Yeah, the chips are socketed. W24M512 is the model number on 'em, according to datasheets they're 64K x 8, plus an AA26256AK 32K x 8 TAG chip. Though, looking up the model number on the TAG chip, I did find an articleindicating it could be fake. (Wayback machine link, since the site now seems to have a paywall.)

So... say the chips are fake, or bad, or whatever. I have a bunch of 61256's that I know to be genuine... will those plug in and work in place of a 32-pin 24512? The pinout looks similar, just with the smaller chip leaving two rows open in the socket. And what about the TAG chip, would I just use another 61256 there, or does it require a specialized chip?

Old Thrashbarg wrote:

So... say the chips are fake, or bad, or whatever.

That's interesting. I tested a PC-Chips 486 PCI board with fake cache. Inside the BIOS setup area, I maxed out its settings - one of which was setting the RAM to "EDO" which probably helped a bit, and then ran speedsys. I got a perfectly acceptable score of approx 24.something using a DX2-66. My point is, what do you get when you run benchmark utilities using your AMD DX4-100 chip? Are the values obtained from benchmarking falling roughly within an acceptable range, or do they seem a bit too low? If they are OK, then perhaps you've got less to worry about than you first thought? - Perhaps you're "good to go"? 😀

Well, the CPU score was 37 point something, and total memory bandwidth was around 100MB/sec, with around 30MB/sec on the main memory... which I suppose isn't bad, but it still bothers me that it's a crippled board.

Since the cache chips are socketed anyway, and they do have traces leading to the chipset, it should be possible to install some working cache into it, whether it's the chips I already have spare, or otherwise. The main sticking point might actually be the BIOS... if it's done in such a way that the cache is permanently disabled regardless of the BIOS setting, I'd have to find a way around that... either swap the BIOS for one from a different board (if I could find one that'd work), or modify the existing one.

EDIT: Well, I'll be damned. I swapped out the cache chips for some spares I had laying around, and now I have 128KB of functional L2 cache. (The chips I had were 32kb x 8, not 64kb x 8... but 128KB is better than 0KB.) Interestingly, the BIOS still reports 256KB installed, so they did indeed doctor the software as well.

Old Thrashbarg wrote:

EDIT: Well, I'll be damned. I swapped out the cache chips for some spares I had laying around, and now I have 128KB of functional L2 cache. (The chips I had were 32kb x 8, not 64kb x 8... but 128KB is better than 0KB.) Interestingly, the BIOS still reports 256KB installed, so they did indeed doctor the software as well.

Ah so this board exhibits the same kind of behaviour as a PC-Chips fake cache board. It's great you can upgrade the fake cache to normal cache. But what would be very interesting to learn is exactly what improvements you get when you run a series of benchmarks - with and without the cache upgrade. It would be great to see some kind of comparison table, for example. If possible, please can you run several benchmarks, including all the old favourites mentioned in Vogons - speedsys, pcpbench, doom -timedemo, etc! 😀

One of my PC-Chips fake cache boards has a fake COAST cache stick. I wonder if I could remove it, and replace it with a proper COAST stick?

I'm pretty sure the PCChips cache sticks are non-standard, at least on their 486 boards. The slot may look the same, may even physically accept a COAST, but I think the only thing that will "work" in them is the official PCChips fake COAST stick.

I'll run some tests on my system once I figure out how all the various cache-related BIOS and jumper settings work... I got it enabled and working but haven't done anything toward optimizations yet.

Old Thrashbarg wrote:

.....
EDIT: Well, I'll be damned. I swapped out the cache chips for some spares I had laying around, and now I have 128KB of functional L2 cache. (The chips I had were 32kb x 8, not 64kb x 8... but 128KB is better than 0KB.) Interestingly, the BIOS still reports 256KB installed, so they did indeed doctor the software as well.

Good to see you've gotten one step further 😉

And about the tag chip, afaik the tag chips are just ordinary sram chips (not sure on the realy small tag chips we see on some older boards).
The BIOS may report 256KB even though now it has 128KB because you need to set the amount of cache with jumpers.

Theres no way I can tell that the board was manufactured by PC-Chips (even though they had the habbit of making it a real challenge to ID their boards) but that doesn't mean the computer company that originally sold this board didn't put fake cache chips on it himself.

Still quite awkward that VLB boards were manufactured in 96. Nvm 🤣, must be me 😜

I suspect you're right about a retailer adding his own fake chips, as the board seems to be quite good quality otherwise... not really the type of thing I would expect to cut corners like that.

