They are all looks fake to me. Perform some benchmarks and post the results. Also, you should mention the CPU model.

The manual says that "parameters such as cache and DRAM configurations and ISA bus clock are automatically detected and programmed by the BIOS so that neither jumper nor BIOS settings is necessary"

Yes please do post some screenshots, first time I hear about a half-fake cache possibility! If indeed 128KB are present, my guess is that some chip(s) from one bank is defective and the automated detection falls back to 128KB. You can try to re-sit and swap the chips to see what happens, and of course try new ones.

Interesting, a VIA based board, not much love for those in general it seems.

I have my doubts about half of the chips being fakes. Frankly, the fake ones I've seen always had something obviously suspicious about them like "write back" being printed directly on the SRAM.

That behavior isn't totally out of whack for either a bad SRAM chip or improper jumper configuration of the cache.

The lack of any cache jumper documentation isn't all that unusual. I have an m919 v 1.5 (came with 256k) that the manual mentions possible cache sizes, but any official documentation just says something like "not specified " when it comes to the configuration of any relevant jumpers. Only by searching high and low and carefully studying photos of like boards was I able to deduct how to set them for different sizes. My thoughts and experiences make me think that often chipsets from this era were picky about cache chips, or about specific cache / tag combinations, and possibly as a result manufacturers/suppliers opted to omit any jumper setting references from manuals to discourage users from messing with it and potentially ending up with a board that may not function properly that they may want to return as a result

First we need pictures...

Post some of the cache chips of your motherboard, if they are exactly the same as the theRetroweb one, they are fake,
see this post :

Rectron RT-4S3 fake cache

As I said, the board looks IDENTICAL to the one on the retroweb; Even the dates on the cache chips are exactly the same. If it wasn't for a different number on the BIOS chip, I'd have thought it was the same board!

Both cachechek and Speedsys say there are only 128K of cache - and doesn't seem like the "automatic detection" actually works, I can remove all chips and the boot screen still says 256KB (but both benchmarks report 0, of course).

I'll try to take some pics tomorrow or in the next few days.

Well, it's possible that half of those memory chips are real, if the manufacturer made both variants.

I believe it is possible, but unlikely (but who knows?) ...???

Do you have an SRAM tester? It's the only way to be sure if the chips are good or not.

There is another way to test : with a multimeter, to see if there is any resistance in the circuit, if there is no chip inside, the resistance will be infinite.

This test must be done out of circuit, remove all the sram chips from the motherboard.

Set the multimeter to resistance measurement. (Caution, not in continuity mode...)

Place the red lead on pin 28 and the black lead on pin 14, If you have something between 1 and 5 MegaOhm, we can be sure that there is a silicon chip inside.
With te test leads in the reverse position (pin 28--> black and pin 14-->red) you will have about half the value (1 to 2.5 MegaOhm)

If the measured resistance is infinite (no resistance) the chip are fake.

I measured with a Fluke 87; the values ​​may be different with another model or brand. It's a matter of testing with the multimeter you have on hand.

Then report the results.

Hi H3llR4iser,

I found this post from the user AndrewK2685 , has information that might help you:
Re: Help identifying VIA ISA/VLB 486 mobo

So, I can confirm that the bank 1 chips are fake indeed. This is a new one - especially considering they still have a 256K tag ram chip on the board.

The chips in question are the darker ones, with the 9412 production date code.

I tested the board as is - BIOS default values loaded, all chips installed. As you can see, it detects only 128KB of cache. Speedsys agrees...

Now the fun starts - I removed the chips from bank 1. The board booted no issue...and of course, it still "detects" 256K of cache (which, I suspect, is dictated by the jumper cluster above bank 0, which is undocumented in the manual)
The benchmarks are exactly the same as above.

Next logical test...swap chips from bank 1 to bank 0! Can you guess what happens? (continued in next message as I hit attachment limits!)

--->continued

If your guess was "Cachechek will say there's no L2 cache", congratulations, you win a new screenshot:

Interestingly, Speedsys reports a massively higher memory bandwidth - I'm not sure how reliable that measure is, because a quick 3D Bench shows that the system is indeed slower without the cache (57.3 fps vs 45.5 fps)

And yes - the resistance is infinite between pins 14 and 28 on the suspicious chips. So yeah, these boards have only 128KB of cache, while advertising 256KB. There's some BIOS hocus pocus to make it work regardless - other boards I've played with, would not boot properly if the installed cache was different from the configured size. In fact, the board works perfectly fine with all the cache ripped out, but still enabled.
What's even funnier, is that the tag ram chip is still a 256kb one - which I would imagine costed more than one with less capacity.

By the way, the board also works with this different 128KB kit I have:

However, there is no demonstrable performance improvement over the -20 chips, even from the benchmarks.

Other than that, the board hangs during POST - just after the energy star logo disappears - if I install these 15ns chips together with the original (real...) 20ns ones, regardless of which bank has which. TBH I'm not sure you can mix&match SRAM chips (I just realized I never tried before).

Finally, a couple of pics of the "star of the show":

Last but not least - it's interesting how dishonest the manufacturer were on this one, as I'm fairly sure the "casual omission" of the cache jumper settings from the manual was entirely premeditated. Afterall, who went to check that all of the cache on their motherboard worked, back in 1994?

Yeah, it was very convenient to leave that information out of the manual wasn't it. Seems, the shadier the manufacturer the higher the likelihood that those settings got left out of the manual...

