That is unfortunate indeed. Lets see if 4036yb handles it better for you.
Not surprised that aha-1522b did better than 1542c#.
Once you get the Advansys / SIIG controller, it should move the needle quite a bit.

Btw, i never sa 1522b, only 1520b. Do you really have 22 ?

pshipkov wrote on 2024-05-19, 05:20:

@MikeSG
What was the point of the flexible ribbon mod ?

I'm simplifying the flush circuit on my design using the Improve Tech circuit. Just missing a few connections.

Will share the gerber files after testing.

Post is at Re: Custom interposer module for TI486SXL2-66 PGA168 to PGA132 - HELP!

pshipkov wrote on 2024-05-20, 06:18:

That is unfortunate indeed. Lets see if 4036yb handles it better for you. Not surprised that aha-1522b did better than 1542c#. O […]
Show full quote

That is unfortunate indeed. Lets see if 4036yb handles it better for you.
Not surprised that aha-1522b did better than 1542c#.
Once you get the Advansys / SIIG controller, it should move the needle quite a bit.

Btw, i never sa 1522b, only 1520b. Do you really have 22 ?

Yes, i actually have the AHA-1522B, which just means it contains the integrated floppy port and related IC's. This saves an ISA slot, assuming you use floppy. If you need a multi-I/O card anyway, then it wouldn't help much.

Edit: Attached photo of the 1522B on testbed:

@MikeSG
I guess you are doing this in the name of completeness, correct ?
I tested these SXL2 CPUs on a LOT of motherboards and with one exception they worked just fine. Not sure if additional improvements are actually needed.
If i don't solder the two chips on the adapter will it still function ?
My point is that simplicity is a positive factor for DIY hobby contraptions given the limited time we can allocate to this stuff.

@Feipoa
Good point that most 386 mobos require multi-i/o card, so 1522 does not bring a lot of value over 1520.

pshipkov wrote on 2024-05-21, 05:59:

I guess you are doing this in the name of completeness, correct ? I tested these SXL2 CPUs on a LOT of motherboards and with one […]
Show full quote

I guess you are doing this in the name of completeness, correct ?
I tested these SXL2 CPUs on a LOT of motherboards and with one exception they worked just fine. Not sure if additional improvements are actually needed.
If i don't solder the two chips on the adapter will it still function ?
My point is that simplicity is a positive factor for DIY hobby contraptions given the limited time we can allocate to this stuff.

I had the same thinking in this regard. I hope Mike makes the invalidation circuit disable-able with a jumper, or minimally a solder bridge. I haven't run into any 386 boards which didn't work with the SXL2 out of the box with the cyrix.exe utility. Do you recall the exact motherboard you tested which didn't work with the SXL2 and cyrix.exe.

Nonetheless, I think there could be some obscure OEM or MCA systems out there that may benefit from this external cache invalidation circuit, whereby MEMW# and/or cyrix.exe won't cut it. There was another user in this thread who had an MCA system and needed to add an inverter and OR gate to the design to get the SXL2 working, however this option was already added to Alpha 1.

Or maybe with the Improve-It invalidation scheme, we can bypass cyrix.exe all together, or get improved AHA-1542 support? And hopefully without sacrificing overclock-ability. I know I tested for this in the past with a similar Improve-It interposer, but no longer recall the outcome.

pshipkov wrote on 2024-05-21, 05:59:

@Feipoa
Good point that most 386 mobos require multi-i/o card, so 1522 does not bring a lot of value over 1520.

However, for all my systems, I use PS/2 for the mouse, thus skipping the multi-I/O card entirely. You can probably see why the 1522B and 1542CP w/bulit-in floppy appealed to me. If there's ISA slots left, I may still add a serial I/O card to test serial mice if needed.

pshipkov wrote on 2024-05-21, 05:59:

@MikeSG I guess you are doing this in the name of completeness, correct ? I tested these SXL2 CPUs on a LOT of motherboards and […]
Show full quote

@MikeSG
I guess you are doing this in the name of completeness, correct ?
I tested these SXL2 CPUs on a LOT of motherboards and with one exception they worked just fine. Not sure if additional improvements are actually needed.
If i don't solder the two chips on the adapter will it still function ?
My point is that simplicity is a positive factor for DIY hobby contraptions given the limited time we can allocate to this stuff.

