I was able to get my original protoboard unit working by replicating my hardware from page 3 of this thread. Using these:

386-VC-H motherboard
- 256K cache
- Cyrix 83D87-40 FPU instaleld (will not work without)
- 8 sticks of 4 MB FPM memory (4 sticks will not work)
- CL-GD5434 ISA graphics card (ET4000AX won't work)
- Cannot use a keyboard controller with PS/2 mouse modification
- 20 MHz or 33 MHz crystal oscillator (all others, even slower ones, won't work)

As mentioned on page 3, I'm able to get passed the BIOS tests, but am unable to boot to floppy or HDD. With 20 MHz, it tries to boot to floppy, but stops after a few seconds. With 33 MHz, it won't get to the boot stage. Note that all these unique restrictions do not apply if I'm using a standard SXL2, or my SXL2-G66 on my taller interposer on the same motherboard.

If using Blavius' adaptor, the system won't light up at all. I'm surprised he was able to get further on his system. I will need to go one-by-one with all the pins to ensure they are connected.

EDIT: I tested all the connections on Blavius' adapter. Only one ground pin, located at location L17, wasn't connected. Fixed that, but still no joy. Also tried J1 on VCC3. It didn't help. I feel like we are missing something fundamental here, like needing to filter the data and address lines. Perhaps add an aluminium Faraday cage over the adapter?

EDIT2: Using a 33 MHz oscillator, if I disable L1/L2 cache in the BIOS, I am able to view the SCSI card's BIOS during startup. The Adaptec BIOS, however, will not detect the HDD with this slow of an oscillator. Same thing occurs when using an original PGA132 SXL2-50. I need at least a 50 MHz cyrstal to boot from any media.

EDIT3: Tried adding a grounded Faraday cage over my original protoboard, but didn't help. Electrical tape added to the inner, outer, and bottom of the chassis to prevent shorting. After 3 hours fooling with this, I'm now having trouble getting my the system to show signs of life, even with 20 & 33 MHz oscillators. But if I insert the PGA132 SXL2-50, no issue.

Blavius, would you be willing to test your adapter on one or two other motherboards? I'm wondering if you are seeing the same fussiness that I am.

The only other thing I have to test this with is a second IBM P70, except this one is slower at 16MHz baseclock (33MHz xtal), instead of the 20MHz I tested the adapter with. Mobo is a little different, so who knows. I'll have a look tonight and let you know.

Edit:
The adapter unfortunately doesn't fit in my other P70, some tall components are in the way. That's the issue with the side-by-side design clearly demonstrated.

Feipoa, if you can get the system to show a sign of life at 16Mhz, problems at higher frequencies could indeed be due to interference. Is there a way to shield the bottom of the pcb that doesn't cause shorts? Maybe you could tape alu foil on the parts outside of the sockets, with the dense interconnect strips.

feipoa wrote on 2022-10-02, 08:56:

- Cyrix 83D87-40 FPU instaleld (will not work without)

sure looks like bus loading issues, that usually means reflections

feipoa wrote on 2022-10-02, 08:56:

- 8 sticks of 4 MB FPM memory (4 sticks will not work)

It shouldnt matter because afaik there are usually 245 buffer chips between CPU bus and RAM data pins?
https://theretroweb.com/motherboards/s/fic-386-vc-h ? those would be U11-14 245 and RP19-26 22ohm resistors
hmm maybe not, you can measure between SIMM socket pin 3 and CPU data pins if there is direct connection

feipoa wrote on 2022-10-02, 08:56:

if I disable L1/L2 cache in the BIOS

try taking cache chips off the board, they are directly on CPU bus with no buffering

feipoa wrote on 2022-10-02, 08:56:

- 20 MHz or 33 MHz crystal oscillator (all others, even slower ones, won't work)

sensitivity for certain frequencies sure looks like reflections/interference.

feipoa wrote on 2022-10-02, 08:56:

I feel like we are missing something fundamental here, like needing to filter the data and address lines.

wont say no to that

feipoa wrote on 2022-10-02, 08:56:

Perhaps add an aluminium Faraday cage over the adapter?
Farady_over_protoboard.JPG

good idea, but looking at your picture is that RED alligator clip supposed to be ground? by connecting it with thin long wire you made it worse, now its an antenna 😀. Kitchen foil is great for RFI/EMI experiments, just make sure you have a real aluminum one and not the cheap Chinese lookalike plastic stuff, and take care to put a paper barrier between alu foil and electronics otherwise you will short something. Shield must be connected to local ground in shortest most direct way possible.

