Just a guess, maybe it's one of those original fake cache 486 boards with a modified bios ?

Do you have another bios for it ?

Can we have a photo of the board?

There is probably a jumper block that configures amount of cache on your board and yours is configured for 0KB.

I can't find any information on your motherboard, your best bet is to find the pinout for the chipset and figure out which pins handle cache amount settings, then do continuity checks until you find the jumpers. Then you experiment.

weedeewee wrote on 2021-04-22, 13:43:

Just a guess, maybe it's one of those original fake cache 486 boards with a modified bios ?

Do you have another bios for it ?

Can we have a photo of the board?

Right. To rule out possibilities, I decided to test another CPU in the board. I'd tested with two i486 DX2/66 CPU's. It's my go to "safe" option. I swapped to an Am5x86, and while changing the jumpers around, I found an undocumented jumper... JP30, which had pins 1 and 2 shorted. Since there's no mention of this in the manual, I decided to remove it, and after booting the board up, I popped into Speedsys, and voila... working L2 cache. So I guess JP30 bypasses cache for some reason.

appiah4 wrote on 2021-04-22, 14:12:

There is probably a jumper block that configures amount of cache on your board and yours is configured for 0KB.

The jumpers for cache amount was set correctly. Seems there was an additional jumper to bypass cache for some reason.

JP30 is part of the cache amount jumper I mentioned above then. Using JP30 and the other blocks cache can probably be set to 0, 128, 256 and possibly 512 depending on which pins are open/closed.

I have noticed that several SiS 471 boards have a messed up BIOS when it comes to L2 cache: in many of them the dirty tag is not set properly so WB L2 doesn't work as it should, and in others the cache timing register values set in the chipset are incorrect, resulting in low L2 speeds. This last problem causes Speedsys, Cachechk and other tools to be unable to detect the presence of L2 cache when DRAM timings are tight, as this makes DRAM to run as fast as the L2.

So one thing you can do is lower the main memory speed and see if L2 is detected. It's not a fix, but it could help you diagnose the problem.

EDIT: just noticed that you got the issue fixed! Nice!

TheMobRules wrote on 2021-04-22, 14:16:

I have noticed that several SiS 471 boards have a messed up BIOS when it comes to L2 cache: in many of them the dirty tag is not set properly so WB L2 doesn't work as it should, and in others the cache timing register values set in the chipset are incorrect, resulting in low L2 speeds. This last problem causes Speedsys, Cachechk and other tools to be unable to detect the presence of L2 cache when DRAM timings are tight, as this makes DRAM to run as fast as the L2.

So one thing you can do is lower the main memory speed and see if L2 is detected. It's not a fix, but it could help you diagnose the problem.

EDIT: just noticed that you got the issue fixed! Nice!

My board has traces for an "Alter RAM" as part of the write back caching policy. There's no socket there though, but the motherboard does mention that you supposedly need an Alter RAM for write-back operation. I need to get hold of a desoldering station to scavenge a cache socket from my donor board, so that I can install it on this board. I'd love to test what adding that Alter RAM would do to the performance of the board. I don't really have very high expectations. My main 486 system uses a Legend QDI (Lenovo) board with Opti895 chipset, which so far has been relatively fast and very stable, and since it's from a manufacturer that used to have a good reputation... QDI was used by lots of system builders due to the stability... I don't really wanna swap it out. It's still fun to test out other hardware though 😀

LunarG wrote on 2021-04-22, 14:13:

weedeewee wrote on 2021-04-22, 13:43:

Just a guess, maybe it's one of those original fake cache 486 boards with a modified bios ?

Do you have another bios for it ?

Can we have a photo of the board?

Right. To rule out possibilities, I decided to test another CPU in the board. I'd tested with two i486 DX2/66 CPU's. It's my go to "safe" option. I swapped to an Am5x86, and while changing the jumpers around, I found an undocumented jumper... JP30, which had pins 1 and 2 shorted. Since there's no mention of this in the manual, I decided to remove it, and after booting the board up, I popped into Speedsys, and voila... working L2 cache. So I guess JP30 bypasses cache for some reason.

Glad to hear you've got it fixed !

Out of curiosity... What is the brand/model of your board ?

weedeewee wrote on 2021-04-22, 15:33:

LunarG wrote on 2021-04-22, 14:13:

weedeewee wrote on 2021-04-22, 13:43:

Just a guess, maybe it's one of those original fake cache 486 boards with a modified bios ?

Do you have another bios for it ?

Can we have a photo of the board?

Right. To rule out possibilities, I decided to test another CPU in the board. I'd tested with two i486 DX2/66 CPU's. It's my go to "safe" option. I swapped to an Am5x86, and while changing the jumpers around, I found an undocumented jumper... JP30, which had pins 1 and 2 shorted. Since there's no mention of this in the manual, I decided to remove it, and after booting the board up, I popped into Speedsys, and voila... working L2 cache. So I guess JP30 bypasses cache for some reason.

Glad to hear you've got it fixed !

Out of curiosity... What is the brand/model of your board ?

