.... and now it works!

What was it you ask? DRAM configuration? Dodgy video card?

Nope.

Turbo button.

This board and the Peaktron PA286-SMT have a configuration jumper for turbo operation. By default the board came with this set to one of the two keyboard shortcut controls. I unjumpered it (reverting it to 'standard' front-panel button operation) and it powered straight on.

I am wondering if using a modern PS/2 keyboard (native, not USB->PS/2) and a PS/2 to DIN connector somehow messes up the turbo control?

I also found a close up image from that Czech site showing the board with 4 SIMM modules installed.. I matched the jumper settings and installed my 4x 1MB modules and it also works fine - detected and counted as normal. I am still persevering to get the 4x 4MB modules working, though it's proving tricky as the same board settings as the Peaktron don't work. I suppose it could be that board had a different BIOS.

Can't seem to enable 0 wait states on this board - it works fine with the jumper open, but as soon as you enable it, the start-up memory test freaks out at random points each time you power on. It may be at 200kb one time, then 2400kb next, or 3891kb, you just don't know.

This is despite using the same 70ns rated 1MB modules that came out of the Headland HT12 board which operated with 0 wait states enabled with those modules for years without trouble.

Hmmm. The annoying thing is that the 4MB modules which I cannot seem to get configured are all 60ns rated and I suspect may get around the issue!

DRAM ram access times vs wait states required is not a strict function of CPU clock, the chipset's memory controller can be relatively slower or faster necessitating faster memory on a different board at the same CPU clock

what clock speed are you running at now? I can't find it mentioned and the picture is too low res to read the oscillator

It's running an AMD 16mhz chip and standard 32mhz oscillator.

Found four 60ns 1MB modules fitted to one of my Transputer interface boards (a Transtech TMB04 with 16x 30pin SIMM slots), I've swapped in the 70ns modules I was using on the 286 and fitted the 60ns sticks from the Transputer and with the zero wait jumper fitted the memory test completes every time now.

The Transputer card should still work as all the other modules were 70ns rated on it. Not sure why these ones were the odd (60ns) ones out. It has been a few years since I populated the RAM on that thing.

It's interesting that this board requires better rated RAM at 0-wait @16MHz than the HT12 board did at the same. I'm wondering what this is going to mean at 20MHz and higher.....

Next steps on this is to fit a DIP socket for the oscillator so that I can run tests with various speeds (looking in my parts box I have 32, 38, 40 and 50MHz oscillators) and several processor brands and speeds (two AMD-16 with different date codes, a Harris-20 and a Harris-25). I expect, based on the 70ns chips not working at 0-wait @ 16MHz, that I may need 50ns chips at 20 or 25MHz.

I really don't want to have to run this with wait states as it will kill the performance.

Only problem is that on this board the oscillator is directly under one of the ISA slots and the back panel. So it's going to need a low profile DIP socket.

Found a memory configuration larger than 4MB which works!

The main jumper SW1 block needs the following jumpers set:

SW1 J5 closed
SW1 J6 closed
SW1 J7 open
SW1 J8 open

According to the silkscreen on the board that configures one bank (J7/J8) for 4MB (J5/J6) memory module sizes. This results in 8MB being detected and verified correctly by the memory test.

This corresponds to the same settings as on the Peaktron PA286-SMT. Note however that having two banks set as 4MB is an unsupported configuration (it would be SW1 J5-J8 as closed, closed, open, closed on the PA286-SMT) and you get memory error beep codes if you attempt to use that setting on this board (PA286-SA1).

Still, 8MB is a huge amount of RAM for a 286.

Bonus, these are my 60ns 4MB modules, so 0-wait still works 😀

megatron-uk wrote on 2021-01-16, 09:07:

Next steps on this is to fit a DIP socket for the oscillator so that I can run tests with various speeds (looking in my parts box I have 32, 38, 40 and 50MHz oscillators) and several processor brands and speeds (two AMD-16 with different date codes, a Harris-20 and a Harris-25). I expect, based on the 70ns chips not working at 0-wait @ 16MHz, that I may need 50ns chips at 20 or 25MHz.

I really don't want to have to run this with wait states as it will kill the performance.

Only problem is that on this board the oscillator is directly under one of the ISA slots and the back panel. So it's going to need a low profile DIP socket.

needing a wait state to run 70ns chips at 16MHz is pretty bad, my 286 runs at 15MHz 0-wait with 80ns memory

First round of benchmarks completed. This is at the standard 16MHz cpu speed with 0-wait enabled, no parity and shadowed video and system BIOS.

- AMD N80L286-16S at stock 16MHz speed via 32MHz clock
- Intel i287XL
- 0 wait state enabled
- system and video BIOS shadowed
- 2x 60ns 4MB SIMM modules
- Trident 9000B VGA (no-name card, 512KB)

Norton Sysinfo CPU 12.1
CheckIt Dhrystones 4042K
CheckIt Whetstones 537.3K
CheckIt Video chars/sec 2913
Landmark CPU 24.67MHz
Landmark FPU 18.70MHz
Landmark Video 2318.49
3DBench 6.1

When I benchmarked my old HT12 board (0-wait, 16MHz) back in 2011, I got a Norton SI speed index of 12.3, so this is within error margins of that board at standard cpu speeds. I don't know how the CheckIt and Landmark scores compare though.

At first glance, it looks like quite a reasonably good set of results, I'd be broadly happy with that as a stock system. I would expect better video results once my ET4000AX arrives.

Next up is removing the AMD processor and fitting a 20MHz Harris and 40MHz oscillator.

