techweenie wrote on 2021-08-16, 02:18:
The pinouts are very different. I've got Vcc at all four corners of the 93C413 chipset, and that doesn't match the 488 pinout. I don't have continuity from the chipset to any pins in the cache area aside from ground and Vcc. The jumpers and chips in that area all connect to each other on various pins, as well as two more 74F245 ICs between the ISA slots. It's not making any sense to me yet. Perhaps if I start probing some known good boards using 74F245 I'll be able to figure more out.
Oh, I'm sorry I didn't read the thread carefully enough. I didn't notice that the data sheet was for a different chipset model than your chip. Before you get starting to probe other boards, let me lay out the basics for those 74F245 /74F244 chips (in fact, I think, 74F244 makes more sense in that position, you will soon see why):
The 74xx245 (where XX often is one of "LS", "HC", "HCT", "ALS", "F") chips are bidirectional disableble (the engineer says tri-state) bus transceivers. They can be used to amplify (in digital electronics terms, one usually says "buffer") data signals, and to forward data only when required, seperating the parts before and after the chip if not required. Other chips with similar function are the '244, the '373 and the '374. The '373 and '374 also have a kind of memory (called "latch" or "register") that can remember the last input data and continously output that data even if the input is no longer present. Threre also are the '573 and '574 chips that work like the '373 and '374, but have more convenient pinout for bus coupling. All of these chips handle 8-bits at a time. All of them are 20-pin chips.
The board has two of those chips near the cache, so these chips only handle 16 bits. This makes it very unlikely that these chips are used for the data bits, as there are 32 data bits, and either all of them or none of them are buffered. This would require four chips. On the other hand, addressing up to 512KB of cache requires 16 bit to select a cache line (using the address lines A4-A19 from the CPU address bus), which would be a perfect match for 2 chips.
Now for the differences of the chips: The '244 can only forward data in one direction, but is split in two parts of four bit each that can be individually disabled or enabled. On the other hand, the '245 can forward data in both directions, and has an input to to toggle into what direction data should be forwarded. Both the '373/'573 and '374/'574 can only forward data in one direction, but they use a different scheme to tell when live input data should be output and when old input data should be held. The '373/'573 is like a video camera with a "freeze frame" button: As long as you push the button, the output stays frozen at the levels that were output when you started pushing the button. While the button is released, data gets forwarded instantly. On the other and, the '374 is like an electronic photo camera: Everytime you push the button, it takes a snapshot of the input that gets output. It doesn't matter how long you push the button. The snapshot is output until you push the button the next time. In electronics engineering terms, we say the '373 is "level triggered" (i.e. it updates its output always when the button is not pressed), whereas the '374 is "edge triggered" (i.e. it updates it output only the moment when the button changed from "not pressed" to "pressed"). To further go away from the camera/picture analogy: "button pressed" actually means feeding a low level (~0V) to the control pin, and "button not pressed" means feeding a high level (3-5V) to the control pin.
If my guess is correct that the two DIP-20 pins next to the cache are used for address lines of the cache (you can easily probe that), the '245 would make the least sense of all of the possibilities, as address always go from the CPU to the cache and never the other way around, so you don't need bidirectional chips. The '244 and the '373 have a very distinctive pattern which 8 pins are output pins, which is again distinct from the pattern of the '245. So probing the pattern of output pins should tell you whether the pattern is '244-like, '245-like or '373-like. The output pattern of the '374 equals the one of the '373, so in that case, much deeper analysis is required. Also IIRC the output pattern of the '573 and '574 equals the output pattern of the '245 if used in one of the two possible directions.
But oops, please take a moment to re-evaluate the situation:
techweenie wrote on 2021-08-16, 02:18:
I don't have continuity from the chipset to any pins in the cache area aside from ground and Vcc.
techweenie wrote on 2021-08-08, 07:43:Those spots were originally blank with the words "ENG SPARE".
Maybe the silk screen is not lying to you! "ENG SPARE" means that these positions just have Vcc and GND hooked up, and the other pins can be wired after production for tests or hotfixes. Your measurement seems to describe exactly that. It's possible that these chips are not needed for correct cache operation.
Anyway, you should find some path from the cache data pins to the processor data pins (maybe through four '245 chips), you should find some path from the cache data address pins (expect two of them) to the processor address pins (maybe through some '244 chips). You should find some path from the cache output enable (aka gate) and write enable pins to the chipset. You should find a path from the two address pins of the cache that are not connected to the processor to chipset pins. You should find a path from the tag data pins to the chipset. If you found all those paths, no further chips are required.