Interposer PCBs just arrived today from JLCPCB...

Today I assembled the PCB and after some TQFP pins solder issues I made it POST! The BIOS detects it as "80486DX/2" at 133MHz. When multi on board is set to 3x it runs at 100MHz and when 2x it runs at 133MHz. L1 cache probably runs in WT mode. Doom runs pretty fast with S3 765VL and Quake is quite playable in 320x200 too.
Doom: 40.91fps
Quake: 12.0fps
I have some stability issues, under the load the LDO (even it has a small heatsink) gets very hot burning my fingers and probably causing some random restart/hang. The CPU is not as hot but would need some heatsink too. I have to make some custom spring to hold some heatsink as my adapter has bigger height than std PGA CPU...

Could you show the pin-side view? The inner circle must have been awkward to solder.

Normally I see these QFP-208 to PGA-168 interposer boards with 2312 and 0805 pads for capacitors. Were you going to add pads?

Place a fan over the VRM and run it at 100 MHz. Does it still crash? Try the interposer on another motherboard.

Soldering of the inner circle was not so hard with a long thin solder tip. I don't use SMD pins but i cut one end of normal pins. There are 10 SMD 0402 caps 100nF around the package between pins and I use 4-layers PCB to have a solid VCC and GND planes inside. I'll make a better photo later...

RayeR wrote on 2024-09-26, 03:01:

Ou, bloody hell, I missed that the footprint on board is not SQFP208 but PQFP196 (for 486SX) and the pinout is totally different so I cannot solder the DX5 there :(

By the way, I had also made this mistake when looking for QFP208 boards for the Am5x86 QFP. There were ample 486SX QFP on PGA interposers.

Can I know what you are using for the pins?

Is there just one PCB in the final product, or do you have two PCB's sandwiched together to create pseudo blind vias?

Are the gerbers available? I may order a 5-pack from JLCPCB.

Ah, some mistakes are repeated again 😀 at least by different people...

I use standard precise pin header by Connfly
https://www.gme.cz/v/1498962/connfly-prs140g- … -kolikova-lista
and there are also cheaper on Aliexpress. The interposer is just one 4-layer PCB, no reason for sandwich. On the photo above are 2 PCBs to show top and bottom side at once. I got 6 pieces from JLC-PCB (std order is 5 pcs but some bonus?) So I have some free available. But the shipping from my country will be much more you would pay to JLCPCB. Gerbers was not published yet but if you wanna make a copy for personal use I could give it to you.

I did some rework under magnifier to fix some poorly soldered and misaligned pins. Now it should be OK...

I mounted a bit bigger heatsink on the LDO and lay down a big one on CPU and then it seems to be stable, an 0.5hr during memtest... Then I pushed it higher to 4*40MHz but it's not stable. First it seemed to be sensitive on cache timing so I relaxed from 0WS 2-1-1-1 to 1WS 3-2-2-2 step by step but it still couldn't finish Quake timedemo and crashes after some seconds. With 0WS also other pmode apps crashed just at start. I also relaxed other timings like RAM, ISA etc. but didn't help in Quake. Doom passed. So one question, are 15ns + 12ns tag SRAM capable to run at 40MHz at fastest timing? And for 50MHz? But I don't have enough 12ns chips anyway... I still suspect the LDO doesn't behave optimal, maybe adjustable version would be better because it could pick up feedback signal closer to CPU. Also adj. would let me set e.g. 3.4-3.5V that would probably help CPU stability. I'll try to look for LM1084-adj but I don't like that it needs to be mounted by cooling side to MB interior that makes mounting bigger heatsink problematic compared to my version where cooling side of LDO is directed outside of MB....

I also chaged capacitors on interposer that I added some 10 and 22u so I have cca 70uF total. I'll have to hook a scope on vcore to see what's going on there...

15ns should be good for up to 66Mhz. 20ns up to 50Mhz. 12ns up to 83Mhz.

Try disabling L1/L2 cache and see what happens in Quake.

Could you show a bottom view of your adaptor with pins soldered? How are you making the pins the same length, if some are through-hole, and others not?

