@pshipkov
I can't find multiplier x4. Mobo supports 33MHz, 40MHz, also 50MHz, but don't know ho to switch to x4 multi (4x33,4x40, 4x50? 😀 )...... Also this motherboard sadly doesn't support EDO ram (have 50ns 32MB, but it won't boot with this memory). How sad....

@anonymous coward

Yes, Alaris Leopard LX.

Btw, just ran an online search for "fastest 386".
Have to say that kind of like how the posted above metrics disprove some myths about the subject.

@barry07
Skim through page 7 and 8 in this thread.
there is simple contraption that allows to force level 1 cache in WB mode and 4x multiplier.
But i am sure there is a setting for that on the board. Just need to find it 😀

barry07 wrote on 2023-02-15, 18:54:

@pshipkov
I can't find multiplier x4. Mobo supports 33MHz, 40MHz, also 50MHz, but don't know ho to switch to x4 multi (4x33,4x40, 4x50? 😀 )...... Also this motherboard sadly doesn't support EDO ram (have 50ns 32MB, but it won't boot with this memory). How sad....

Lots of 486/586-boards seems to have problems using 50 ns RAM, regardless of the capacity and regardless of the BIOS-settings.
Some of them can start, but neither Windows can be installed successfull, nor it can be used successfull (if non-50ns-memory is used in advance for the installation itself).
For many years, only one of my boards does work properly with a 32-MB-stick (ZIDA Tomato 4DPS Rev. 1.0).
Interestingly, the Memtest86-test shows no problems for the modules I have (in case a 586-board is booting with them).
Sadly, I don't know exactly the reason for this problem.
What is your experience about this?

72-pin 50ns SIMMs are EDO memory types.
At least i never saw 50ns FPM memory modules. They usually go up to 60ns.
First you need to look at the chipset specifications, but that is not an auto-guarantee by itself.
Board circuitry and/or BIOS (hard to say really which one exactly, or a combination of both) are a big factor as well.
As for Zida Tomato 4DPS - you are not the only person who experienced problems with motherboard.
In fact you achieved much better results with it compared to at least me and Feipoa.

pshipkov wrote on 2023-02-16, 16:01:

As for Zida Tomato 4DPS - you are not the only person who experienced problems with motherboard.

My experiensce above was regarding 50ns-modules only 😉
I meant a 32-MB-stick with 50ns (not 60ns/70ns) on the ZIDA, that's is very successfull 😀
Maybe it was a little bit misleading, sorry...

pshipkov, in preparation for binning SXL2-66 chips for 90 MHz op, I decided to first run the SXL2-80 at higher ISA values. If you are interested in ISA speeds over 10 MHz, here's some additional numbers for your chart.

1DTK PEM-4030Y [Symphony Haydn II, GD5434, 256K]
2SXL2 - 80 MHz at 2x40 and,
3ISA at 13.3 MHz, 4.03 Vrms, then DOOM = 3396 = 22.07 fps
4ISA at 16 MHz, 4.03 Vrms, then DOOM = 3259 = 22.92 fps
5ISA at 20 MHz, 4.03 Vrms, then DOOM = 3155 = 23.67 fps

I noticed that DOOM results are faster when both banks filled. Haydn II supports memory interleaving?

Pretty sure memory interleave is supported. Not many chipsets support both bank interleave and writeback L2, but this one did...that that's part of the reason why it's special.

@gonzo
Gotcha.
I am really surprised you were able to get Zida 4DPS to play nicely. Good job there.

@Feipoa
So +7MHz to ISA results in mere 1.5 fps increase in Doom. Not a lot, but still a +.
Added your numbers. Well, the top score only. Charts are getting too long already, so trying to keep them compact.
One thing that i would like to get solid confirmation from you about is long term stability. I think you mentioned somewhere that things are in the green zone, but cannot remember exactly.
As you know my stance on it - things must be real-real. : )

@Anonymous Coward
Agreed.
On a similar note - i think i finally stumbled onto a good 486 VLB mobo based on Symphony Wagner. An elusive target.

