Some boards only like tinned lead SIMMs for whatever reason. It is possible these are slightly too thin to make proper contact. One of the chips might be bad.

BTW these are 120nS chips. They might be too slow for use in 386/486 systems.

You suggest I could try with a 286?
Gonna do then 😀

I saw one post of a guy asking for a spare in an Amiga forum. Maybe a machine specific set?

Tried with my 286 mobo and didn't even boot.
The 256kb x 4 that were in were 80 ns 🤣, faster than these 120ns... Are these maybe 80806, 80186 era?

Agree ! Have had many issues with the gold plated 30 pin simms and 120nS are way to slow for even a good 286 unless it is one of those old 6 or 8Mhz versions.
The Creative AWE32 and Goldfinch cards use 80nS or faster 30pin simms.
A lot of XT's used 120nS or 150nS DIP ram so maybe some XT with 30pin? Maybe they were in an old Apple, have some odd 30pin that came from early Macs...
Slowest 30 pin in the spare ram box are 100nS 1Mb 9 chip with label CHP-1MBSIM1-42158PDP and have no clue where I got them 😀
added: those are Mitsubishi dated 1994 were definately for something specific as most 30pin at that time were 70nS or better and 72pin were already quite common..

120ns SIMMs? That's impressive. It's very rare these days to see anything slower than 80ns. The slowest I have ever seen were the ones that came with the XT286 motherboard. I think those are at least 150ns.

Nexxen wrote on 2020-12-19, 18:00:

I have issues with 4 sticks of SIMM, post says 640KB of ram but halts and can't even get to boot memtest.

I'm not sure of what I have, as they should be 1MB sticks. 9 chips maybe is some ECC unsupported by the chipsets?

Those sticks are 256K sticks. They make up for 1MB of RAM. If your 386/486 boards support memory relocation (possibly you need to disable BIOS shadowing), the memory test should count up to 1024K, but without relocation, the 384 KB between 640K and 1024K are only usable as shadow RAM for faster ROM access or maybe as UMBs, provided proper initialization. So the BIOS counting up to 640KB might be expected behaviour.

Furthermore, you need to make sure your memtest is old enough. Newer versions of memtest try to measure memory throughput or CPU clock speed using the CPU integrated cycle-counter that didn't exist before the Pentium, so these versions crash on boot on a 486 processor. According to this thread memtest86 that works for 486 , the last version of memtest working on 486 class processors is "memtest86+ 4.10". Notice the plus, that one is a fork from the original memtest, and you need to go back even more in time to find a working non-plus memtest. And last, but not least, I am unsure whether configurations without any extended memory (so 640KB only) are supported by memtest at all...

These points add to the already mentioned fact that the access time of your modules is astonishingly slow, so you might need to add maximum wait states and maybe even underclock your mainboard to get them working properly in a 386/486 class machine at all. At least (I checked the datasheet to verify) these modules do support page-mode access, which 486-class chipsets might expect from any SIMM you plug into the mainboard.

Don't waste your time with these things in a 386/486s. I think those might be good for a 10-12MHz 286, but that's about it.
I also wouldn't bother with memory relocation. In a 386/486 you are going to lose a tremendous amount of conventional memory with relocation enabled. You need that space to load drivers.

Anonymous Coward wrote on 2020-12-20, 12:46:

Don't waste your time with these things in a 386/486s. I think those might be good for a 10-12MHz 286, but that's about it.

I agree. 120ns memory in a 386 or fast computer makes no sense. But some of us here are trying things just to see whether they are possible at all.

Anonymous Coward wrote on 2020-12-20, 12:46:

I also wouldn't bother with memory relocation. In a 386/486 you are going to lose a tremendous amount of conventional memory with relocation enabled. You need that space to load drivers.

I strongly disagree with this statement, though, although it is partly true. If you enable memory relocation, it makes you unable to use a driver like UMBPCI that re-purposes shadow RAM as UMBs. If you have a board where UMBPCI or UMB_DRVR works on, and you don't need EMM386 for other purposes, by all means leave relocation disabled and use these tools.

My impression is that most VOGONs user do not have these tools set up (if they work well on their computers at all), and get UMBs provided by EMM386 (or a similar memory manager) instead. In that case, enabling relocation does not do any harm at all. Instead, it provides the user with 256KB/384KB more extended memory that can be used by DOS extenders (with 16-bit windows in 286 or 386 mode counting as one) or memory managers to provide it as XMS/EMS.

And in case of just 1MB of RAM (if you are running a 386/486 board on 256KB sticks just for the fun of it), you need to enable memory relocation to be able to use "DOS=HIGH", because you won't have a HMA without...

