I have slowly also started to save the BIOS images of my older motherboards and even my older ISA graphics cards.

My original 486 motherboard (that I bought back in 1993) had a BIOS that became corrupt and it took me a while to find a replacement BIOS. So, although most EPROM chips do last quite a while, there can be the odd occasion where one does fail.

However, if you have a 286 or higher PC, then you can also try Navrátil System Information (NSSI 0.60) to save the BIOS image of both the motherboard and the graphics card. Thereby, bypassing the need to physically remove the EPROM chip.

For my XT motherboards, I've been using my TL866CS Pro.

I looked at NSSI and while it dumped the video and mobo bios correctly, it made a single file. My mobo bios and videocard bios each are two chips each. HI byte and LO byte I guess. Interleaved or accessed in parallel for faster speed. I would somehow need to split the images into "A" and "B" chips.

So I just made an image of each chip directly.

Now if your BIOS chip is a single-chip then nssi would work straight away.

Even single chips can be internaly interleaved. My Trident TVGA8900D is one such example, the NSSI dump is just a linear image but the actual ROM data has even bytes at +0x0000 and odd ones at +0x4000. Seems like emulators too expect a linear image so just because a dump works in one doesn't mean it's usable as-is for burning a ROM replacement chip.

Also, dumping mobo BIOS via NSSI will only work up to 64k images (so many 486+ boards will not be dumped correctly), and only if the ROM shadow is OFF. And even then the chipset might be internally mapping some SRAM onto it and allow writes to store some BIOS config values. So the dumped image will have wrong checksum and fail to boot. Guess how I know.

Yes.

While I was dumping EPROMS from my 486 there were instances where I got different checksums from the same chip. Reseating the chips in the programmer's zif socket didn't fix the issue. Are these chips going bad? And what's the right image/checksum? Repositioning the chip in the zif didn't do much except yield yet another different checkum.

Turns out that dual-wipe sockets read from 2 surfaces of the pins. That's fine. And these surfaces are the ones you'd typically clean. But most all programmers read from the 2 sides NOT EVER touched by the socket contacts. The edges are 90 degrees apart. Perfect "hiding place" for long-term oxide buildup. Especially over 25+ years. These edges never see and friction/cleaning action from insertion and removal.

So. Solution is simple, you must clean the sides of the pins which are never touched by the socket contacts. After doing that each BIOS chip I was dumping read reliably 10x in a row, with or without a little wiggling and jiggling and repositioning.

Summary:
IC sockets contact one area of the chip's pins, but 'prommers contact a different area.
Therefore ALL 4 sides of each pin must be clean in order for your system to work and for your 'prommer to grab data.

I'm willing to bet that many bad dumps over the years happened because this issue wasn't addressed. Not so much with soldered-in chips, those are clean on all 4 sides after removal usually. But I'd imagine arcade cabs are ripe for letting muck and dirt build up. If just pulling the chip and ripping it with the 'prommer, you're reading through decades of oxide accumulation, and dust. Unless you clean all 4 sides of each pin.

TRIVIA: Reading an EPROM with a bright light shining into the window will result in a massively bad read.