It's mostly for completeness. It's a "wire free" cache invalidation in case the Cyrix utility doesn't work, but I guess it works 99.9% of cases anyway.

I'll just delete the extra chips.

The extra cache invalidation circuit may also be useful if the user is wanting to run a non DOS-based operating system which cannot run the cyrix.exe utility, like Windows NT, OS/2, or CP/M. I'm not sure if CP/M and OS/2 can run cyrix.exe)

BTW other OSes can be usually booted from plain DOS via GRUB4DOS, so you can configure your HW with DOS utils, call GRUB4DOS.EXE and chainload other OS like NT. It can be all automated in autoexec.bat and grub menu.lst to select some boot item within a given timeout.

RayeR wrote on 2024-05-21, 20:11:

BTW other OSes can be usually booted from plain DOS via GRUB4DOS, so you can configure your HW with DOS utils, call GRUB4DOS.EXE and chainload other OS like NT. It can be all automated in autoexec.bat and grub menu.lst to select some boot item within a given timeout.

This is what I do, but not everyone wants to put in this extra effort.

Updated my post containing PCB design req, Bill of Materials with the new simpler layout. Re: Custom interposer module for TI486SXL2-66 PGA168 to PGA132 - HELP!

Ordered them today and will arrive between Jun 20 - 30 to test. Once it works I'll upload the gerber files.

Only doubts are whether the press-fit holes are within tolerance. There was an option when ordering to ensure +/- 0.05mm but was +$280AUD.

The first fully assembled prototype is built now.

The clearance of the press-fit 168-socket is perfection for some PCBs but slightly too tight for others. I only tested two PCBs for fitment out of seven, but expect at least half to fit well. It comes down to PCB manufacture.

Added soldering notes at Re: Custom interposer module for TI486SXL2-66 PGA168 to PGA132 - HELP!

I will add the gerber files after testing.

The only problem is my 386DX board is a repair job and it doesn't work right now. I'm buying an oscilloscope next week. After 2-3 weeks if nothing has changed I'll be open to either sell/trade to someone to test it.

Tested voltage regulator no-load voltage, no-cpu power on, and cpu power on.

Looks clean!

For which scenarios are NA-SEL# jumpered to 1-2 or to 2-3?

It is unfortunate that you don't have a working 386 motherboard to test this design on. What motivated you to create this interposer without a working 386 system?

Looks like your motherboard is a Biostar MB-1333C rated for 33 MHz. I have this board as well and was planning on swapping the F82C351-33 MHz northbridge for a 40 MHz part at some point. I hope you can get your board working; it seems unlikely that someone will pay or trade for an untested interposer design.

I noticed that you only have 4 SIMM sockets filled. Some of these old 386 boards don't function unless all 8 SIMMs are filled. In some cases, they will work with only 4 SIMMs, but depending on the motherboard, you may need to install the SIMMs in all Bank A, all Bank B, or half in Bank A and half in Bank B. I remember the CHIPS board you have is fussy about this. Do you have 8x 1 MB SIMMs? Start with a standard AMD 386DX and 8 MB; what are your error symptoms?

Only use 64K L2 on this board. I haven't figured out exactly whic SRAM chips will work for 256K, but they aren't the regular 32kx8 chips. I have a few options on order for experimenting with this.

Hm, I never seen yet a 386 MB that would require 8 modules, I don't see reason for it as 30p SIMMs are 8bit wide and 386 is 32bit wide so 4*8=32. It may be tricky how to populate 4 modules to 8 sockets, there may be different schemes but nothing hard to try out. Even without a RAM the BIOS should send some POST and beep codes to see a sign of life-if CPU works at bare minimum ok...

