@maxtherabbit

Good stuff. 😁

i held myself few times from getting dipp only 286 boards. especially fast ones like yours.
i will be very interested to see if you can get it to tick well at 20 and especially 25mhz.
Saying this, because the chipset is very demanding towards RAM when running at 25mhz with 0-wait.

The board megatron-uk is getting is 16 rated, but works great at 25mhz with 0-waits.
I already warned him to be ready for prolonged RAM exploratory work if long-term stability is desired.

it's printed on the chips themselves, VL82C20x-yyQC, where yy is either 16 or 20

Yep.

Btw, while at it ...

I was able to achieve long term stability with a 286 VLSI board at 27.5MHz and 0-wait states.
Was thinking to post about it in the coming days, but here is a video showing a boot to couple-of-tests roll with full gear: https://www.petershipkov.com/temp/retro_pc_im … 6_vlsi_27.5.mp4
Somebody has to dust-off that case. Notice the sticker one of the kids taped on it. This PC is named Timmy. 😀

The 2-3 guys who bothered to check my "blog"-thread probably noticed that the best achieved performance until now with the bunch of examined 286 motherboards was with the VLSI ones running at 25MHz and 0-waits. The same boards and few other good runners can tick at 30MHz, but with 1 or more wait state, which results in lower performance.

The improved ventilation that allowed complete stability at 25MHz: Re: 3 (+3 more) retro battle stations
The first post in the same thread shows the perf numbers.
Also, there is a post with upgrade module + Ti486SXL2 CPU running at 60MHz. Technically CPU operations are much faster, but the system is compromised in several ways = paper tiger.

Why i am saying all this ?

I am pretty sure that this particular VLSI chipset can do 30MHz at 0-wait, but there is no memory that can withstand it.
To get it running stably at 25MHz was a challenge.
27.5MHz required memory curation on a whole different level. 😀

I usually stay away from bombastic statements, assuming everything shaking out there, but this can very well be the fastest 286, or at least one of the top pretenders.

One thing that will make finding suitable DIP DRAM a bit easier for me is that I have a Chroma memory tester. It can evaluate and stress test chips at a variety of access times and displays their "true" speed, which is often much faster than their rating.

Sadly I don't have a 25MHz rated CPU, my Harris is only rated to 20MHz so we will have to see if it can handle an overclock as well as the memory issue

Do you have 0-wait writes enabled as well on Timmy or just 0-wait reads?

I'd love a 30/72 pin memory tester - I've a big bag of the stuff that it would be useful to grade by actual speed (and for the 72 pin stuff, EDO/FPM).

I did just order a hot air rework station though, so removing that stubborn 286 surface mount processor should be easier.

Also, the VLSI chipset (201/311) has EMS support built in, according to that site which lists all of the chipset features (I'll find a link in a sec). But does it need a driver to enable it, or is it pure hardware emulation?

megatron-uk wrote on 2021-01-24, 14:55:

Also, the VLSI chipset (201/311) has EMS support built in, according to that site which lists all of the chipset features (I'll find a link in a sec). But does it need a driver to enable it, or is it pure hardware emulation?

it's fake news, just a software emulation - at least with the 200 series

Damn the fake media and their fake news! 😁

So EMS on a VLSI chipset needs either a EMS card, or something like EMM286? Or are you saying that it's like the Headland and the chipset emulates it for you?

megatron-uk wrote on 2021-01-24, 17:00:

Damn the fake media and their fake news! 😁

So EMS on a VLSI chipset needs either a EMS card, or something like EMM286? Or are you saying that it's like the Headland and the chipset emulates it for you?

headland emulated it in hardware AFAIK, but the VLSI driver just sticks a page frame in memory and copies data back and forth to extended - no actual page mapping occurs

Looking at various test results - looks like 0-wait for both read and write.

Memory tester will be helpful.
Ratings on memory chips are inaccurate.
Specific brands of 70ns ones often do better than other 60ies.
Especially true for IBM RAM modules.

Finished the final round of testing for the Eurosoft HT12 board. Replaced the surface mount AMD286 with a PLCC socket and have tested 20 and 25MHz processors.

20MHz (40MHz/2) is solid, as far as I can see - nothing untoward during the testing and it is still happy running at 0-wait with my 60ns RAM.

