gonzo wrote on 2024-06-17, 09:47:

feipoa wrote on 2024-06-16, 10:30:

In short, you will want to find an unused PGA-168 socket. Cut off all the pins on that socket which correspond to the incoming voltage (Vcc) from the motherboard

Am I right to use a desoldered used socket, too...? :)

Only if the used socket is in fairly good condition, nothing with solder blobs on the pins. Left-over solder on a desoldered PGA168 socket can out stretch the retaining mechanism on socket 3's.

gonzo wrote on 2024-06-19, 08:14:

Thank you, pshipkov. I am a little bit disappointed with the long time for delivery from China, so I had a look at this one buck […]
Show full quote

pshipkov wrote on 2024-06-18, 16:06:

Maximum values for Am5x86: ~5V, ~1A, ~4W.
Looks like ~1.5A buck-regulator will be fine.

Thank you, pshipkov.
I am a little bit disappointed with the long time for delivery from China, so I had a look at this one buck-regulator offered from a German-trader:
https://www.conrad.de/de/p/az-delivery-lm2596 … ductDescription

Input-voltage: 4 to 40 V
Output-Voltage: 3,3 to 24 V
Output-Amperage: 2 A constant
Very comfortable because of a display for the Output-Voltage :)
Price per piece: about 6 Euro

Would this be a good choice?

By the way: how must be the negative pole (ground) of the buck-regulator connected to the PGA-168-interposer? I assume, all the Vcc-pins (23-24 pins) are all only the positive pole...

I like the display. It shows Vin and Vout. It would be even nicer if it could also show current.

Looks like Am5x86 at 3.3 V and 133 Mhz typically draws 825 mA. For over 5V and approaching 200 MHz, maybe around 1.5 A? I think the buck reg. you found should be OK.

gonzo wrote on 2024-06-19, 18:57:

pshipkov wrote on 2024-06-19, 16:42:

Nice device. Will be ok.
CPUs VCC pins go to vout+.
In/out negative poles of the regulator to ground.

So the V(out) negative pole of the buck-regulator does not have any contact with any pin of the interposer? So it goes directly to ground somewhere on the mainboard?

This was one of the questions I had in the SXL thread. It seemed to me like connecting Vout's GND to the PGA might create a sort of ground loop, but other commenters seems to think it would be OK. Worked fine for me connecting Vout's GND to the PGA interposer.

feipoa wrote on 2024-06-20, 02:32:

This was one of the questions I had in the SXL thread. It seemed to me like connecting Vout's GND to the PGA might create a sort of ground loop, but other commenters seems to think it would be OK. Worked fine for me connecting Vout's GND to the PGA interposer.

Thank you, feipoa, this sounds safe enough to be done once again here 😀

Today I have another question about the interposer.

I would use it for an AMD486Dx5-133 (for frequencies 180+/200/200+ MHz) AND an Intel486-DX4-100WB (for frequencies 120-133 MHz @ 4,xx Volt, but less than 5,0 Volt, as I do not have any appropriate i486-CPU starting @ 5V).

Comparing the datasheets for these two CPUs, there is a very interesting pin J1 (attached the relevant points from both datasheets) .

In the AMD-datasheet, this pin is CPU-internal not connected (INC).
In the Intel-datasheet, this pin is a Vcc-pin.

If I understand it right, in case of using an Intel486, this J1-pin have to be connected to the buck-generator, too, but this connection does not matter if an AMD486 is used later instead of the Intel486 (so this J1 can always stay connected to the buck-generator).

Am I right?

and the last pic

pshipkov wrote on 2024-06-19, 20:13:

I can get 240MHz with Peltiers blazing at 12V, but was never able to complete POST.
With that said, the flexibility to use crystal oscillators (or other clockgens) with more inbetween frequencies can potentially move the bar further up from where is now - somewhere between 200 and 230 MHz.

Yes, this was the reason for me not to aim 240 MHz - this is just too much, I think (utopical...)

Can you maybe explain in short words, what is your experience in case of NOT using a Peltier (what are the symptoms)?
- when does the CPU get too hot (instantly; after few minutes; after one hour...)
- can you do something in Windows (some software etc.) until the CPU get too hot
- something else....