Sometimes when maxing out ram you have to use a larger number of smaller sticks. Maybe that’s what happened, but I also thought that 386 required 2 or 4 sticks. (Weren’t there some with 16-bit busses?) feipoa has messed with these quite a bit, I’m sure he knows what he is talking about, though I would like to understand why some boards need 8

Yeah, repair that board with a standard basic cpu first.

386SX - external 16bit bus, usually works with 2 modules
386DX - external 32bit bus, usually works with 4 modules

Yes I know, 386DX should only need 4 modules to account for the 32-bit data width, but sometimes I could not get 4 modules working. When I could get the 4 modules working, I had to try all sorts of positions on these older boards, e.g.

SIMMs
1 - filled
2 - filled
3 - filled
4 - filled
5 - empty
6 - empty
7 - empty
8 - empty

or

SIMMs
1 - empty
2 - empty
3 - empty
4 - empty
5 - filled
6 - filled
7 - filled
8 - filled

or

SIMMs
1 - filled
2 - empty
3 - filled
4 - empty
5 - filled
6 - empty
7 - filled
8 - empty

or

SIMMs
1 - empty
2 - filled
3 - empty
4 - filled
5 - empty
6 - filled
7 - empty
8 - filled

I didn't figure out those last two until a year or two ago. If you've tried all of these with 4 sticks and it still doesn't work, the BIOS or chipset may be insisting on all 8 SIMM slots being filled.

I’ve been pondering that it may be possible to make this interposer to fit entirely within the cpu socket. A lot of work though.

Sphere478 wrote on 2024-06-28, 01:12:

I’ve been pondering that it may be possible to make this interposer to fit entirely within the cpu socket. A lot of work though.

I don't see value in this unless someone finds a source for these QFP144 SXL2-66 chips:

I've tested alpha1 (the larger PCB) on a lot of 386 boards and haven't run into any hard barriers. I've had to tilt a tantalum out of the way one or two times. For the worst case, I had a jumper on the motherboard sitting right under the interposer PCB. To clear it, I had to snip off the ends of 5 PGA168 pins, shown here:

The PCB itself can fit over motherboard jumpers, but it is the protrusion of the PGA168 pins themselves which can cause clearance issues. For this situation, it wouldn't matter how small the PCB is, unless we can find a source for these rare QFP SXL2 chips.

feipoa wrote on 2024-06-27, 08:59:

For which scenarios are NA-SEL# jumpered to 1-2 or to 2-3?

NA-SEL is for motherboards that use memory pipelining with the NA pin. The TI486SXL manual says some motherboards require a 10-20k pullup - NA jumper 1-2, or to be disconnected with a 10-20k pullup - NA jumper NC, or to be disconnected with a 100r pulldown - NA jumper 2-3.

feipoa wrote on 2024-06-27, 08:59:

It is unfortunate that you don't have a working 386 motherboard to test this design on. What motivated you to create this interposer without a working 386 system?

I have one 386sx and one 386dx board left (sold the 486 boards) and want to fix/tune them up with 486 upgrade chips, run performance tests, test duke nukem 3d, c&c red alert, and some other games, and as engineering projects/talking points they're interesting...

feipoa wrote on 2024-06-27, 08:59:

Looks like your motherboard is a Biostar MB-1333C rated for 33 MHz. I have this board as well and was planning on swapping the F82C351-33 MHz northbridge for a 40 MHz part at some point.

In searching a little while ago I found the F82C351B/40 and the F82C351/C. I tried to buy the C version but was way outside the price I was willing to pay. Only HK/Chinese suppliers. https://www.digipart.com/part/F82C351

feipoa wrote on 2024-06-27, 08:59:

I noticed that you only have 4 SIMM sockets filled. Some of these old 386 boards don't function unless all 8 SIMMs are filled. [...] what are your error symptoms?

Will test that. I know the board is hardware configured for parity RAM.