Not all Am5x86-133 chips can handle 160 MHz. I have one QFP Am5x86 on an Evergreen interposer which needs 3.9 V to run stable, while others can handle 3.45 V at 160 MHz. Try running at 4 V.

Better photos are here...
I simply cut off the pins on one side of doublesided header by sharp Knippex. Then I put solder blob at 1st and last pin of the row, align it to position and solder on PCB. Then soldered all remaining pins. I used empty PGA socket to fit and align all pins when soldering through-hole headers.

As Quake passed with slowest cache timing it would pass with disabled cache. I definitely change the fixed voltage LDO to some adjustable one so I could give it a little bit voltage boost...

BTW I'm curious if some 486 desktop machines used SMM/SMI. This board don't have routed this pins at all. My interposer has a LED at SMIACT to indicate the CPU is in SMM. Maybe only laptops used this mode for handling some PM events...

Thanks for the photos. It would probably help to also provide a 45-degree angle shot of the bottom inside corner.

I think I know what you've done. To confirm, the unsoldered end of the PGA male-male pin is pyramidal and the cylindrical end is soldered? If yes, how much larger (diameter) did you make the surface mount pads for the inner perimeter pins comapred to the cylinder itself? For and soldering to work well, these inner perimeter surface mount pads should be wider than the cut cylinders. If they are not wider, then did you use solder paste?

What voltage is your motherboard's VRM outputting? Are you able to confirm that a standard cermamic ADW or ADZ also needs to have these reduced timings at 160 MHz on your motherboard?

When you say you are using 0402 100 nF capacitors, I assume you are referring to the imperial sizing?

0402 metric = 1005 imperial
0402 imperial = 1005 metric

1005 (0402 metric) = 0.4 x 0.2 mm
0402 (1005 metric) = 1.0 x 0.5 mm
0603 (1608 metric) = 1.6 x 0.8 mm
0805 (2012 metric) = 2.0 x 1.25 mm

I noticed that you have two different capacitor sizes. Did you end up using different capacitance's?

I never bothered to check if any of my motherboards have SMM (SMI#/SMIACT#) routed to the CPU, but I have always wondered about this. Does the MB's BIOS need the SMM related pins connected to throttle the CPU clock down when idle/suspend timers are reached? Or is there alternate means of achieving clock throttling?

Do you have an AWARD BIOS with the clock throttle feature in the power management section of the BIOS?

Are you are using a 390-ohm resistor for the SMM LED? What are the nominal pad sizes for LED and resistor?

I noticed that you have some CLKMUL pads. What do you think about making these through-hole, perhaps with the same pitch as the header found on a SCSI HDD ID jumper block (2.00 mm)? Alternately, standard 2.54 mm could fit if the black mounting block on the header was discarded.

feipoa wrote on 2024-11-14, 08:52:

Thanks for the photos. It would probably help to also provide a 45-degree angle shot of the bottom inside corner.

I think I know what you've done. To confirm, the unsoldered end of the PGA male-male pin is pyramidal and the cylindrical end is soldered? If yes, how much larger (diameter) did you make the surface mount pads for the inner perimeter pins comapred to the cylinder itself? For and soldering to work well, these inner perimeter surface mount pads should be wider than the cut cylinders. If they are not wider, then did you use solder paste?

Heh, a lot of questions. I think better would be to quick-draw a simple picture 😀
The trick is that I didn't flattened the pin bottom but leave a short tail that elevates the header above PCB surface and let me put a thin solder tip inside to do soldering job. I use pad dimensions - THT: 1,4/0,7mm, SMD: 1,6mm

feipoa wrote on 2024-11-14, 08:52:

What voltage is your motherboard's VRM outputting? Are you able to confirm that a standard cermamic ADW or ADZ also needs to have these reduced timings at 160 MHz on your motherboard?