24MHz (48MHz/2) is hit and miss. I did initially get almost a full run of benchmarks at this speed and with the Trident card (one reboot from all of the tests). Switching to the Tseng card made things much worse; I couldn't complete any VGA benchmarks with it fitted at this speed.

25MHz (50MHz/2) is a no-go. It will power on and POST, but won't boot.

There are some odd results in the VGA speed testing in Comptest at over 16MHz (ie it drops off a cliff), but all the other benchmarks (especially 3DBench and CheckIt) show VGA throughput increasing in line with clock speed.

I also re-ran all of the FPU benchmarks with the FPU clockspeed jumper at both the standard (8MHz) jumper position, as well as at the alternate (14.318MHz) position. As expected the latter shows a significant performance increase as the FPU will actually be running substantially higher internal clock with that input.

Conclusion:
The board runs solidly at 20MHz. I reckon you could get a couple more MHz out of it, but it's stable at that speed. FPU performance at 14.318MHz is brilliant. I'll need to do some captures of Wolfenstein on this setup, as I imagine it will be smoother than my original 16MHz HT12 system which I recorded several years ago:

https://www.youtube.com/watch?v=7lBHxfsCqWw

Damn, those last set of results were without VGA BIOS shadowing, which explains the fall in metrics from the 16MHz results for comptest. I'll need to re-do them.

Caught out by BIOS settings resetting to defaults between runs - since none of my boards have batteries on them at the moment.

Yeah, much better:

Metric / 16MHz 1 wait / 16MHz 0 wait / 20MHz 0 wait
CheckIt Video chars sec VGA 1 / 6816 / 8860 / 11311
CheckIt Video chars sec VGA 2 / 6995 /9166 / 11557
CompTest Video BIOS VGA 1 / 28717 / 37759 / 47537
CompTest Video BIOS VGA 2 / 26995 / 35564 / 45026
CompTest Video Direct VGA 1 / 10204 / 12987 / 16393
CompTest Video Direct VGA 2 / 10989 / 14085 / 17857

Just running through some of the game benchmarks now at the 20MHz, 0-wait config using the Tseng card, since thats the fastest-but-stable configuration of this board.

Wolf3D benchmark for 20Mhz, 0-wait and the Tseng card comes back with 14.9FPS.

Some visual representations of how the Eurosoft HT12 board scales at various clock speeds

(16MHz + 1 wait, 16MHz 0-wait, 20MHz 0-wait, 24Mhz 0-wait, 25MHz + 1 wait)

It's clear that if this was stable at 25MHz 0-wait it would be a damn nice performer, but it just can't reach it.

Good FPU perf in there.

Can you run a wolf3d and F1 tests - some real world stuff.
Synthetic tests are often misleading.

Wolf3d does a consistent 14.8fps at 20MHz 0-wait. I didn't bother testing at 24mhz 0-wait (too unreliable) nor at 25mhz 1-wait (significantly slower than 20mhz 0-wait).

I can't figure out how to get the F1GP benchmark to run 😁

For F1 - hold "o" (the letter key) to get CPU utilization on screen. Wait for a moment until the cars start moving - that's the number.

Okay, I can get the benchmarks to run at 16-1, 16-0 and 20-0 and 25-1, but not at 24-0.

Wolfenstein 3D (FPS) / 9.2 / 11.6 / 14.8 / NA / 11.4
F1GP Bench 1 (CPU %) / 208-227 / 164-181 / 129-140 / NA / 135-151

For the F1GP benchmark I recorded the lowest number seen before the cars start as well as the highest seen as they start moving.

cool graphs.... I love 286... nothing is compatible with it. 😁 , I have a Toshiba T3100e with 80286 12mhz on it and CGA plasma orange screen... unable to start win3/1 on it but its great.... games that run I exit them with reboot , and if it hangs on a program(app) i Just shut it down if (ctrl+alt+del) dose not reboot it.

P.S. here are some low benches from my t3100e
NC sysinfo Disk speed
This computer : 16.6
Disk Index Average 9.5 , Track-to Track 1.78ms Data transfer rate: 2.044,4 Kilobytes/sec

(I will check cpu and report back if you wish , I did only my upgraded hdd)