This is a perfect time to whip out a scope
-connect ground lead VERY close to the points you are measuring
-start probing CPU BUS while resetting system
perfect setup would consist of trigger on a probe on a cpu pin signaling its in control of the BUS, while another one is in your hand checking D0-31 and address bus. That way you could distinguish who is talking on the bus.

Thanks for your detailed reply. Yeah, the red clip is connecting to the PSU ground. The clip is only 30 cm long. I've use this [temporary] trick a few times while designing prototypes at work and it improved the situation tremendously. In both cases, I used an temporary aligator clip of the same length. To make it shorter in this particular application, I'd have to cut and solder a wire. Maybe I can get that 30 cm down to 15 cm.

I can probe with the scope. What is the objective, to observe TM reflections and confirm their existence? I don't think there's anything I can do to resolve all of them. For my wire-bound protoboard, I suppose I may be able to fix a few. When you say to probe the CPU BUS, which specific pin are you referring to? And which pin to act as trigger? Thanks.

EDIT:
What I find curious is this 23 unit stack of PGA132 sockets on my BL3-75/VLB system. Added contact resistance, inductance, etc. but I haven't had any issue running it like this. The socket lengths are much longer than the traces on either my SXL protoboard or Blavius' PCB.

feipoa wrote on 2022-10-02, 20:34:

socket_stack_for_BL3.JPG

absolute king shit

Here in the next few days I’m gonna do beta 3.141

Let’s have someone order it and give it a whirl, see what we got.

I think the only thing better would be a smd design.

feipoa wrote on 2022-10-02, 20:34:

EDIT:
What I find curious is this 23 unit stack of PGA132 sockets on my BL3-75/VLB system. Added contact resistance, inductance, etc. but I haven't had any issue running it like this. The socket lengths are much longer than the traces on either my SXL protoboard or Blavius' PCB.
socket_stack_for_BL3.JPG

That's nuts! Quite an elegant look though, especially with the CPU card sitting on the VLB card 😀

debs3759 wrote on 2022-10-02, 23:04:

...especially with the CPU card sitting on the VLB card :)

Since I took that photo, I added another socket or two so that it doesn't sit on the VLB card. The Trio64V+ card I had was a little taller than my Diamond Trio64.

Yeah, it is rather ridiculous, but functional and fits in the case. It was luck of the draw for which direction the BL3 upgrade module was oriented. I call that system the CN Tower build. It might look better if all the sockets were black.

Sphere, regarding SMD, I suppose it is possible to cut off the machine pin skinny leads on a PGA168 socket, then get those wider stubs all the same length with some smoothing. However, I'm not sure how to surface mount such a socket by hand. I think the ABS would melt. When I tried to hot air solder some PLCC sockets, the ABS didn't stand up so well and I had to use an iron tip. Alternately, it may be possible to surface mount the very outer edge of a PGA168 socket, while leaving the two inner squares through-hole.

Are you able to upload the design to Oshpark, PCBway, or some similar facility so we can order the PCB?

I’ll upload the next beta gerbers you just upload them yourself to the place you want to order them from.

I made this with jlcpcb specifications. So prob use them.

Anyone have a PCB outfit in Canada to recommend?

feipoa wrote on 2022-10-02, 20:34:

Thanks for your detailed reply. Yeah, the red clip is connecting to the PSU ground. The clip is only 30 cm long. I've use this [temporary] trick a few times while designing prototypes at work and it improved the situation tremendously. In both cases, I used an temporary aligator clip of the same length. To make it shorter in this particular application, I'd have to cut and solder a wire. Maybe I can get that 30 cm down to 15 cm.