I put fixed voltage LDO 3,3V there but going to change it...
I don't have any other 5x86/DX5 CPUs, of course if I would have a ceramic one I will never mess with this interposer 😀

feipoa wrote on 2024-11-14, 08:52:

When you say you are using 0402 100 nF capacitors, I assume you are referring to the imperial sizing?
I noticed that you have two different capacitor sizes. Did you end up using different capacitance's?

it's 0402 imperial but 0603 can be soldered between the pins as well. Yes, I decided to use 3 capacitor vaues: 100 nF, 10 µF and 22 µF, cca 70 µF total. I didn't have time to unpack scope and check voltage ripple yet but surely I'll check it.

feipoa wrote on 2024-11-14, 08:52:

I never bothered to check if any of my motherboards have SMM (SMI#/SMIACT#) routed to the CPU, but I have always wondered about this. Does the MB's BIOS need the SMM related pins connected to throttle the CPU clock down when idle/suspend timers are reached? Or is there alternate means of achieving clock throttling?
Do you have an AWARD BIOS with the clock throttle feature in the power management section of the BIOS?

If I remember well I never seen this throttling option in any 486 board I had play with. I guess such features was reserved for laptops domain...

feipoa wrote on 2024-11-14, 08:52:

Are you are using a 390-ohm resistor for the SMM LED? What are the nominal pad sizes for LED and resistor?

R is 390ohm 0603 and LED is 0603 (imperial) too.
pad sizes for LED: 0.8*0.8mm , center distance 1.6mm
pad sizes for resistor: 0.94*0.7mm , center distance 1.575mm
I use this footprints with cooperation of one PCB assembly company and they are OK with it.

feipoa wrote on 2024-11-14, 08:52:

I noticed that you have some CLKMUL pads. What do you think about making these through-hole, perhaps with the same pitch as the header found on a SCSI HDD ID jumper block (2.00 mm)? Alternately, standard 2.54 mm could fit if the black mounting block on the header was discarded.

I think it's not needed. My MB has CLKMUL routed on a jumper and if not I would just put a solder blob or 0R jumper over it...

Wouldn't it be easier for soldering to let the SMD-PGA solder pad protrude into the inner circle, not needing to stick a solder tip into a small crevice at an angle?

If you are still running at 3.3 V, I can see why 160 MHz isn't shaping up for you at tight timings.

If I recall right, my M919 and MB-8433UUD socket 3 desktop motherboards have these throttle down options in the BIOS.

Having a CLKMUL header on the interposer would mostly be for convenience, switching between 3x/4x.

feipoa wrote on 2024-11-14, 21:59:

Wouldn't it be easier for soldering to let the SMD-PGA solder pad protrude into the inner circle, not needing to stick a solder tip into a small crevice at an angle?

Maybe, but I wanted to have vias close enough to SMD CPU pads, not routing longer traces to center. And as I didn't have serious problems with soldering...

feipoa wrote on 2024-11-14, 21:59:

If you are still running at 3.3 V, I can see why 160 MHz isn't shaping up for you at tight timings.

Yes I read that some users have issues with running it stable with stock voltage some not, depends how good piece you got...

feipoa wrote on 2024-11-14, 21:59:

If I recall right, my M919 and MB-8433UUD socket 3 desktop motherboards have these throttle down options in the BIOS.

Having a CLKMUL header on the interposer would mostly be for convenience, switching between 3x/4x.

I guess that most of advanded MBs supporting DX2/4 CPUs already have a jumper on board for this. I just routed some pins to test points if I have a need to try something...

It is up to you, but I think extending the SMD pad, perhaps 1-2 mm towards the centre would help a few people who want to assemble one of these. As a consequence, you'll also be able to get all the PGA pins the same length if you file off that stub. To be clear, the SMD pad would go from a circle to an oval, or one side round, the other square.

The CLKMUL header suggestion was just as an extra convenience for users who don't want to find the CLKMUL pin on their motherboard, or if their board doesn't have one. Most commercial interposers had the CLKMUL jumper, but it isn't really important. I think extending the SMD pad for PGA is more beneficial as you'd be able to use most larger soldering tips.

Ok, got it. This pad mod would need a lot of rerouting work of current design (moving lot of vias), I don't plan to make any further PCBs...

I know, you'd have to manually drag inward nearly all the vias. Maybe someone else will come along who wants to do this...

Not so simple bcause of corners where it would cause crossing and need further moving...

It is certainly possible if someone wants to put in labour. If you upload the gerbers, one such individual may come along.