At present, I have only done more extensive testing at 13.3 MHz ISA. 16 MHz checked out in Win 3.11 w/loading IE5 google search. I will run your tests once I bin my SXL2 chips.

Most of the symphony boards with vlb I see come from young Microsystems. They’re usually also DOA.

There is supposedly a DTK vlb board with the symphony chipset. I have literally been looking for it since the mid 90s. I haven’t ever seen a single one…even in the early 2000s when eBay was overflowing with 486 boards. The model was pKM 0038Y. I only ever found PKM 0038S, which was the SIS version.

Was there a reason this combination was rare? Maybe it was flawed somehow.

pshipkov wrote on 2023-01-30, 22:43:

Converted that PC to POD100, but the software setup is still the same. DOS 6.22 […]
Show full quote

Converted that PC to POD100, but the software setup is still the same.
DOS 6.22

config.sys:
DEVICE=C:\...\HIMEM.SYS /TESTMEM:OFF /M:1
DOS=HIGH,UMB (optional)
DEVICEHIGH=C:\...\EIDE2300.SYS /T /M0:8

autoexec.bat - nothing special/relevant to this:
PATH C:\...

I can spin it again in one of the next days to double check.
I remember that user Chadti99 mentioned something about the Quake 1 score too - cannot remember the details anymore.

Hi, a small update from my side.
I am still not sure which graphics card to use for a build based on the Asus VL/I-486SV2GX4.
For DOS benchmarks it is pretty clear what to use, a card with the ARK chipset, it is really great.
I use this Octek branded one:

I found the jumper settings here on Vogons:
"....
2.4 Jumper Setting
JP1 IRQ9 1-2 Disable 2-3 Enable
IRQ9 is used by some applications and is activated when vertical sync signal is generated. Usually, it is disabled.

JP2 VL-Bus Speed 1-2 Normal 2-3 Turbo
VL-VGA-1000 can run at zero wait state. But at 50MHz or on some motherboards, one wait state can be inserted by setting JP2 to (1-2), so that it can run more stable.

JP3 Monitor 1-2 Auto 2-3 Mono
VL-VGA-1000 uses automatic detection for monitor type. In case there is trouble, set it to (2-3) for monochrome monitor.
"
I had JP2 on 1-2, setting it to 2-3 improved for example the Wolf3D benchmark:

My "preliminary values" for 160MHz with this card:
3DBench 1.0c: 105.1
Chris Benchmark: 75.8 Chris Benchmark 640x480: 23.7
PCP: 28.3 PCP 640x480: 11.1
Quake Timedemo 640x480: 18.4
Doom Max details: 1134 ticks => 65.9
Wolf3D 2nd run 151.9

So finally one benchmark where I could beat a value from pshipkov 😉 Wolf 3D is always > 150.
Still behind is my doom value, but I think, I now know why:
It is mentioned here:
https://www.complang.tuwien.ac.at/misc/doombench.html
One of the first sentences:
"...
Note that performance is not only determined by processor, video card, and motherboard but also by the sound system, the memory manager, the mouse driver and the disk performance (e.g. SMARTDRV).
..."
OK, I don't use a soundcard for now, but I use a conventional hard drive and no caching mechanism / smartdrv. So running from a fast Compact flash could probably be the difference here.
Also the 18.8 Quake from pshipkov are outstanding, maybe there is also some disk access during the bench run?
But that's all fine for now. But what I am (again- like before with the Trident 9440 card) struggling is the Windows 98 installation.

First of all I was surprised that there are really no integrated for ARK1000 cards in Windows 98SE (which came out in 1999). Strange.
I tried the ones which are here on Vogons Drivers:
http://vogonsdrivers.com/getfile.php?fileid=2 … 1&menustate=1,0

In principle they work, but somehow not opimal:
I can only choose 256 colours, when I set for example 800x600 16bit my mouse pointer gets stuck.
The card should support also 1024x768 16bit.
I use an 1280x1024 TFT Eizo monitor, which normally should use 60Hz. Have good experience with vintage graphics card with that monitor.
There is also a utility included to set the refresh rate, but it does not work under Windows 98SE.

