Are those boards actually making use of EDO DRAM's bandwidth, or just using it as FPM DRAM (with correspondingly lower burst/page bandwidth). Latency should be the same for similar speed grades, so the wait states would also be similar for random (not sequential/burst/page) accesses, while you'd still miss out on the advantages of EDO for burst access. (meaning roughly 1/2 the bandwidth EDO has the potential for -theoretically, FPM is a bit faster than that, but with DRAM controllers limited to 33/40 MHz, timing granularity limits that much more) ie 50 ns FPM DRAM might be just as fast for such a system.

Hmm, actually, running at 40 MHz FSB may be the reason for the 60 ns wait-state problem, since you'd need a 33 MHz bus (or 50, or 66) to produce a 60 ns pulse (2x 33 MHz cycles or 3x 50 MHz or 4x 66 Mhz) whereas 40 MHz would get a bit fiddly and you'd need 3 40 MHz cycles (3x25=75 ns). A 40 MHz bus would need 50 ns latency for 2-clock access times, so using 60 ns RAM would have to drop to 75 ns (the next step down), so you might as well be using 70 ns RAM.

On another note, the Red Hill computer guide/museum lists the 2 motherboards they used heavily with the AM5x86 (and Cyrix).
http://redhill.net.au/b/b-96.html
The SIS based Chaintech 486SPM and VIA Pluto based FIC PIO-3, both 1996 vintage boards, with the PIO-3 being preferred for the AM5x86.

Indeed that's a strange issue to be understood somehow, probably later when the performance riddle is solved. The only board I've tried to the moment that has showed any affinity to EDO was EXP8449 by DataExpert. But this one was not actually tried on with AMD-133, yet.
It had a DX2-66 Intel CPU.. and a bad battery. So it was given for a 2032 vertical battery holder to be soldered on, instead.
it was not used for anything but the standard 33 mhz bus...
The more surprising is the 'need' for 50 ns memory to operate without errors @__@

These Redhill pages look familiar, thanks! 😀 Though VIA was always a bit suspicious to me... VIA-specific bugs and prolonged suffering for its users seem to stand out just as long as VIA exists. Chaintech should be the solution, they were always more reliable.

The board seems fine but was experimented upon. Surely not in the name of horror but getting a PS/2 mouse port working.
http://img17.imageshack.us/img17/7265/dsc02225aw.jpg

Will have to dismantle it or it doesn't fit into AT case. Fortunately all other components are supposed to be OK. Here I cherish a hope to see a good AMD-160 performance at last 😀

But as we know it, SiS was not exactly the very fast 486 chipset.

For the AMD X5-133 + SIS 496/497 + decent VGA I'd expect the following PCPBench scores:

133/33 MHz: about 8 FPS
160/40 MHz: at least 10 FPS

I reckon PCPBench is memory/bus limited at this point, because I get nearly identical scores from an Intel DX4 at 100/120 MHz.

> 133/33 MHz: about 8 FPS

thanks! well... a score of 6 was easily attained on UMC at 120mhz.
So I have to think the other 3 setups are just not using AMD X5-133 properly (though having the proper CPU model shown at boot up.)

What might be the possible reason for mistake, here?
3 boards can't be equally bad and I hardly could misinterpret the jumper manuals in such a consistent way, either.

Just thinking that poor BIOS versions might be the reason. It seems that finding a real good 5x86 board is even harder than could be expected?

5u3 wrote:

For the AMD X5-133 + SIS 496/497 + decent VGA ...

might I ask if that's only for SiS 496? Is UMC going to be faster for any practical purpose? 😕

The numbers above are specific for SIS 496/497 boards. UMC/ALI may be faster if configured right, but not by much.

About the poor performance of your SIS boards - could you post a Speedsys screenshot?

5u3 wrote:

The numbers above are specific for SIS 496/497 boards. UMC/ALI may be faster if configured right, but not by much.

About the poor performance of your SIS boards - could you post a Speedsys screenshot?

Sure ^ ^ just in checking this an interesting fact came up, this LS board seems to like at least some EDO chips.
1 (double sided) module booted but locked up while starting Speedsys, another (also double sided) didn't start at all, but these two double sided 32 mb chips seem to start and pass ~all~ speedtest checks. Neat.

I've never heard this 4-ISA slot LS model had any support for EDO.

http://motherboards.mbarron.net/models/486pci/ls486e.htm

the site expressly states "not EDO"! Surely they aren't taking some variations into account.

Here are the screens, just sorry I did only some photos of the screens by a camera.
not pretty parallel but still very readable.

the 2nd DOS boot screen. On the first screen of BIOS boot up there's the same thing, the CPU is identified correctly.

http://img26.imageshack.us/img26/4765/dsc02228dg.jpg

the Speedsys screen.

http://img41.imageshack.us/img41/9631/dsc02230ar.jpg

Yes it's 66mhz, and both A-Trend boards have it here too. A terrible, inexplicable mistake.
Can anything but a BIOS fix defeat this error? The jumpers were set correctly here and with A-Trend too I suppose.

What happens if you put on the turbo button jumper? What are your cache/memory timings set at? Not all motherboards have the off-jumper default to Turbo mode. I've seen some boards with 3-state turbo buttons.

Some SiS-based boards are equivalently as fast as UMC-based boards.

feipoa wrote:

What happens if you put on the turbo button jumper? What are your cache/memory timings set at? Not all motherboards have the off-jumper default to Turbo mode. I've seen some boards with 3-state turbo buttons.

Some SiS-based boards are equivalently as fast as UMC-based boards.