The SiS471 board is from "Iston Computer Corp." I suspect that's an OEM of some kind that delivered boards to system builders, but who the heck knows. I can't find any useful information about it. I bought the board years ago supposedly "works perfectly" but when I received it, it had a faulty keyboard controller. I replaced that though, so no issues with that. The jumper information link on elhvb.com was unavailable for years, so I couldn't configure it, so I had to go with the settings it had when it arrived, which was for DX2/66. But a while back, the link came back up, so I've wanted to do some testing, but just couldn't find the time to do so. And now when I finally had the time/energy, I ran into the whole cache issue.
Well, fortunately I've now been able to run a whole bunch of benchmarks. All wait states at the tightest timings available, cache set to write-thru with 15ns chips and a single stick of 60ns 16MB RAM, and Speedsys reports memory bandwidth in excess of 100MB/s, which seems pretty damn good. Unfortunately, that's only half the story, since it's still scoring a tiny bit lower in Doom, 3DBench and PC Player Benchmark than my QDI board. Then again, that could be the difference between the Intel DX4 and the AMD DX4 I suppose. I don't want to take the Intel DX4 out of my main 486 just for testing.

Oh, forgot to post a picture of the board as requested.

The plot thickens. It occurred to me that I hadn't actually tried booting with the AMD 5x86 and a jumper in place on JP30 pins 1 and 2. So I did that now, and it's still got functioning cache. So who the heck knows why it's not working for the DX2 then. *shrug*

It's probably not properly configured to operate with a DX2

Can you please send a photo that is not 8bpp and also a picture of the POST string (or preferably a BIOS dump using this http://cd.textfiles.com/microhaus/mhblackbox3 … MORY/GETROM.ZIP) so we can try to identify it and add this to UH19 ?

Deksor wrote on 2021-04-22, 16:56:

It's probably not properly configured to operate with a DX2

Can you please send a photo that is not 8bpp and also a picture of the POST string (or preferably a BIOS dump using this http://cd.textfiles.com/microhaus/mhblackbox3 … MORY/GETROM.ZIP) so we can try to identify it and add this to UH19 ?

The original picture wasn't 8bpp. It seems to have been converted when uploading to the forum. Perhaps due to size limitations? Dunno.

Try this link: https://drive.google.com/file/d/1YvBHykpcvYC1 … iew?usp=sharing

According to the manual, it was set up correctly for DX2. But the manual may be for a different revision board. As mentioned, it's not the most high profile motherboard, and so information has been sparse. When I change settings for these old boards, I draw out my own jumper diagrams, fill them in according to the manual, and double check both my diagram before changing any of the jumpers, and on the board after changing. However, there are several types of Intel DX2 listed in the manual. There is the standard i486DX/DX2, the P24D, the P24S or"SL-Enhanced 486 DX2." Information on identifying each individual model that is available online is incomplete, so sometimes one has to resort to guesswork, or simply choose one. Now that I've at least been able to confirm that there's working cache, I will spend some time over the weekend trying to see if I can get it working properly with the DX2's as well, although it's not that important, as I'm more than happy to do all future benchmarking with the AMD 5x86.

UPDATE: Right. My DX2's are both the "SL-Enhanced" version with power management features. Gives me something to work with. I'll do a BIOS dump tomorrow and I'll take a picture of the POST string. It's getting late 😀

One thing I'm wondering about is : what if the problem with cache was not L2 cache but L1 cache actually ? The 5x86 supports both Write-Through and Write-Back cache. But That DX2 might only support Write-Through ? Causing the board to fail to POST 😀

Thanks for the photo, we'll add this board soon. A BIOS dump is welcome as well :p

Deksor wrote on 2021-04-22, 20:06:

One thing I'm wondering about is : what if the problem with cache was not L2 cache but L1 cache actually ? The 5x86 supports both Write-Through and Write-Back cache. But That DX2 might only support Write-Through ? Causing the board to fail to POST 😀

Thanks for the photo, we'll add this board soon. A BIOS dump is welcome as well :p

I mean, the system did boot, and it ran fine, and Cachechk reported that the L1 cache was working, but that there was only L1 cache in the system. I've done all I can be arsed today.
The board has a pretty nice layout and it's actually got better jumper layout than at least one of my other boards, so it's good to see it working properly(ish) with the AMD 5x86. The manual doesn't have any jumper settings for running 133MHz unfortunately.

LunarG wrote on 2021-04-22, 20:18:

The manual doesn't have any jumper settings for running 133MHz unfortunately.

I can't read the label of the chip that is in the top right corner on the photo, nearest to the label OSC2. If we can find the datasheet on that, I'm guessing, it will provide some insight on the possible frequencies that can be set.

weedeewee wrote on 2021-04-22, 20:43:

LunarG wrote on 2021-04-22, 20:18:

The manual doesn't have any jumper settings for running 133MHz unfortunately.

I can't read the label of the chip that is in the top right corner on the photo, nearest to the label OSC2. If we can find the datasheet on that, I'm guessing, it will provide some insight on the possible frequencies that can be set.

Would the multiplier depend on the oscillator? The board has settings for 25, 33, 40 and 50MHz.
It's labelled KDN-0147C. According to the manual it could be either "KDS-0147x" or "AV9107-xx" is located in U25 or "MX8315xx" in location U26.
I can't find the datasheet for KDS-0147C, but I did find one for AV9107.
https://drive.google.com/file/d/1fM38kLzAVrlY … iew?usp=sharing
I'll see if I can do a BIOS dump as soon as my coffee is done. 😀

Never tried GETROM before, so I hope I did it correctly. Ran both the AT and XT executables just to be on the safe side.

The POST string is as follows:

(C) American Megatrends Inc.,
40-PF02-001630-00101111-072594-SIS471G-F

I also did a BIOS dump using NSSI, just because...