. If wipe-sockets then one needs to be careful to ensure good contacts.

While it's certainly possible that the vendor has a production-wide systemic problem, IMO that's highly unlikely ... particularly as the (much!) higher density parts are undoubtedly a completely different process. Else I doubt that they'd still be in business given a bad reputation for reliability. It's also possible that the 32Kx8 QC problem was batch-specific and you got "lucky" with your tubes :-{.
 
Back
Top