That has caught me out a number of times, it's worth a check, check, and re-check.

I would check the turbo switch first, but maybe you have a fake 5x86-133. Maybe it's really a DX/4 100.

You should investigate to see if you are indeed running with a 2x multiplier.

Anonymous Coward wrote:

I would check the turbo switch first, but maybe you have a fake 5x86-133. Maybe it's really a DX/4 100.

You should investigate to see if you are indeed running with a 2x multiplier.

If it was a remarked DX/4, that would be a 3x multiplier, not 2x. And, the CPU does appear to be set to 3x in several of those boards. (hence 120 MHz at 40 FSB)

Anonymous Coward wrote:

I would check the turbo switch first, but maybe you have a fake 5x86-133. Maybe it's really a DX/4 100.

You should investigate to see if you are indeed running with a 2x multiplier.

Sure, I started to have such thoughts when the 2nd board displayed the same "curse". Like the ADW cpu is defective, fake, gone cachemoduleless(tm) 😀
After swapping the CPU to ADZ-133 and seeing no improvement... well I had to think of the other jumpers. Yeah, that was the forgotten Turbo Switch!
one should never forget it...
(posted some screens)

Thanks a lot minna-san, everyone!
That was it, I completely forgot about the Turbo Switch. Indeed that was the reason for slow operation.

reminds of a possibility such as "There once was a little hedgehog, one day he forgot how to breathe, and he died..."
Fortunately now that you mention it, the trouble is gone 😄

With the TS connected it's mainly quite nice. Even the first CPU that I've tried reaches the good 160mhz speed. The video bench works great, even with this unplanned EDO memory there are no lockups or any other problems.

Even a medium 486 CPU fan w/aluminum compound seems sufficient and cool with the bench.
http://img221.imageshack.us/img221/7950/dsc02235a.jpg

The performance screens are taken in WT mode.
This way the board has much better memory caching potential (afair SiS 496/97 can cache only 32mb when in WB mode)

Well I surely feel much happier with this 😀 Thanks again everybody!

Heh, wow, that'll do it.

Had you forgotten to connect a turbo button/switch, or just to activate it?

the sis496/496 requires the turbo switch to be short for full memory/cache speed. there's not a jumper on all boards. some other board which lack the jumper have it hard-wired in the "short" position.

while 10.9fps is comparable with other x5 @ 4*40mhz, your memory throughput still sucks. the chipset is able to achieve ~45-47mb/s shown in the "memory throughput" line in speedsys. maybe you can tighten the timings. post your advanced bios settings for more advice.

cheers,
fab

kool kitty89 wrote:

Heh, wow, that'll do it.

Had you forgotten to connect a turbo button/switch, or just to activate it?

Connecting the switch in place as it is on the picture was pretty enough to get it.
Appreciated the pin indicator printed on the switch, the 1st pin marking on the board and even an inscription telling the JP number for the switch... no need to see the manual.

When pressed, it makes it "de-turbo". lits the yellow indicator and halves the speed. I wonder it it could be reversed with this UTT case. (Making full speed when lit).

FGB wrote:

the sis496/496 requires the turbo switch to be short for full memory/cache speed.

it looks that I was reminded of this feature the hard way 😀 Pentiums make one forget about this switch and SiS is not forgiving, at all.
What's nice is the fact it works, quite efficient in slowdown and nothing is hard-wired.
on the other hand ASUS (in making the same SiS chipset board) was said to forgo this switch... and it's a pity. Better have it operational.

The memory used here was a bit of surprise because the Mbarron site specifically states that EDO is not supported. So I left it on defaults to see its general stability while running the video bench, for some 20 minutes. No errors.
I will surely explore it further. Also a Riva128ZX by Diamond was obtained for this 486 box, too. An advanced member here was recommending its great VESA compatibility, also the speed is of interest here (compared to Matrox Mil2).

The advanced settings mainly seem to follow the manual. Mbarron has its pages assorted in a way
http://motherboards.mbarron.net/models/486pci/ls486e.htm
and also each page is packed itnto a .pdf 😵 (effectively preventing ACDsee and similar picture pack viewers from working)

I did away with this uncomfortable facade, making a usual series of .gif pages.
http://www.divshare.com/download/19174114-c15

Sure I san make photos of the DRAM settings screen a bit later too... If you deem some settings to be of particular importance here, please point me to it.

The first thing that comes to my mind when I get to it is the wait states. Good memory setup can work with zero waits and remain stable...
Surely there should be other, also important.

I'll take a look at the BIOS-settings later but the SiS 496/497 does support EDO in the final Revision of the chipset. The northbridge chip is marked with a small "P" and a second letter, the timestamp should state some date after 9650. The chips on my SiS EDO Board are marked "PR" on the 496 chip from 9713 and "NU" on the 497 chip from 9711.

Great thanks for the push to improve memory here... Now the memory subsystem has stopped this vampire-related type of activity. Quite to the contrary, it seems to perform well 😀

the settings screen currently looks like

and the result is

So the memory throughput has gone all the way from 30 MB/s to 39 and 47MB/s at the moment.

What is tragic is the fact that it almost didn't help the 3D performance. The PCP bench was 10.9 and now it's 11.3.
Interestingly, the Diamond Viper Riva 128 was losing a fraction of the frame to Matrox (10.8 vs. 10.9) but with memory throughput only slightly raised they started to show 11.3.
Any further memory throughput increase doesn't seem to have any practical effect here. So the faster type of memory is not needed probably, neither a faster video card. Everything seems to be hindered with the processor power, that is, the lack of it.

We need more CPU power huh... Just where we might get it I wonder 😀