proposal: the development of a 386SX interposer for cheap TI486SXLC2-G50 CPUs.

[super-sama]

I'm putting this out there as a general call-out to anyone interested in a challenge. simply put, there are a lot of NOS TI486SXLC2-G50 chips that are available on eBay. (listing provided as example, there are others with more available for up to $18 shipped in some cases.) These are 3.3v, with 5v-tolerant I/O. These can't be used in a 5v system without a regulator and I don't know what else. some interposer boards with such a CPU on them have a couple extra chips probably for compatibility reasons, but what they do is beyond the scope of my knowledge.

386SX interposers in general are getting very scarce and people are wanting stupid amounts of money for the ones that are left, especially the ones meant for IBM PS/2 systems like the 50 and 60 models. All of the IBM boards I've seen use the FPU socket for the upgrade, and this may also be because they were originally designed for 286 systems, but I can't tell. From what I understand, these work in a different way to the normal interposer boards out there and are only specifically compatible with IBM PS/2 systems and do not work on any others, so this would not be the target audience of the proposed board.

the proposal is to develop and make available a board which either sockets atop a BQFP100 386SX or compatible clone, or uses the PLCC-68 FPU socket to disable the onboard 386SX and take over, with 5v to 3.3v voltage regulation and whatever else is necessary to get these properly enumerated at boot time.

The datasheet for the TI486 chips are here: http://bitsavers.informatik.uni-stuttgart.de/components/ti/TI486/1994_TI486SXLC_and_TI486SXL_Microprocessors_Reference_Guide.pdf

what the PDF doesn't go into too much detail about is the VCC5 pin, which is present on the 144-pin QFP package chips, meant to provide 5v for I/O. According to another PDF here: https://www.versalogic.com/products/manuals/M4863.pdf the schematic for the SXLC2 calls for pin #45 to be attached to +5? and yet all the TI official PDF says is that pin #45 on the QFP100 package of other models of the TI486 are all NC, so this is likely *just* an SXLC2-G50 feature since it is a 3.3v chip with 5v-tolerant I/O.

These two images here explain where it is in the two PDFs of the conflicting information, but the TI PDF is older and the SXLC2 was still listed as "advance information" in this. the 144-pin QFP has the pin for I/O voltage listed however.


now, I may be 100% in the wrong here, but is something like this doable? I'm almost certain it'd become a multi-layer board, and it likely will not work for everyone since these chips require a utility to flip a register to activate clock doubling and not all boards like that. likewise they also need a utility to enable the 8K of cache on the chip to get the extra speed gains. This would be for people with a 33-50MHz crystal already on their board, a 386SX-16/20/25, and an optional 387SX at 16-25MHz. however I can also see it being a good way to get say, an am386sx-40 onto a board too, with less hassle.

the only problem I see being faced is the connection to the PLCC-68 socket with pins, or finding a snap-on socket for a BFQP-100 chip. the former seems doable, but the latter seems a bit more exotic.

 

[eeguru]

The difficult part, as you calculated, is the physical adaptation. Ironwood makes a lot of the socket adapters that are needed. But they are very expensive. Most hobbyists wont be able to solder the SXLC on to the adapter either. Which means someone would have to buy them from eBay and do it as a commercial en-devour; further increasing costs.

Certainly doable. I'm a bit surprised by all the NOS SXLCs too. But I'm not sure it could be done cheaply.

 

[super-sama]

Certainly doable. I'm a bit surprised by all the NOS SXLCs too. But I'm not sure it could be done cheaply.

yeah, I figured it wouldn't be cheap. this is more of a "hey, these are cheap right now in numbers, and they'd work with a few extra bits, only problem is getting to the end result on the cheap" PSA for anyone that may see this that might be interested in taking it upon themselves to make it happen. I looked into the stuff from Ironwood a little bit back when I first had this idea and I get how expensive that's going to be. but, assuming one can disable the onboard CPU through the FPU socket, and assuming both that one can find pin headers in the right pitch and enlarge the board slightly for through-hole, it should be possible to do without an expensive snap socket for the CPU itself, exactly like IBM's upgrade. the problem from there is getting a PLCC-68 through-hole socket installed and again, minding the size of the board for having at least two sets of through-hole components.

it would bring the cost down considerably if this is a plausible course of action, and the CPUs would yes, still need to be soldered, alongside any other surface-mount components needed to make it work. not everyone can solder surface-mount, so this does put the hypothetical kit or finished product way over the head of the average person that'd want this, and this can be helped a bit that larger parts are still available that one can hold with needlenose pliers while soldering (i.e. tantulum capacitors), but they do rob board space if a small footprint is desired. it would make it less than ideal for use in a laptop for instance where there isn't any space to begin with for such an upgrade.

but, I also found these: https://www.ebay.com/itm/Cyrix-Processor-Cx87SLC-V25Q-New-trays-of-80-pcs-10-trays/293039407955
looks like 80 CX87SLC-V25Q FPUs for $50 in a tray. at the time of this writing they've sold a single tray, but these would also be useful if the PLCC attachment is taken. this would be another surface-mount part to drop on the board if that were the route, and honestly this would be the best since it's basically a miniature FasMath if I'm not mistaken.