I would first check to see what voltage your motherboard is sending to J1 when an Intel DX4 is installed, and again when an Am5x86 is installed. Set jumpers accordingly for each CPU.

If the DX4 needs 5V on J1 always, and the AMD Am5x86 is N/C (but spec sheet says no voltage to this pin), then I would not cut off the J1 pin from the PGA168 socket, and also not wire it to the buck regulator.

gonzo wrote on 2024-06-20, 10:42:

Can you maybe explain in short words, what is your experience in case of NOT using a Peltier (what are the symptoms)? - when do […]
Show full quote

Can you maybe explain in short words, what is your experience in case of NOT using a Peltier (what are the symptoms)?
- when does the CPU get too hot (instantly; after few minutes; after one hour...)
- can you do something in Windows (some software etc.) until the CPU get too hot
- something else....

Symptoms are the usual set of instabilities - no lights at all, POST hangs, post-BOOT system hangs, software (apps/games) exiting to DOS prompt abruptly, or behaving incorrectly.
The first 3 are the most frequent.

Cannot think of anything special to add about running Windows OS and apps/games in it at 180/200MHz.
It is another use case that has its own set of system requirements - type and quality of components, cooling, electrical supply, etc.

The 486 architecture is very CPU bound, so any user activities will result in increased processor power draw and heat dissipation.
This usually happens seconds after heavy task is started.
Air and water cooling cannot go below the surrounding temperature. Often much higher than it.
Stability can be very different in a hot summer day, or a cold winter one.

Look at subject 2 in this post.

The path to success is much longer with air cooling, often prohibitive.

feipoa wrote on 2024-06-20, 11:40:

I would first check to see what voltage your motherboard is sending to J1 when an Intel DX4 is installed, and again when an Am5x86 is installed. Set jumpers accordingly for each CPU.

If the DX4 needs 5V on J1 always, and the AMD Am5x86 is N/C (but spec sheet says no voltage to this pin), then I would not cut off the J1 pin from the PGA168 socket, and also not wire it to the buck regulator.

Check of J1 on the HOT-433 rev. 1 is done: for both CPUs, the voltage is always 5 Volt (even the CPU-voltage is jumpered at "standard" 3,45 V). Measured on the pin-side of the socket (bottom of the mainboard).

Does this mean, for this board, no connection of J1 to the other Vcc on the interposer is nesessary? I ask once again, because I am not sure about the "role" of this j1-voltage.
If an Intel486DX4 is used in the interposer - is they any reason for connectng of J1 together with the other Vcc-pins to the positive V(out) of buck-genrator (e.g. if the J1-voltage is 3,3 V on another mainboard)?

About grounding of the negative V(out) of the buck-generator to the interposer: will be one (single) connection to one (single) Vss-pin enough?

gonzo, if your motherboard is always sending 5 V to J1, no matter what CPU is installed, then do not cut off the J1 pin from the PGA168 interposer and do not wire J1 to the buck.

I'm not sure how many of the Vcc pins to gang up and wire to the buck; probably as many as you are willing to. I'd think at least 10. Same for the GND pins. If there are any Vcc pins next to a GND, you could also solder an SMD 100 nF filter cap between those pins. Probably size 0805.

DTK PEM-4036Y was discussed multiple times in this thread as being one of the finest 386 motherboards.
It is the finest really ...
Ever since i stumbled upon it and understood its qualities i have been looking for a backup.
Years later found two 4036YB models. They are great on their own, but not quite 4036Y.
Along the way developed lasting interest in Symphony chipsets which lead to the inspection of many other 386/486 motherboards based on them, but never again saw another 4036Y assembly.
In fact, the only 4036Y pictures online are from this thread - not encouraging.

DTK PEM-4030Y

Few weeks ago it shown up on Ebay for very reasonably price. Bought it.
It came in great physical condition but was dysfunctional.
Had a quick sync with Feipoa who has the same model. Wanted to confirm several things beforehand.
The reanimation effort, briefly:
EPROM was corrupted and RTC battery was dead. Fixed them right away.
Everything checked out with traces and voltages.
Oscilloscope returned proper readings, but sometimes the keyboard controller signals were inconsistent. Replaced it with trusted chip.
Board came to life.

Followed a long period of careful testing - the usual grind.
The outcome ?
This board is a mirror of PEM-4036Y in terms of performance, stability, and scalability.
For my great delight.

