With the set I assume is good (because I've been using it for years) it would usually show 256kb, although it intermittently showed 512kb for the 1024kb setting. It's possible that one or more of the chips went bad I suppose. I should probably get a tester.

CMOS settings were of course turned down.

I tried several combinations of the 10ns chips as well. Some of those chips are definitely bad. With one combination the system didn't even turn on.

Save yourself a lot of frustration and your DIP sockets from wearing out, get a tester. I thought you had decided on one some time ago. Even SRAM chips which passed Himem could not fool the stand-alone memory tester. For such chips, Himem would only occasionally throw out an error.

Did Himem indicate bad memory?

I didn't see any himem errors (when I made it that far). I'd have to check the config files to see if I'm even loading it.

I've had a tester picked out for some time. I just haven't gotten around to buying it.

I wouldn't completely rule out the possibility that the jumper settings for the Tyan board on th99 are wrong or that the board wasn't properly tested with 128kx8 SRAMs.

512k and 1024k cache settings solved. The TH99 settings are infact correct. The problem lies in the sockets. This board is designed to take both the 300mil and 400mil variants of SRAMs. Unfortunately, 2 pins on the 300mil sockets are not hooked up!!! I guess because at the time only 400mil 128kx8 SRAMs were available and Tyan figured nobody would ever bother to try the 300mil parts if and when they became available. I bridged the connections between the sockets and now everything works as it should.

I also got around to testing the Asus 486SV1 Award BIOS on this board. It works okay, except it does not recognise the 1024kb cache setting. The actual SV1 motherboard only has sockets allowing up to 512kb. While the Award BIOS itself is capable of supporting 1024kb (I decompiled the ROM), it appears this support is disabled. I have not yet figured out how to re-enable. So if you want to run 1024kb stick with the AMI BIOS for now.

Also, I bought an SRAM tester and tested every single SRAM in my box.

I will need to look into the writeback cache issue again. Apparently the chip I tested in the dirty bit socket was actually 64kx4. I will need a proper 64kx1 chip re test. I found 15ns parts so far. Still looking for 12ns.

Seems like Tyan was a little sloppy with their PCB design.

If you are wanting to use the AWARD BIOS to add PS/2 mouse support, perhaps see if you are successful with this before bothering with changing the BIOS code to accept 1024K. There could be some other instabilities associated with the 486SV1 BIOS that you've yet to uncover.

What was your pass/fail ratio of your 10 ns 128kx8 SRAM chips?

It's certainly not made to the same level of quality as the Super EISA. On the Super EISA they took fewer cost saving measures and socketted just about everything save the 245s.

I got writeback L2 working on this board.

Last time I tested I used a 64kx4 instead of 64kx1 dirty bit. I found a proper 64kx1 on a different board and tested it out.

The scores were basically identical, except for main DRAM which was 5MB/sec faster using write through mode. (using speedsys)

I think I will stay with write through.

Seems write-back cache had a much greater impact with L1 cache instead of L2 cache. I've yet to see an example of L2 cache in write-back mode having much improvement over write-through. I think GLQuake scored 19.3 fps with L2 WB and 19.1 fps with L2 WT.

Looks like you swapped out the AMI Megakey controller for a Holtek on the Super EISA. That and your board appears to be newer as you have a white Socket instead of the blue Overdrive Socket 2. My Tyan board is funky too. The Keyboard controller appears to be bad, which manifests itself by randomly beeping the PC speaker and locking up the machine while playing digital audio via my AWE32, but only if the machine is in real mode! Loading EMM386 hides the problem. Swapping out the keyboard controller fixes it too.

I swapped out the KBC on the SuperEISA to get one that I knew for sure had mouse support. The old one worked fine. As it turned out, the old one had mouse support too. Infact, just about all of them do unless you're using something really ancient.

The KBC on my Tyan was defective just like yours. I replaced it almost immediately.

Unrelated: at some point I will have to make a post about the VL/EISA Pentium system I've been playing with to compare to the 486 boards. In most ways the 486s are better systems,especially when we are talking about L2 cache and DRAM. It appears the L2 cache and DRAM are only using 32-bit datapaths, basically a 486 design. I don't beleive it is doing DRAM interleave either, which is pretty stupid on a system that can go up to 512MB. I just did an oscillator replacement to bump the FSB up to 66MHz for the PODP5V133 I have, but I was kind of hoping to get a second one for an SMP demonstration in NT4 before doing a writeup. What would be better is if I had a Powerleap PL-54CMMX so I could drop in a K6+

The Windbond W83C42 keyboard controllers found on some [many?] 486 boards will not work with PS/2 mice.

As far as I can tell, all AMI "model F" KBCs and onwards will do the mouse support. I haven't tried earlier ones, but in theory anything from 1987 onward (when the PS/2 came out) should have it.

The SuperEISA is back from the dead (for now at least). The best I can tell is that the BIOS and socket both had a little corrosion going on. It was a white coloured residue that's easy to overlook as it is close to the original pin colour (but not shiny at all). I noticed many of my old ROM chips have this problem.

The nice thing about the Super EISA is that the clockgen is socketted and I can easily switch between that or a metal can oscillator. This lets me experiment with non standard frequencies. As I have not been successful at running a Cyrix 5x86 at 2x66 (and don't have a 4X part) I thought it would be interesting to test my chips out with a 44.236MHz crystal using a 3X clock. At 44.236MHz my board is much more stable than at 50MHz, and the memory performance is quite epic. I am getting an average of 57MB/sec in speedsys. After confirming stability of this speed with an AMD 5x86 (3x44.236 = 133MHz) I decided to give my four Cyrix 5x86 chips a good.

I have an IBM 5x86C 100HF, a Cyrix S1R3, and two Cyrix S0R5 chips. The 5x86C and S1R3 were the only two that were semi viable at 133MHz. I am able to complete speedsys, but running demanding 32-bit games will lock the system up within a few minutes. Keep in mind this is at 3.45V. I will attempt to modify my VRM for 3.75V and try again.

I tested amd 5x86 at 4x44.236=180MHz as well, but none of my 3 chips could handle it at 3.45V. I suspect even at increased voltages it will still not happen.

That is epic, but you still need to confirm stability. Try running some graphic benchmarks in Win9x or NT4. The installation of NT4 is another good test.

My best stable speed with the IBM 5x86c-133/2x is:

L1: 191 MB/s
L2: 73 MB/s
RAM: 51 MB/s

Agreed. I will need a larger SCSI drive though. I guess that is probably the next thing on the agenda. I've been talking about getting one for so long but never actually buy...time to take some action. I should have some extra money for frivolous spending later this week...

NJRoadfan wrote:

A plain old stack of Intel DX2/66s, nothing special. They all came with the EISA boards (some with their warranty void if removed stickers in tact!) so I doubt they are write back chips. Heck, I didn't even know they made write back ones.

Yea, they did. I did not know this myself until I opened up an easter egg system I got earlier this year.

Nice little surprize... I will look to test it again soon, I remember when doing initial testing on the (UMC) board, The CPU was a little nippier than a conventional DX2-66.

Anonymous Coward wrote:

What would be better is if I had a Powerleap PL-54CMMX so I could drop in a K6+

So that's what it was for!?

Have you tested a K6 as yet?

This question is best answered in my VL/EISA Pentium thread, but yes I tested it. The results aren't good. There seems to be a BIOS incompatibility. Even with real Pentium chips the L2 cache is automatically disabled. With any kind of K6 chip the system doesn't even POST.