elianda wrote:

I recently bought a 386 mainboard with an SIS chipset.
The curious thing is that it has 64 kB L2 Cache with 8x MT 5C6408-20 chips and twice the same chip as Tag RAM - and the board runs uncached.

While I'm not exactly familiar with the applied concept of TAG-RAM, shouldn't those chips be of a different kind? From what I gathered TAG-RAM is a completely different chip family (and concept) compared to Cache SRAMs. Maybe setting the jumpers for the TAG-RAM wrong on purpose and/or removing them helps circling in the issue there.

shock__ wrote:

From what I gathered TAG-RAM is a completely different chip family (and concept) compared to Cache SRAMs

I thought these were exactly the same?

Tetrium wrote:

shock__ wrote:

From what I gathered TAG-RAM is a completely different chip family (and concept) compared to Cache SRAMs

I thought these were exactly the same?

Like I said, I'm not exactly familiar there. But from what I gathered cache SRAM saves actual values, while TAG-RAM saves adresses where those values are located (either L2 cache or main RAM). Also according to wikipedia it's a kind of content adressable memory which works pretty much the other way around compared to traditional RAM.

Most of the times in 386/486 boards I've seen, if the stock cache chips are for example 20ns, then the tag chip might be faster(15ns). When I replaced the 20ns chips in some boards with 15ns ones, it worked fine.

I upgraded my 386 board from 128kb to 256kb cache. Originally, it came with a 15ns TAG RAM chip, and the rest were all 20ns. I pulled some cache off a dead 486 motherboard which were all 20ns chips. The TAG RAM chip I used was identical to the other RAM chips.

Generally your TAG RAM chip needs to be a bit faster than the rest of your cache chips, however according to my motherboard manual this wasn't necessary for the bus speeds I was running at- 20ns was sufficient.

Here are some images that might help:
http://mail.lipsia.de/~enigma/forum/vogons/board.jpg
http://mail.lipsia.de/~enigma/forum/vogons/l2cache.jpg
http://mail.lipsia.de/~enigma/forum/vogons/tagram.jpg

And the graphics card, I am looking for a driver:
http://mail.lipsia.de/~enigma/forum/vogons/n9.jpg

To the best of my knowledge, tag ram chips were just regular SRAM. Though it is usually not necessary for the density of the tag chip to be very high (depending on the caching scheme), so it in the early days when the cost of SRAM was very high, it seems a lot of companies used low density SRAM chips for their tags. They had varying amounts of pins, and usually non-standard pinouts. These days it can be very difficult to replace tag rams on boards like that.

Most of the later 386 and 486 boards (not all) just used the standard 28-pin SRAMs for tags. I have a very high end VLB/EISA 486 motherboard that unfortunately used THREE non-standard SRAMs for tag chips. They were all 20ns. Several years later I managed to find 12ns replacements.

In my opinion, you should really keep standard tag rams in mind when buying older boards, especially if you get stuck with 20 or 25ns parts.

The driver for that huge card should be easy to find. It is actually an early accelerated grahphics standard called "TIGA" Texas Instruments Graphics Adapter. It was actually a competitor to 8514/A, and at the time was expected to win the battle for VGA replacement.

TIGA drivers should come with windows 3.0 and 3.1.

I have tried a bit around with the cache rams. Removing them seems to make no difference, except that it freezes when using a memory manager. Though there is no jumper setting for 'no cache'. So I have put them in again...

As for the Tiga card - yes there are drivers that come with Win 3.1, but they require an installed TIGA software package. Especially Win wants that the TIGACD software version 2.05 or later is installed. It drops back to DOS with an error if it isn't installed.

My feeling is that the standard #9 GXi drivers will work with your card. Give this a shot:

http://driver.bandi.co.kr/graphic/number_nine … 31/gxiin115.exe

I was fiddling around with some version from siemens I found, but it was somehow modified already. Your source seems to be the default setup and it correctly detects the card.
The software detects 4 MB VRAM and 1 MB Instruction RAM. This seems to be ok and fit to the chips on the board.
Thanks ac.

The board goes up to 1280x1024@64K color. The Win3.x driver works, but supports only up to 256 colors. Still, very nice.
Also seeing the demo program running accelerated animation in 1280x1024@64K color on this 386 is awesome.
(looks like there have been also some StarTrek fans at #9)

Well the docs say, that there is also an extra Windows 3.x driver disk that has a few more drivers than Win 3.x itself. f.e. tigatc.drv that supports also high color modes.

Found something here:
http://ftp1.us.dell.com/video/
Filename: 9gxi_wn.exe

Anonymous Coward wrote:

Most of the later 386 and 486 boards (not all) just used the standard 28-pin SRAMs for tags. I have a very high end VLB/EISA 486 motherboard that unfortunately used THREE non-standard SRAMs for tag chips. They were all 20ns. Several years later I managed to find 12ns replacements.

Been there ... done that 🙁 Some 68030 based DIY project I was considering of soldering uses 2 TC5588-15 chips for TAGs. Still haven't found those for a decent price yet. I guess that's what got me confused there.

Back on topic: Interchanging chips didn't do anything either? Also what's J14 used for?
Other than that I'd go with what Markk & DonutKing said.

Ok the Tiga part works now very nicely in 1280x1024@64K colors in Win3.x.

For the Cache chips, I do not see an easy solution. If I get hand on another full set of chips, I can try to replace all ICs. (I don't want to take them from a working and setup machine). But it might be aswell the board that has some issues there. If it was really about the cache chips I would have expected something like a freeze if they were defect and enabled.

Ok I found it.

Turbo Switch is the answer....
I was checking with NU sysinfo7 and the score was 25 with turbo off and 41.3 with turbo on, but always 386DX 40 MHz was shown.
This made me a bit curious, because usually for f.e. 8 MHz the score drops much more.

So this board does not clock down, but disables cache and maybe a few other things. As you know Cachechk and also Speedsys show the MHz nicely...
And Speedsys scored at 5.89 CPU which could even be due to no caching. (score now is 6.85 a bit tuned BIOS)

L2 cacheable area is 16 MB.

Cachechk Results are:
39 cached / 53 uncached.
Cache speed 27.9 MB/s
Main Memory Speed 20.7 MB/s (RAS 2T, 2WS Read)
Access times
Read 101 ns
Write 62 ns

So RAS 2T speeds up from ~120 ns to ~100 ns.

in comparison to the results here
3DBENCH CPU Benchmark Database

Wieso einfach wenn's auch schwer geht 😉

Certainly an interesting way of switching the clock speed.
If you still need spare SRAMs let me know ... I have 3 sets of 8 chips each sitting here unused (256k).

I would be much more interested to increase the L2 cacheable area to 32 MB.

shock__ wrote:

Wieso einfach wenn's auch schwer geht 😉

Certainly an interesting way of switching the clock speed.
If you still need spare SRAMs let me know ... I have 3 sets of 8 chips each sitting here unused (256k).

Excuse the thread revival.... 2011!

The SRAM for the card... I presume with there 8 chips of 256Kb, that they are ( 32K x 8 bit ) for each IC... this would be ok for the sockets on the GXiTC card?

If this is the case, and given there are only four sockets on the card, this will imply a maximum accessible 128Kb for the 34082 co-pro.