Any hints about that?
Are the better drivers?
Or shall I switch to Windows 95? Will it work better with Windows 95?
But I have also still a 17" CRT monitor, maybe I should try with that one (?)

Ark1000VL does not work in each and every motherboard if overclocking is at play.
Compared to it S3 Tro64 is miles better at that, but at small perf impact. There is BIOS imposed wait state which helps with the improved reliability.

Very good numbers !
Top dog stuff.
Updated the Wolf3D charts with your best number. Marked it gray for now, indicating peak-perf but at compromised stability.
The wording in your post suggests you are still figuring out stability issues. When you solve the issue will turn it blue.
Intention is to be objective about things.

As for benchmarks - yes, i run them without audio and mouse drivers in the loop.
I use CF cards, but there is no local storage access in the code (gameplay loops) of Doom and Quake, so that won't be a factor.

Win98SE came in ... well ... 98/99.
Ark1000VL came in 1993 and was not a huge seller.
Not surprised it was not properly supported 5-6 years later.
For VLB systems i stick to Windows 3.11 most of the time.

---

@anonymous coward

PKM-0038Y is on the short list but haven't seen one for sell.
30-pin SIMMs and 256Kb level 2 cache only are lowering my expectations, but who knows ...

pshipkov wrote on 2023-02-18, 20:00:

Ark1000VL does not work in each and every motherboard if overclocking is at play. Compared to it S3 Tro64 is miles better at tha […]
Show full quote

Ark1000VL does not work in each and every motherboard if overclocking is at play.
Compared to it S3 Tro64 is miles better at that, but at small perf impact. There is BIOS imposed wait state which helps with the improved reliability.

Very good numbers !
Top dog stuff.
Updated the Wolf3D charts with your best number. Marked it gray for now, indicating peak-perf but at compromised stability.
The wording in your post suggests you are still figuring out stability issues. When you solve the issue will turn it blue.
Intention is to be objective about things.

As for benchmarks - yes, i run them without audio and mouse drivers in the loop.
I use CF cards, but there is no local storage access in the code (gameplay loops) of Doom and Quake, so that won't be a factor.

Win98SE came in ... well ... 98/99.
Ark1000VL came in 1993 and was not a huge seller.
Not surprised it was not properly supported 5-6 years later.
For VLB systems i stick to Windows 3.11 most of the time.

---
.....

Hi again,
I think you are bit off with the release date of the ARK Logic cards. ARK Logic was founded 1993, this card I use is clearly from 1995 (so more or less a late vesa local bus card). Maybe those cards were already available in 1994, but not in 1993.

No stability issues, with 160MHz everything works with fastest settings and lowest wait states.

One good stability test passed, it is the installation of Windows 98SE itsself (a lot more demanding than dos benchmarks 😉
System has a minimal setup: 1 Vesa local bus I/O card (only one IDE controller) with one 6GB harddisk and an IDE DVD drive (not really period correct 🤣), a 3.5" floppy drive, PS/2 mouse connected, the mentioned graphics card and an ISA 3Com 3c509b combo card, that is all.
4 times 16MB FPM 60ns and of course 1024kB L2 cache. Both L1 and L2 write back. And the BIOS with the "L2 write back fix".

But let in on grey for now, it is fine. I did not perform tests like your lightwave 3d under Windows yet. Also this Autocad test was not done yet, also a really good stability test.

But I would really like to use Windows 9x (easy networking, I like to transfer files via FTP from my normal workstation) with 65000 colours.
I keep on searching.
An alternate option would be using a ATI Mach32 VRAM card which I have, too. That is probably nice under Windows 9x, but DOS throughput is really bad in comparison to the ARK card.