The symptoms are the board runs at 4.8v instead of 5v, everything gets warm except for the BIOS, no BIOS post analyser codes, extensive varta damage/repair. I need to check under the kb controller and one or two other chips. Also the BIOS I'm using is for the F82C351B and I have a F82C351A.

feipoa wrote on 2024-06-27, 08:59:

Only use 64K L2 on this board. I haven't figured out exactly whic SRAM chips will work for 256K, but they aren't the regular 32kx8 chips. I have a few options on order for experimenting with this.

They are 16Kx4 for the 64KB chips. I bought some 15ns versions "MCM6290P15" to replace the 25ns's. Tested both... I think the 256KB versions use 64kx4.

The board overall is in good shape, just need to find that one thing stopping it from booting.

I noticed the pins aren't through-hole for your jumper. Are you using a special surface mount header for this? I didn't see flared feet on your photo.

I see you wrote the following,
NA-SEL#.
Default is jumpered 1-2 or No jumper. No jumper = NC and 20k pullup. 1-2 = Connected & 20k pullup. 2-3 = NC and 100r pulldown.

However, I am still a little confused. Which jumper position is used for the SXL and which is used for the i486SX for the situations in which, a) the chipset does not support pipelining, b) the chipset supports pipelining and drives NA#, and c) the chipset supports pipelining but does not drive NA# ? Seems like there are 6 conditions.

From what you've written, and viewing D-4 in the processor databook,

No jumper:
The NA# pin from the female side of the interposer is not connected to the male side of the interposer. In addition, the female side of the interposer has NA# pulled up to 3.6 V via a 20K resistor on the interposer. This corresponds to i486SX (chipset does not support pipelining).

1-2:
The NA# pin from the female side of the interposer is connected to the male side of the interposer. In addition, the NA# pin is pulled up to 3.6 V via a 20K resistor on the interposer. This corresponds to SXL (chipset does not support pipelining)

2-3:
The NA# pin from the female side of the interposer is not connected to the male side of the interposer. In addition, the NA# pin is pulled down to GND via a 100-ohm resistor on the interposer. This corresponds to i486SX (chipset supports pipeling but does not drive NA#). Is there an equivalent SXL option in which NA# is connected from male to female sides here (w/pulldown)?

Is there also an option in which the NA# pin from the female side of the interposer is connected to the male side of the interposer and there aren't any pullups or pulldowns? This would correspond to the chipset supporting pipelining and driving NA# directly (SXL). Alpha 1/2 were wired in this manner.

However, when I measured the NA# pin on a few motherboards, I noticed the following:

Rabbit, 4.7K pullup to 5V
UMC, 10K pullup to 5V
Dtk Symphony, 10K pullup to 5V
Symphony 2, 1K pullup to 5V
C&T, 4.7K pullup to 5V
SiS460, 4.7K pullup to 5V
VIA, short to 5V
ALi, 10K pullup to 5V

which would imply to me that the chipset does not supporting pipelining. Which chipsets used on a 386DX supported pipelining?

Do you know what the advantage of the C&T F82C351/C is compared to B1? Were all the C chips rated for 40 MHz? I see the F82C351/C going for about $19 USD.

From my experience with Varta leaks on these 386 boards, you need to remove any components which you suspect may have trace damage. I've had to remove ISA slots, AT power connectors, SIMMs, DIP sockets, etc to fix bad leakages. Be sure to neutralise the spill using a vinegar bath.

I had one C&T board in which the DRAM ran very hot and the board eventually died. I never troubleshooted the issue and used the board as a parts donor.

Yes, the MB-1340C uses 64Kx4 for the 256K option, however, there are at least 3 variants of 64kx4 that I am aware of. Some are DIP-24 and some are DIP-28. Of the DIP-28 format, there are two different pinouts, that is, one has an extra CE2# pin, while the other is N/C for the same pin. To further the complication, the motherboard has DIP-28 sockets for both the DATA and TAG sockets. I don't have the DIP-28 variant which has the extra CE2# pin.