Everything said about 4036Y applies to 4030Y too, so no need to get into further details.

Looks like you are comparing the silkscreen model name with sticker model name. It is probably best to compare sticker name with sticker name, and silkscreen name with silkscreen name. For the record, the sticker name of this board is the DTK PEM-4030Y and its silkscreen name is the PEM-0030Y. Very confusing.

So the PEM-4030Y and the PEM-4036Y are equivalent in terms of OC performance in every regard?

You have a point. Aligned the names.

To your question - yes. Very satisfying.

The silkscreen name is the actual model of the board PEM-0030Y. The sticker is an update that tells you which CPU it came with from the factory, in this case a 40MHz 386. PEM-4030Y.

Anonymous Coward wrote on 2024-06-23, 07:07:

The silkscreen name is the actual model of the board PEM-0030Y. The sticker is an update that tells you which CPU it came with from the factory, in this case a 40MHz 386. PEM-4030Y.

Do you think the sticker name has anything to do with the FSB stability of the chipsets?

Should we be referring to them only by their silkscreen names?

feipoa wrote on 2024-06-23, 07:27:

Should we be referring to them only by their silkscreen names?

Thinking about this a little more... the sticker name PEM-4036YB and the sticker name PEM-4036Y perform differently w.r.t. overclocking and contain different package IC's on the PCB, yet the silkscreen name on both are PEM-0036Y. This suggests that the sticker name provides more information as to the PCB itself.

So is it this particular motherboard implementation or something about Symphony Labs SL82C460 that makes this board so great? Since SL82C460 is a 486 chipset with VLB support I wonder how it compares to other 486 chipsets.

feipoa wrote on 2024-06-23, 07:27:

Do you think the sticker name has anything to do with the FSB stability of the chipsets?
Should we be referring to them only by their silkscreen names?

Back when chipsets were a new thing, they would stamp maximum rated speed on them. I heard these early chipsets were relatively fragile, and prone to failure which is why they were normally socketted.
If we're talking about stuff from the 80s, then it would be a good idea not to exceed exceed the manufacturer's ratings. After 1991 or so, you'll notice most chipsets were no longer stamped. By that time the technology and yields improved, and you can generally assume these chipsets will work at up to 40MHz (unless otherwise stated in the datasheet). Since the DTK boards in question were likely never used outside of their own products, my guess is that the sticker was some kind of internal marking to tell the assembler which CPU to stick into the socket.

I would go by the silkscreen, not the sticker.

I have a board that had a 33Mhz only sticker on it 386/486 but not sure if that was a chipset limit or because it had 25ns cache on it...

486LC1 board, SiS 82C206 and two Contac chips I can't read on my phone pic, don't seem to have speed grades marked.

The 25/33/40 labeling sounds reasonable.

I have a new in plastic bag DTK PEM-4036YB with tons of seal-labels everywhere - CPU, L2 cache, RAM, EPROM, KBC.
While the label says 40, the CPU is Intel 386DX-33.
It is possible of course the CPU was downgraded by the customer (Toshiba Corporation), but that does not sound very plausible - you get 40MHz chip from the manufacturer to swap with 33MHz one.

@rasz_pl
I assume you don't have the stamina to flip through some of the post linked in the directory on the first page, so here is the gist of it:
After gathering a lot of empirical data i can safely say that Symphony Labs SL82C362/461 is the fastest 386/486 ISA chipset.
Boards based on it vary from intermediate to impressive
The intermediate ones are clock-to-clock faster than anything else, but don't overclock very well. Usually max-out at 45MHz.
The impressive ones (the two in question here) overclock easily to 50-55 MHz.
If you check the combined 386 performance charts, you will see what that means.
Symphony Haydn II and Wagner chipsets support VLB as well, but the implementation is ok at best.

@BitWrangler
Pass a picture please.
386 Contac boards are of great interest here after MSI MS-3124.

pshipkov wrote on 2024-06-23, 19:14:

@BitWrangler
Pass a picture please.
386 Contac boards are of great interest here after MSI MS-3124.

This is a pic I had online already, if you want a clearer one it has to wait until I can get it's box out again, it is the one right in the center...

The one in the middle, right ?
I have a dead one like it here.
It worked for very short while and then fell silent.
Is your working ?