Indeed there shouldn't be any measurable difference between -20 or -15 sram at the same settings, The speed number only tells you how fast it should be able to run before having problems. For it to actually run faster you have to play with the cache timings. Sometimes what happens is the cache is set too slow, by a degree that direct access to ram is just as fast - in these cases boards will often silently default to cacheless operation, even while still detecting said cache during boot up.

Yeah figured it would make zero difference, I was just curious. Also, I've tested some of the setting AndrewK2685 posted that @SDumas linked - indeed, as it's said in that post, the BIOS always shows "256KB" for L2 cache, seems like the string has been hardcoded to hide their mischief. I've now "correctly" configured it for 128KB, just for the sake of it. I should have a kit of 4x256k chips in the post - provided they DO arrive and they're not fake (AliExpress...), I'll be curious to see if this board works with these.

H3llR4iser wrote on Yesterday, 23:40:

Interestingly, Speedsys reports a massively higher memory bandwidth - I'm not sure how reliable that measure is, because a quick 3D Bench shows that the system is indeed slower without the cache (57.3 fps vs 45.5 fps)

Your RAM speeds in cachechk with L2 enabled (100 us/KB) are horrendous--like "oops I left the turbo button off" slow--while your RAM speeds in cachechk with L2 disabled are inline with a 33 MHz bus.

This could either be an "always dirty" cache configuration, or DRAM wait states set really high aka "Slower" or "Slowest" depending on the BIOS/chipset.

Fill the two banks with the 4 GLT7256M08 and the 4 good ones, they are all 32kx8.

Change JS5 and JS6 according to the table below.

Run Speedsys to verify the results.

...........................JS1 ... JS2 ... JS3 ... JS4 ... JS5 ... JS6
1M......................1-2 .. 1-2 .. 2-3 .. 2-3 .. 1-2 .. 2-3
512K.(single).....2-3 .. 1-2 .. 2-3 .. 1-2 .. 1-2 .. 2-3
512K.(dual)........1-2 .. 1-2 .. 2-3 .. 1-2 .. 1-2 .. 2-3
256K.(single).....1-2 .. 2-3 .. 1-2 .. 1-2 .. 1-2 .. 2-3
256K.(dual)........1-2 .. 1-2 .. 1-2 .. 1-2 .. 1-2 .. 2-3 <-- you need this..
128K.(single)......1-2 .. 1-2 .. 1-2 .. 1-2 .. 2-3 .. 1-2 <-- you have this...

I've done all that already...the "darker" 9412 chips are fake. Infinite resistance 14-28, and they don't even work on their own.

If I install the GLT chips together with the four known good chips on the board, it doesn't get past the POST screen, regardless of the jumpers configuration. There's something iffy with the BIOS of the board, it always shows 256KB of ram regardless of what you set on those jumpers (even the original poster of these settings says the same). It's clearly been hardcoded to show that.

I have some DIP32 512kb chips coming in the "future" (shipped from China...), 'cause I'm curious to see if it will actually use 256KB in a single bank configuration as the settings suggest.

jakethompson1 wrote on Today, 03:04:

H3llR4iser wrote on Yesterday, 23:40:

Interestingly, Speedsys reports a massively higher memory bandwidth - I'm not sure how reliable that measure is, because a quick 3D Bench shows that the system is indeed slower without the cache (57.3 fps vs 45.5 fps)

Your RAM speeds in cachechk with L2 enabled (100 us/KB) are horrendous--like "oops I left the turbo button off" slow--while your RAM speeds in cachechk with L2 disabled are inline with a 33 MHz bus.

This could either be an "always dirty" cache configuration, or DRAM wait states set really high aka "Slower" or "Slowest" depending on the BIOS/chipset.

I know - the BIOS doesn't have much in terms of DRAM settings; It's set as "Fastest" with 0 WS. I initially thought it was the case that the board needed the turbo jumper to be closed, but then in other tests, the speed more or less checks out, such as 57/58 fps in 3D Benchmark. Now, there IS a strange situation where just the Doom test is quite slow for a DX4/100, something like 29fps.

For reference, the exact same CPU, on a cacheless (well, fake cache - I seem to have a knack for those!) PCChip M912 does the same 58 fps in 3D Benchmark, but around 34 fps in doom. I'm guessing it might be related to the RAM speed issue.
I've checked with other ram sticks - the speeds don't change at all.

If you check these posts...
Re: Help identifying VIA ISA/VLB 486 mobo
Very slow 486 board - FIC 486-KVD

The results check out. I'm guessing there is a reason why those VIA boards are unpopular. Now, that said...this is us looking at those with 21st century eyes. Back then...who would have actually noticed a 2fps difference in DOOM?

It's also possible you have write through or 7+1 tag mode available for the external cache as hidden options, either of which should give that RAM speed a boost.

jakethompson1 wrote on Today, 21:10:

It's also possible you have write through or 7+1 tag mode available for the external cache as hidden options, either of which should give that RAM speed a boost.

The cache is set as write back; The BIOS has no 7+1 setting available (first time I see that not being an option), haven't tried anything in terms of hidden options. These VIA boards appear to be just slow - check the linked threads.
Funnily enough, if I set the cache as Writhe-Through, the ram speed increases in cachecheck/speedsys, but in 3D Bench and Doom, the overall real performance drops.

H3llR4iser wrote on Today, 22:00:

The BIOS has no 7+1 setting available (first time I see that not being an option)

The UMC 481, 491, and SiS 460/461 are like that. Anyway, if someone figures out the pinout on this VIA chipset (e.g., by tracing a board that provides a dirty RAM socket) and wants to modify a board, they could add the results to my thread here: UM481/UM491 "Always Dirty" modification HOWTO