Damn. Went to fit the Harris chip and the 40MHz crystal I was going to use has had a leg snapped off whilst in the box of other oscillators.

It works perfectly on the 32MHz crystal, as you would expect (with identical benchmarks results since there is no architectural difference between AMD and Harris 80286 chips), but no-go at 50MHz, which is my next available oscillator can.

I've ordered a 40 and 48MHz crystal to test the 20MHz chip with, as well as a 54MHz crystal so that I can check my Harris 25MHz (still in the post - delayed!) with 50, 54 and 56MHz oscillators.

On further testing I'm finding that this system has unreliable IO (both floppy and CF/hard drive) with the 0-wait setting left enabled.

Just trying to get DOS 5 installed with 0-wait enabled results in random freezes and hangs whilst copying floppy contents to the hard drive / CF card during setup. I've also tried DR-DOS 7 (for FAT32 support) and it proved equally as flaky while doing hard drive IO. I've got the latest XTIDE (r604) release (ide_atl.bin) on a 27c128 EPROM on a 3C509B. It detects cards perfectly.

Disabling 0-wait and it seems more stable, so far.

The benchmarks were done booting direct to a basic DOS 5 boot floppy and running from there. I'm now thinking I may have to go back and re-do the tests and/or rethink the memory configuration; for example try the 'standard' 4x1MB 60ns configuration.

As expected, the performance of the system with 1-wait is dog slow. Norton SI drops from 12.1 to 9.3, Landmark CPU drops from a synthetic 24.6MHz to 18MHz, and so-on and so-on.

Approximately 25% slower across the board, and disk IO is a pathetic 900KB/sec. Compared to my old rock-solid HT12 board, that's no good for me.

I'll try my other SIMM modules, but this board is turning out to be not quite what I wanted.

ARGH!!!!

Same damn thing.

With 0-wait state when doing long directory listings (such as the DOS directory, or something like Norton Utilities 😎 you get random hard lockups. Switch back to 1-wait state and they complete every time; the only pause being the free-size calculation for the drive (slow as expected on a reasonably large partition).

Urgh. This is going to be a waste of time trying to build a speedy 286 with this board.

Recommendations please for another board? My criteria:

- Baby AT format
- Socketed processor
- SIMM sockets
- Minimum 4MB RAM capacity, but hopefully higher
- Chipset which can run at least at 20MHz, but preferably 25MHz
- Minimum 6 ISA slots (at least 5x 16bit)
- Zero-wait state essential!

It would be nice to have hardware EMS support.

So I think that the source of the crashes may actually be coming from having a combination of 0-wait state enabled and system + video BIOS shadowing turned on.

I'm trying various things that were locking up the system previously and without shadowing enabled they seem to be working okay. The DOS and Norton directory listings which repeatedly locked up the system now complete successful (just as they do in 1-wait state mode) with 0-wait and shadowing disabled.

Norton SI is back to the previous performance rating, although I'm a little disappointed with the hard disk rating of ~1MB/sec, compared to my Headland board which was several times that figure (~3MB/sec in my best optimised settings). That said, there's almost nothing going to be IO bottlenecked on a 286 system at 1MB/sec disk IO speeds - it's faster than any drive I would have had at the time I bought my original 286 system new.

I'll continue my testing and see if this configuration seems stable long-term. Any suggestions for soak tests?

Ignore that. One more reboot and then the system failed to load command.com. This just isn't stable at 0-wait at all.

I think you're expecting too much.
If the CPU runs at 16 MHz, but Landmark benchmarks it at 18 MHz, then it's not bad.
Yes, it could be better, but back in the era nobody would complain about such a result.

BTW, Checkit 3.0 memory test with "Quick Memory Test Only: N" and appropriate "Number of Test Passes:" can be really helpful!

Compared to the HT12 board it's horrendously slow, even at the same 16MHz clock speed. The fact that it isn't reliable at all at 16MHz with 0-wait enabled and fast 60ns RAM is pretty poor. The CPU/FPU benchmarks don't get much of a hit at 1-wait of course, because they're largely not RAM or Disk intensive, but anything RAM based just crawls. The difference is night and day.

To run reliably, it needs 0-wait disabling, which then puts it at about 2/3 the speed of the HT12 board. At the same clock.

I wish I could get the HT12 board up and running, but it just won't work past the memory test (keyboard becomes unresponsive and it wants CMOS reset - no battery).

I've done all the testing I can, running anything now from CF/HDD ultimately ends with the system locking up - it's a crap shoot whether it loads from disk or not.

Config 0 in the results is AMD 16MHz with 1-wait.
Config 1 is 16MHz AMD with 0-wait.
Config 2 is 20MHz Harris (clocked at 16MHz) with 0-wait - this CPU was due to then be clocked at 20MHz for further tests. No point doing that now.

HT12 is a good chipset, so is VLSI Scamp. Try to find a different board that uses one of those

Also, CheckIt 3 reports failures on the extended memory test:

maxtherabbit wrote on 2021-01-17, 15:19:

HT12 is a good chipset, so is VLSI Scamp. Try to find a different board that uses one of those

I know about the HT12, mine served me well since around 1991/92!

I've seen this one, it's a HT12 variant. Lots of ISA slots (but still baby AT), no signs of battery damage, at least 3 oscillator pads, including one that looks like an asynchronous FPU clock. Bad points - soldered on 12MHz Siemens processor (I hate replacing these things with PLCC sockets), and 4x SIPP sockets, rather than SIMMs. Another bad point - no reference in the total hardware 99 guide that I can find.... but HT12 boards are usually supporting up to 4MB and (at least) 16MHz.