• Please review our updated Terms and Rules here

Bad or Fake L2 Cache in a 386 System

D

dvanaria

Experienced Member
Joined
Apr 11, 2013
Messages
110
Location
Denver
I'm building a system based off an Am386DX-40 with a motherboard (VIP M319P1 TAM/40-P1) which has 64 KB of L2 cache installed on it.

A strange problem: when launching certain programs (like Wolfenstein 3D), the computer immediately reboots (no error screen or warnings).

Through trial and error, I got this to stop happening by disabling the L2 Cache in the BIOS Setup (American Megatrends 11/20/1991).

It looks like I have the cache set up correctly (correct number of chips, jumper settings on the motherboard match the manual for 64 K of L2 cache, cache enabled in BIOS).

Any ideas what the problem is? I've heard of "fake" cache chips on 386 motherboards, and it could be that one of the chips is bad - what's the best way to test this?

IMG_7363.jpgIMG_7367.jpgIMG_7380.jpg
 
You need to find a good RAM diagnostic that will run on your board with the cache disabled. It should also work just as well if the cache is enabled.

I would start by carefully popping and reseating the cache chips (in case some oxide buildup is preventing good conductivity). Be very careful because I have seen a few sockets like those come apart when trying to pull the chips out after many years.
 
Download one of the earlier versions of memtest86 which still supports floppy booting. Test with that. If there is a cache or other memory problem, it should turn up.

Check if there is a newer patch for Wolf 3D. I remember that some software used coding trickery that was incompatible with cache.
 
krebizfan said:
Download one of the earlier versions of memtest86 which still supports floppy booting. Test with that. If there is a cache or other memory problem, it should turn up.
Click to expand...

Ok I've tested with MemTest86+ versions 4.10 and 4.20, with some scary results. I have a bootable floppy to run the test, and the test begins to run but then the computer reboots within seconds after starting. Actually, sometimes it is a few seconds, sometimes 10 seconds - it seems to be random.

So I swapped out all the 30-pin SIMM modules with new ones, thinking it must not be the cache. But.. same problem (the computer restarts). I swapped out my VGA card thinking maybe its some on-board memory there that's causing the problem but... same problem (computer reboot).

I'm not sure what to try next, but I'm having a sinking feeling that there's something wrong with a motherboard component. I've swapped out the RAM, video card, and disabled the cache. I suppose its worth a try to remove the cache entirely just to make sure they are not still being polled by the test, so I think I'll try that next.

Any ideas are welcome! Thanks everyone for your help so far.
 
If possible, go into your BIOS and turn the cache OFF and then rerun your mem test. That's how I found a bad cache on a 386 board last year. P.S. Got my money back on that one.
 
Memtest86+ does the same on most systems I tried it on as well (modern ones). It either freezes before it starts to do something, or the system does a reset after some seconds. Old versions may run better, dunno.

For old PCs, I'm using ctramtest V5.1, which is simply the best ram testing tool ever created. It even copies itself to video ram and runs from there so that it can test the whole memory, not just unused parts. It also checks parity, cache and everything else that is related to memory. The only downside: there's no English version...
 

Attachments

  • ctramtst.zip
    33.2 KB · Views: 1
smeezekitty said:
Memtest86+ 4.x doesn't work on 386/486. Grab version 2.11 which should work
Click to expand...

Thanks for the tip. Version 2.11 does work without restarting the computer, but only if I disable the cache (in the BIOS Setup) and turn off the Turbo button (which I believe disables the cache as well - if it's on, it seems to be engaging the cache even if the BIOS Setup is set to cache=disabled).

I've spent a few days on this - the only conclusion I can really come to is that there is a bad cache chip (or two). I've tried resetting the chips (to make sure it wasn't oxide buildup) as well with no luck.

At this point, the system is stable and runs Wolfenstein3D without restarting, as well as MemTest+ ver 2.11 without restarting (the test reports no errors and doesn't seem to lock the machine up ever). Its just dog slow with the cache and turbo disabled - seems more like a 286 speed-wise now.

Has anyone had any luck ordering new/replacement cache chips for these types of motherboards? I think I need 8 individual "8k x 8" chips and 2 smaller TAG chips that are "4k x 4". Is ebay the best place to look for these?
 
I had a dickens of a time trying to get my 386 and 486 boards upgraded with cache and memory due to defective chips I bought on ebay. The only solution I can say worked for me was to buy more than I needed, so I could exchange chips until I got the working ones. But I DID finally upgrade both boards to 256 k of L2 cache, and maxed out both boards with memory (32 MB and 128 MB respectively.)
 
I think I might try to just find a replacement motherboard that comes with an Am386DX-40. That way I won't have to worry about trying to match up the cache and tag chips, plus there's always the possibility that its not the cache itself but something on the cache management circuitry on my motherboard that is causing the problem. A replacement motherboard would cover both potential problems, assuming the cache works on the replacement.

What a bummer to work for 4 weeks on a system only to find this problem - I should have run the motherboard through some initial tests to begin with - live and learn.
 
You must log in or register to reply here.
Back
Top