Thats an antenna. Solder a wire to ground pin/trace directly on the motherboard as close to the new shield as possible.

feipoa wrote on 2022-10-02, 20:34:

I can probe with the scope. What is the objective, to observe TM reflections and confirm their existence? I don't think there's anything I can do to resolve all of them. For my wire-bound protoboard, I suppose I may be able to fix a few. When you say to probe the CPU BUS, which specific pin are you referring to? And which pin to act as trigger? Thanks.

The objective would be to probe working setup, swap to the interposer and compare. No idea about the pins to trigger off of 🙁. But you definitely need a solid consistent trigger to grab same transaction for good comparison. Perhaps M/IO# and triggering on first IO operation in the BIOS would lead to consistently probing same spot. As for trigger type runt might good to try on the interposer setup to try and catch bad signals https://www.youtube.com/watch?v=KiGHdqapP7o

Trying hard to remember the past, I think you are correct, I did have my jumper cable tucked inside the EMI enclosure with GND connection from within. So I tried the SXL2 enclosure again, but with a small cable connected from the inside, but unfortunately it did not help the situation.

With respect to visualising and trying to correct for transmission line reflections, I think this effort may be best utilised after testing Sphere's minimally sized PCB. In the past, I remember having run into a similar issue with CLK when attaching my custom 2x PLL based multiplier on a motherboard. Rather than taking any measurements or pulling out the smith chart, I first added a series trim pot and adjusted it until the thing worked (matched impedence to some degree).

For the case at hand, I find a 33 MHz OSC more reliable than the 20 MHz, but I need at least a 50 MHz crystal for the board to be able to boot from a flopppy. I know this because I tested it with a regular SXL2. Unfortunately, I don't have anything in the 34-49 MHz range. I may have a 47.x MHz crystal oscillator on an ISA card somewhere, but I don't want to hunt it down and desolder it.

Why do you think I'm having slightly better luck with my protoboard adaptor compared to Blavius' PCB? I figure his traces, on average, or shorter, however the trace-to-trace spacing is undoubtedly smaller.

Are you absolutely certain that the machine pins you used to fabricate your 386PGA socket are making contact with the socket on the motherboard? I know that the PGA132 uses machine pins too, but perhaps the machine pin rows that you've used are slightly substandard and not making good contact.

I had an issue with worn holes in a machine pin socket holding a keyboard controller. It took me forever to figure it out.

Indeed, I have ordered some NOS PGA132 sockets not too long ago for the purpose of that CN Tower build. I ordered from 4 different sources. One source, which had about 8 sockets had the male pins with a slightly smaller diameter than standard. It made the sockets fit loosely and I ultimate couldn't use those sockets. However, on my protoboard, the pins fit into the socket VERY tightly, much tighter than with an ordinary PGA132 CPU. So I am fairly confident the issue is not with machine pin contact.

feipoa wrote on 2022-10-03, 07:21:

Why do you think I'm having slightly better luck with my protoboard adaptor compared to Blavius' PCB? I figure his traces, on average, or shorter, however the trace-to-trace spacing is undoubtedly smaller.

sadly its called black magic for a reason https://books.google.pl/books/about/High_spee … id=H5SsQgAACAAJ

Maybe I'm just seeing things, but it looks like you fabricated your own socket using machine pin rows rather than an actual PGA132.

Yes, I'm using these. They fit tight.

How else would you assemble the male end? Is there a single unit gold male-to-male PGA132 connector I could have used?

I haven't checked to see what parts are available in your region. I think I saw some real PGA132s on taobao a while back, but they were certainly not gold. It probably won't make a difference. There's nothing wrong with gluing together your own I guess, as long as you're careful, which it seems you were. It just kind of blows my mind that this adapter is not working. There's electrically no difference between a PGA168 and PGA132 486SXL, so you'd think a passive adapter like this would be a piece of cake.

I’m not sure how you would use a PGA132 socket for this application unless you surface mounted it. It needs male to male to solder on, not male to female. If you see a male-male pga132 online, preferably gold, please show me.

I’m also surprised the adapter isn’t working better. Aren’t the trace lengths on the motherboard much longer than on these side by side adapters?

EDIT: Forgot to mention that the straight pin rows aren't glued; they are just butted up and soldered onto the PCB.