I still haven't figured out how to set the cache size with jumpers, if that's indeed what's going on. There are only two jumpers near the cache chips: the one right below the tag chip has to be open, otherwise DOS won't load, and the one up next to the top cache chip reduces the bandwidth numbers a bit when it's open. What exactly they do, I dunno. (Remember, I'm working blind, here. 😁 )

Here's a quick Speedsys test of the current state of things:
[Read/Write/Move/Average]
Cache Level 1...94.41 MB/s...31.25 MB/s...121.59 MB/s....82.42 MB/s
Cache Level 2...36.50 MB/s...31.16 MB/s....23.22 MB/s.....30.29 MB/s
Memory...........31.45 MB/s...31.67 MB/s....13.46 MB/s.....25.52 MB/s

And a screenshot attached. Something's still a bit off... The L2 read latency setting is at 3-2-2-2, the slowest of 3 choices, which can't help anything, but that's the only setting I could get to work. As for why the L1 'move' speed is faster than the read speed, I have no idea... that's really strange.

Old Thrashbarg wrote:

I'm pretty sure the PCChips cache sticks are non-standard, at least on their 486 boards. The slot may look the same, may even physically accept a COAST, but I think the only thing that will "work" in them is the official PCChips fake COAST stick.

Attention all connoisseurs of sleaze! PC-Chips 486 mobo spotted on ebay:

http://cgi.ebay.co.uk/ws/eBayISAPI.dll?ViewIt … em=190395601841

This one looks particularly nasty, because it looks like there's 2 different types of fake cache. The twin chips in the top left hand corner, plus the COAST stick to the left of the CPU. Unbelievable!

I wonder if the DIMM slots on this UMC powered mobo can accept EDO RAM? I know my later revision board does. However, this one looks like an earlier version, because of the presence of SIMM slots, plus the PCB version number looks like V1.1, or something similar. I think mine is V3.4, or something similar. (Check out the bottom left hand corner - you can just about see the white V letter & numbers.

Gotta love that 486 memory bandwidth. You can see why local buses with 133MB/s theoretical bandwidth aren't exactly going to be saturated on a 486. 😁

That memory bandwidth seems a bit low, though, even for a 486. I don't expect amazing results, but something's not quite right.

I think it's middling but not horrible, believe it or not. My 486 PCI board with Am5x86 @ 40x4 (160) scores about 40MB/s on "memory throughput" in Speedsys when the cache and RAM timings are tweaked to the max.

Well, I'd say the 31MB/sec main memory read speed is about right. And the 36MB/sec L2 read speed seems in line with the crappy timings I'm having to use. But that 13MB/sec move speed... and the L1 is faster at moving than reading? I can't think of any way that would be normal.

It's possible that Speedsys is not working correctly...

swaaye wrote:

It's possible that Speedsys is not working correctly...

I've often thought that.

It's certainly a possibility. Know of any other good general-purpose system test programs that would give a 'second opinion' of sorts? I know of plenty for Windows, but Speedsys is the only noteworthy one I know for DOS...

I'm not sure if this will help, but I thought I'd mention it. I used speedsys on a FIC 486 mobo, and got strange results. Here -

Re: FIC 486-PIO3 mobo (Page 2 - FIC_AMD.png)

A suggestion was to try CTCM.exe; here -

http://motherboards.mbarron.net/download/ (ctcm_int.zip)

Interesting.. I'll have to check that out later.

So I made a couple changes to the configuration... I swapped out the Cirrus card for a #9 GXE64, S3 864-based 2MB VLB card, and I also pulled the finicky AVA-2825 in favor of an AHA-2842A.

I'm having some problems with the new SCSI card, though. I'm trying to get a 9.1GB IBM drive going, by way of an SCA->50 pin adapter board, but the 2842 won't recognize there being a device present. The drive and adapter work fine on a different system with a PCI 2940 card, and the 2842 has no troubles with the couple older narrow-SCSI drives I tried on it. Is there some sort of limitation on these 284x cards that I'm not aware of?

Is the termination set correctly?

It seems to be. I mean, I have termination enabled on the card, and the 'TERM' jumper set on the drive. And the drive worked fine on another card with the same setting. I also tried without the jumper, instead using a cable with an active terminator on the end. Again, worked fine on the other system, but not on the 2842.