I have a 486 motherboard that officially supports up to 128Mb fpm or edo sticks and those generally have 16 chips which would mean 8Mb per chip.
All the older 486 motherboards seem to only officially support up to 2Mb per chip ones.

32mb FPM sticks (16 chips) will work fine practically on anything.

I had or have a few 486 motherboards that supported a specific amount of RAM:

Past boards:
PCChips M912 v1.7 w/ real L2 cache: up to 64MB (either SIMM-30, or SIMM-72) EDO or FPM
Shuttle Hot 443 (PCI + ISA): up to 256MB EDO or FPM SIMM-72

Current board:
Packard Bell PB450M (Micro Firmware 4.05.10 BIOS): up to 128MB SIMM-72 FPM (not sure if EDO memory is supported now or not). --> currently has 32MB SIMM-72 installed.

The Serpent Rider wrote:

32mb FPM sticks (16 chips) will work fine practically on anything.

They don't work even on some i430VX motherboards I have (recognized as 16MB); I even have 1 sided, 2 chip 16MB EDO sticks that register as only 8MB.. Is it really this simple?

I even have 1 sided, 2 chip 16MB EDO

Here is your problem. Those chips were used in 128mb EDO sticks.

They don't work even on some i430VX motherboards

I found it very hard to believe, because practically all of them mention at least 32mb sticks in the manual.

Back in about 1998 or so, I upgraded our family 486 to 64MB (2x32MB), which is the maximum it supported (Even though it had four slots, no more than two 32MB modules could be used, the motherboard manual specifically stated this). It had a QDI V4P895GRN/SMT motherboard running the Opti 895 chipset, and I'm pretty sure (but can't be absolutely certain) that I was using EDO for the upgrade, since those exact same modules eventually found their way into my MediaGX-based Compaq a few years later.

I once had a 486 board (with UMC chipset iirc, a very late PCI board) that I was able to upgrade to 256MB RAM with some huge 128MB FP SIMMs (32 chips per SIMM). I put a Pentium Overdrive onto it and installed Windows XP. Because I could 😉

Thing is, it's not the CPU but the chipset that determines both max chip density, max depth and max number of banks. Also the BIOS can limit stuff. This 486 board could handle more ram than most early Pentium chipsets, but many cannot. I've been playing around with my current 486 boards with Via 486A and SiS 497 chipsets. Neither seems to support more than 64MB, at least I can't get 64MB SIMMs working on them. Biggest they accept is 32MB with 16 chips, so 16Mb chips (4Mx4).

dionb wrote:

I once had a 486 board (with UMC chipset iirc, a very late PCI board) that I was able to upgrade to 256MB RAM with some huge 128MB FP SIMMs (32 chips per SIMM). I put a Pentium Overdrive onto it and installed Windows XP. Because I could 😉

Thing is, it's not the CPU but the chipset that determines both max chip density, max depth and max number of banks. Also the BIOS can limit stuff. This 486 board could handle more ram than most early Pentium chipsets, but many cannot. I've been playing around with my current 486 boards with Via 486A and SiS 497 chipsets. Neither seems to support more than 64MB, at least I can't get 64MB SIMMs working on them. Biggest they accept is 32MB with 16 chips, so 16Mb chips (4Mx4).

I think those are bios limitations because manual of my SiS 496/497 motherboard says that it supports up to 128Mb FPM and EDO simms.

I've seen many 486 boards and their manuals and they mostly count on 32MB SIMM sticks, giving 4*32=128MB as the maximum configuration. But I never had any bigger SIMMs so I couldn't test it it can be exceeded. My bet is that most 486 chipsets won't handle them correctly. But as dionb wrote, there must be some that do support more.

Also keep an eye on the amount of on-board cache memory, please.
Normally, the "cacheable area" is between 16 and 64MiB, depending on kind of mainboard/installed cache.
Installing more memory than this may cause a performance drop.
Also note that Win9x/DOS7 Himem and Win311/DOS6 Himem use memory from two different directions,
so this issue might not be apparent at first if you install more memory than what can be hold in cache (see this older thread).