Amazing, great job!

Don't know of any specific tools for L2, sure you know of all the standard ones. Speedsys should show a graph of the cache, cachchk7 will show in txt, WinTune 2.0 looks pretty if you want to use Windows.

Looking forward to seeing the results posted.

so whats the main goal here? any performance comparison to show off?

I have to hand it to ya... that is a great, wilful and persevering job you did there, I like it!

I love modding 386 era hardware as and when I need, I can relate.

Crafty goodness! 😁

mrau wrote:

so whats the main goal here? any performance comparison to show off?

I never had a 386 with that much cache. I will compare it to an identical one that has only 128K, but I will have to wait until July to do that as I have several PSUs (and other parts) waiting in Poland for me which I will bring back, somehow.

i see, so you got some plans there, powodzenia 😀

i wondered a bit about modding too, i was thinking to what extent those now slow as a fly in tar cpus could be used years later if we improved their supporting hardware and used adequate software;

Now that it's the middle of winter, and my summer car (and project car) are firmly stuck in storage (due to snow), I decided to revive the 486DLC project.

I started by removing the shit work I did on the tag chip, carefully desoldering each lead and then putting in an IC socket first, THEN putting in a new TAG chip (actually, I bought 10, and just used the spare).

First shot, stable system operation.

This looks a hell of a lot better, and less messy:

Also, it works a lot better 😁

Nice work, and I appreciate the effort of replacing components on the board... but I've been running 256KB cache in my 386 for over 7 years now.....

386DX40 build

As I'm only running 8MB of RAM I don't see any performance or benchmark increase at all from 128KB, but its always nice to max out your system.
I also have 1MB of L2 cache in my 486 but again, I don't see any difference from 256KB.

BTW - here is a utility to test and benchmark L1 and L2 cache: ftp://69.43.38.172/mirrors/cd.textfiles.com/c … ty/cachchk5.zip
It also includes a utility to enable/disable L2 cache but IIRC it only works on 486 or better.

Sure, a few % can't be seen but it can be measured in benchmarks. Doubling the cache size usually gains around 3-5% - depends on the app used.

That's a nice, clean Cyrix chip you have there.

Thanks! I have the matching co-pro, 83C87, which the POST summary doesn't see but any and all diags report it and it passes tests.

I don't get it. That's a PC Chips M321, isn't it? Why any soldering, that board supports 256 kb L2 cache without modification..?

M321 apparently came in many flavours - Here's the original one I removed from the system:

The one I had to mod to get 256K was an eBay special, with 64K initially. What I liked about this board is that ALL the DIPs (except the TAG) were socketed, meaning I only had to unsolder the TAG chip vs the TAG chip plus 4 cache chips if I were to do that on the original board pictured above.

I suppose I could have just stuck in an extra 4 256K modules here, but the speeds wouldn't match up (I have 15ns) - I doubt that would make a performance difference, but aesthetically it would - I don't even like mixing RAM speed ratings.

Curiously, the board with 64K had a more advanced BIOS revision (autodetect hard disk, RAM timing, etc) so that one was swapped to the other 256K M321 board.

So, to address possible stability issues, I've added some heatsinks to the CPU (which gets arbitrarily hot) and to the FPU (which remains at room temperature most times).

I've also put in heat sensors into both of them to get a readout. I plan on replacing the LED readout that tells me the speed of the computer with a tiny LED readout that is programmable via arduino. I already have it and it fits in the stock location just fine. I plan on it displaying the speed and +12 +5 rail voltages, then flipping to FPU and CPU temperatures, alternating every 3-4 seconds.

I cut the heatsinks to fit from a larger heatsink I got from ebay.

For now though, I'll run the system with these on and a big ass fan pointed at them with the case cover off to see if it is indeed a thermal event that's locking up my system.

Some thermal paste and:

This JUST clears my AWE32

Jed118 wrote on 2018-05-10, 16:28:

So, to address possible stability issues, I've added some heatsinks to the CPU (which gets arbitrarily hot) and to the FPU (which remains at room temperature most times).

Sorry to threadsurrect-- but did you find it was a thermal event?

Also, some of the Cx486DLC-40s had a problem when run with certain Cx387DX-40s:
http://www.faqs.org/faqs/pc-hardware-faq/chiplist/part2/
"2.30.1 Cyrix Cx486DLC CPU

First generation 40 MHz CPUs had a bug: using a NPX (Cyrix FasMath EMC87 NPX,
Cyrix FasMath Cx83D87 NPX (until November 1991), IIT IIT-3C87 NPX) caused
crashes. These are caused by synchronisation errors in FSAVE and FSTOR
instructions. Later, improved CPUs have an AB prefix printed in the lower
right corner. The Cyrix FasMath 387+ NPX (European name for
Cyrix FasMath Cx83D87 NPX from November 1991) causes no trouble when
co-operating with a bad Cyrix Cx486DLC CPU."
Maybe that's the source of the instability.

AlaricD wrote on 2020-12-03, 23:44:

Sorry to threadsurrect-- but did you find it was a thermal event? […]
Show full quote

Jed118 wrote on 2018-05-10, 16:28:

So, to address possible stability issues, I've added some heatsinks to the CPU (which gets arbitrarily hot) and to the FPU (which remains at room temperature most times).

Sorry to threadsurrect-- but did you find it was a thermal event?

Also, some of the Cx486DLC-40s had a problem when run with certain Cx387DX-40s:
http://www.faqs.org/faqs/pc-hardware-faq/chiplist/part2/
"2.30.1 Cyrix Cx486DLC CPU

First generation 40 MHz CPUs had a bug: using a NPX (Cyrix FasMath EMC87 NPX,
Cyrix FasMath Cx83D87 NPX (until November 1991), IIT IIT-3C87 NPX) caused
crashes. These are caused by synchronisation errors in FSAVE and FSTOR
instructions. Later, improved CPUs have an AB prefix printed in the lower
right corner. The Cyrix FasMath 387+ NPX (European name for
Cyrix FasMath Cx83D87 NPX from November 1991) causes no trouble when
co-operating with a bad Cyrix Cx486DLC CPU."
Maybe that's the source of the instability.

Sorry, that motherboard is now in my "dead" pile. It was warped for some time, I guess it finally just quit. I tried replacing the BIOS, and then I did a dumb thing and put the BIOS chip in backwards, and it got really hot and some of the label melted when I turned it back on. I did get a similar BIOS for an SX machine off eBay, plonked it in, and put my analyzer card out, it showed some codes so I put it in the "to repair later when I get good at it and have some spare time" pile (which is looking more and more like an eBay lot of damaged non-posting boards starting at 99 cents auction)

🙁

Ahhh, bummer 🙁
But now we know... the rest of the story