You're right. I haven't really used UMBPCI, mostly because it came around after people had largely moved on from DOS. When I had Socket7 and Slot 1 systems I had basically zero interest in running DOS on them.
Isn't it somewhat chipset dependent? Will it even work on a 386/486 ISA/VL motherboard?

1MB RAM on a 32-bit system is silly, and probably shouldn't be used for running 32-bit software anyway.

Anonymous Coward wrote on 2020-12-20, 15:52:

Isn't it somewhat chipset dependent? Will it even work on a 386/486 ISA/VL motherboard?

It is very much chipset dependent. UMBPCI is called that way, because it mostly works on PCI-enabled 486/Pentium chipsets. UMB_DRVR is for older chipsets. Not all chipsets support using shadow RAM as regular UMBs. Possible problems include:

• There might be just one master "shadow RAM write enable" control bit, so if you enable writing to shadow RAM at D000-DFFF (to use that area as UMBs), you might also enable writing to the BIOS shadow area at F000-FFFF.
• The shadow RAM might not be targetable by legacy ISA DMA. This will cause problems when trying to transfer data from/to floppy.
• The shadow RAM might not be targetable by ISA busmaster DMA. This will cause problems when trying to transfer data from/to SCSI devices using a controller like the AHA-1542.

As most chipset designers didn't intend the shadow RAM to be used as standard RAM during PC operation, limitations like these are commonplace. They don't hurt the use of that memory as shadow RAM. Because using UMBs this way is a kind of clever hack that often is outside of the design ideas of chipset and mainboard manufacturers, it's not that common, but there are 386/486 VL mainbaords you can use this way.

Catching up, very interesting answers.
I have memtest ver. 2.0.

Sometimes I'd like to know where you learnt all this nice stuff.

Nexxen wrote on 2020-12-26, 02:21:

Catching up, very interesting answers.
I have memtest ver. 2.0.

Sometimes I'd like to know where you learnt all this nice stuff.

A 386 and 486 can run memtest v4.00 from my experience, My 486's will not run v4.37. Any version after 4.0 requires a Pentium or better AFAIK.

Horun wrote on 2020-12-26, 03:40:

A 386 and 486 can run memtest v4.00 from my experience, My 486's will not run v4.37. Any version after 4.0 requires a Pentium or better AFAIK.

Quite spot on, but it is worth pointing out that "V4.00 runs on 386 and 486 processors" is not about memtest86, but memtest86+. memtest86+ 4.10 also runs on 486 processors, I just tested it the other day. There are no significant advantages in V4.10 over V4.00 regarding 386 or 486 systems, so either V4.00 or V4.10 is fine.

mkarcher wrote on 2020-12-26, 08:50:

Horun wrote on 2020-12-26, 03:40:

A 386 and 486 can run memtest v4.00 from my experience, My 486's will not run v4.37. Any version after 4.0 requires a Pentium or better AFAIK.

Quite spot on, but it is worth pointing out that "V4.00 runs on 386 and 486 processors" is not about memtest86, but memtest86+. memtest86+ 4.10 also runs on 486 processors, I just tested it the other day. There are no significant advantages in V4.10 over V4.00 regarding 386 or 486 systems, so either V4.00 or V4.10 is fine.

Thanks yes meant memtest86+. Did not know v4.10 also worked on 386/486.

Anonymous Coward wrote on 2020-12-20, 12:46:

Don't waste your time with these things in a 386/486s. I think those might be good for a 10-12MHz 286, but that's about it.

No . They are odd ball simms. My 286/12 ones are 90ns or better.

Minimum timing is 100ns for 10MHz 286, 12MHz to 25MHz means 80ns.

The 200ns to 120ns were appropriate for 8088/8086 boards.

Get another set of memory at least 80ns for your 386, but 486 really prefers 70ns for best timings.

Cheers,

Possibly for a 8-bit ISA ram expansion board?

The guy that sold them to me had a few Apple II, and sold many parts as spares.

I have a lot of 1MB simms, it's just that I wanted to understand why they didn't work.
Guess problem solved: the slow timings indicate the use for slow machines like 8080/86.

pentiumspeed wrote on 2020-12-26, 17:19:

Minimum timing is 100ns for 10MHz 286, 12MHz to 25MHz means 80ns. […]
Show full quote

Minimum timing is 100ns for 10MHz 286, 12MHz to 25MHz means 80ns.

The 200ns to 120ns were appropriate for 8088/8086 boards.

Get another set of memory at least 80ns for your 386, but 486 really prefers 70ns for best timings.

Cheers,

Fast 286s were sold with 100ns RAM, they just used wait states, which were common practice at the time.
Have you ever seen an original 6MHz IBM AT? I could be remembering wrong, but I think those used 250ns DIP RAMs.