I stand corrected.
Ark1000VL seems to came to market sometime in 1994.
Thanks for pointing out.
With that said - i have no rational explanation why it was not covered by standard Windows drivers.
---
Updated the Am5x86 at 160MHz chart for Wolf3D. Your score is now on top in blue !
---
I think you will be better off with the Ark1000 card over any ATIs. For this class machines DOS VGA perf is central.

pshipkov wrote on 2023-02-19, 08:10:

I stand corrected. Ark1000VL seems to came to market sometime in 1994. Thanks for pointing out. With that said - i have no ratio […]
Show full quote

I stand corrected.
Ark1000VL seems to came to market sometime in 1994.
Thanks for pointing out.
With that said - i have no rational explanation why it was not covered by standard Windows drivers.
---
Updated the Am5x86 at 160MHz chart for Wolf3D. Your score is now on top in blue !
---
I think you will be better off with the Ark1000 card over any ATIs. For this class machines DOS VGA perf is central.

I agree with your last sentence, I think there is nothing better for DOS. Windows is basically only used for file transfer via FTP 😀

I now use the Windows 9x driver arkw95_1.29_newer.zip
downloaded from retronn.de
seems to be very new (1997).
I still have to adjust it to my monitor Eizo Flexscan S1911.
More values:
DOS: Autodesk 3D Studio chevy.3ds: 111 seconds
Lightwave 3D: blade (under Windows 98SE) 1390 seconds. Here I am about 100seconds slower than you (SV2GX4 - 160MHz value), I think it has something not only with Windows 98SE but also with disk performance.
I believe that is also the reason, why the Asus PVI-486SP3 is on top of your Lightwave 3D statistics. It seems to have a very good IDE controller.

Wintune 2 under Windows 98SE 800x600 16 bit: 7797kPixel / second.

No stability issues.
I will try to improve Windows graphics, need the correct inf file for the monitor.
So still playing.

The graphics card I should not change anymore, but maybe I can get a faster VLB disk controller and some more compact flash cards....
Then I will re-install everything. I am really curious how much some benchmarks go will go up.

Last thing will be adding a sound card. But honestly spoken, I am not yet sure, if I will put it into a case.
Running the benchmarks is the main point for me, I don't need a sound card for that.

Offline graphics tests - the 3D Studio R3 one is on par - 111 seconds.
The LS3D one is 100 seconds slower. Wonder why ?
It requires ~24Mb RAM to execute the entire process in memory. You are running with 64Mb RAM so there won't be a local storage component during computation.
It can be different Windows version.

A good EIDE VLB controller with the right CF card will bring the PC to a whole new level.
There is a post in this thread with information about that. Take a look.

About sound card.
Having one in place makes the PC feel complete.
If this is not a thing that moves you - skip it.

pshipkov wrote on 2023-02-19, 18:31:

Offline graphics tests - the 3D Studio R3 one is on par - 111 seconds. The LS3D one is 100 seconds slower. Wonder why ? It requi […]
Show full quote

Offline graphics tests - the 3D Studio R3 one is on par - 111 seconds.
The LS3D one is 100 seconds slower. Wonder why ?
It requires ~24Mb RAM to execute the entire process in memory. You are running with 64Mb RAM so there won't be a local storage component during computation.
It can be different Windows version.

A good EIDE VLB controller with the right CF card will bring the PC to a whole new level.
There is a post in this thread with information about that. Take a look.

About sound card.
Having one in place makes the PC feel complete.
If this is not a thing that moves you - skip it.

Hi,
Maybe not very exciting, but this is what I use now:

So I have the possibilty to lower both recovery time and activity time.
I will try it.

These are all right, a bit on the slow side but will do the job for sure.

CoffeeOne wrote on 2023-02-18, 19:00:

I use this Octek branded one:
IMG_20230218_192815.jpg

It's been a while since I compared, but I am pretty sure the image quality of the Octek (while still not great), is much better than the 2TheMax/Bali32, which is complete ass.
I wonder if any other VLB cards using the ARK1000 chipset exist. I'm only aware of the two.

Hercules Stingray and the unidentified model from the first page of this thread. That unidentified model has only 4 RAM chips which i like better compared to the other models 10 chips.
I use 19” inch Dell LCD monitor and so far the Ark1000VL cards offer no different image than most other VGA adapters.
At least didnt notice something glaringly bad to get my attention.