Hey Gonzo,

M919 is clunky but not that clunky.
Something does not feel right there.
I don't use cachecheck so cannot comment on its reliability.
One easy way is to run SpeedSys and see what the memory performance diagram looks like.
You should see 3 cascades - L1, L2 and bare memory speeds.
If you can - post a picture of that.

Feipoa,
V3 on M919 is a no go.
Also, i think M919 damaged one of the Voodoo3 cards here.

---

I consistently see now-a-days M919 listed on ebay for over $300 USD.
Am i missing something there ?
It is true that it overclocks easily to 180MHz and when paired with 1Mb L2 cache becomes one of the fastest things at that frequency (minus its weak IDE controller).
But also, the board is just not very good.
There are so much better motherboards out there.
Like significantly much better.
For significantly less money.

I think the love affair with the m919 is the infamous reputation of PC CHIPS. These boards came with DIP sockets, DIP sockets fake cache, QFP fake cache, and QFP bare pads. Having each variant of this board would make an interesting exhibit. It was also the only 486 board to use a proprietary COAST module, that is special. Another attribute is the UMC 8881 chipset with VLB and PCI. From what I've seen, most of these VIP boards were on SiS chipsets. Maybe the ultra thin PCB also has some appeal, for its raw cheapness. Personally, I wouldn't pay more than $20 for an m919 and I don't think any of these unique points put it in the $300 range.

@WJG6260

Finally read your message with comprehension.
You are right. Integration and BIOS microcodes optimization matters.
So disqualifying Miro cards since they are apparently not indicative of best possible outcome.
With the remaining Diamond and Paradise cards we can say that for DOS interactive graphics:

1             Diamond S3 Vision868  >  Paradise S3 Vision864  >  Diamond S3 Trio64
2   wolf3d:          122.1                   121.8                   120.3
3     doom:           57.1                    57                      57
4PCP Bench:           26.8                    26.8                    26.7
5  Quake 1:           17.9                    17.9                    17.9

As you can see the performance differences are within the margin of error.

For Windows however the situation is different and Miro's S3 Vision964 comes on top, so:

1Miro S3  Vision964  >  Diamond S3 Vision868  >  Diamond S3 Trio64  >  Paradise S3 Vision864

The picture is starting to sharpen for me.

Hope you can find Diamond 968 as well.

Also, what happens if you slap a Diamond BIOS on Paradise 964 ?
Basically trying to further normalize the results.

---

The ATI Match32 and Match64 cards look weak. These two have probably the most disproportional ebay_price/perfromance ratio out there.

---

TDK PKM-0033S voltage options: 3.3, 3.45, 3.6, 4.0, 5.0

---

No idea about the Chicony CH-471B story.
These are not common boards, so i doubt they were on the market for long and in quantity.
I can see why : D

pshipkov wrote on 2022-12-10, 01:41:

Diamond S3 Vision868 > Paradise S3 Vision864 > Diamond S3 Trio64 For Windows however the situation is different and Miro's S3 Vi […]
Show full quote

Diamond S3 Vision868 > Paradise S3 Vision864 > Diamond S3 Trio64
For Windows however the situation is different and Miro's S3 Vision964 comes on top, so:
S3 Vision964 > Diamond S3 Vision868 > Diamond S3 Trio64 > Paradise S3 Vision864
That is with performance differences within the margin of error.

Thank you for this simplification. Are you able to interject some DOOM numbers in brackets right after the card, e.g. Vision964 [55.54 fps]? And do you have a Trio64V+? My memory is a little foggy, but I think 64V+ was a hint better than the regular Trio64.

I don't have S3 Trio64+.
Have a blank PCB that waits population. One day.
Actually, do you know if simply swapping the Trio64 chip with Trio64+ one in the Diamond DRAM VLB card will work ?
It should, right ?

Probably, if also swap the BIOS chip, but I don’t know for sure. Madao would know better. The answer might even be in his thread on the subject.

Well, it seems my V3-2000 is a "lucky card" - it still works.
Starting it at the M919 shows no post (as already told), with a little bit black-green dark-lighted empty screen.
Fortunately, I switched off the system 2 seconds later 😉

Another of my experience with a PC-Chips-board (a Pentium-I-class one) more than 20 years ago was, that this board was NEVER able to adjust the correct CPU-speed at the system-boot.
The POST shows "fantastic frequencies from neverland" of the K6-2-CPU mounted on it - a few MHz more or less than adjusted, even with a decimal point (e.g. 302,4 MHz, to show the correctness of incorrectness - very funny) 😀
I cannot remember the exact model anymore, but it was an ATX-board without AGP, if I remember correctly.

The crazy ebay-prices for the M919 are for me not understandable - except maybe of a dealer-interest to make big money with old/rare hardware (regardless of the quality)...

I think the ever increasing number of vintage computer youtubers is also assisting inflation. 10 yrs ago, there weren't any. All it takes is one youtube video from a modest youtuber (e.g. 2K subscribers) on a particular hardware piece for demand to shoot way up.

pshipkov wrote on 2022-12-03, 03:24:

....
And the usual set of benchmarks:
.....

Hi, sorry, I quoted you because of your usual set of benchmarks....

I made some progress with the setup of the PVI board, I wanted to install Win98 on it.
So before I found out I can have a "all fastest configuration, cache 2-1-2" with 256kB double banked cache with 50 (x3) setting.
It was unstable with Windows 98. I switched to 4x40MHz, no stability issues.

The existing Win98 installation was a bit strange, there was no driver for the PCI S3 Trio64+ card and other things.
OK, I decided to re-do a complete new fresh Win98 installation, but it was very difficult.
Long story short: Removed the Advansys ISA PnP ISA SCSI controller (was just for the CD ROM) and switched to an IDE DVD drive.
Also removed the non PnP Sound Blaster 16 card to minimize resource conflicts.

Now the installation seems to be fine: The 3Com 3C509B ISA PnP works, graphic card works. I can transfer files from/to my main workstation via FTP.
So I have the first values of Winstone 2 graphics:

1) PVI 160MHz S3 Trio 64+ 1MB 800x600 16bit: 6818kPixel/sec.
2) PVI 160MHz Trident Vesa Local 9440AGi 2MB 800x600 8bit: 4023kPixel/sec.

So you can see the first question: Is there a driver for the Trident for Windows 98??? It is detected automatically by Windows as "Trident Super VGA", but when you search in the details, it says only 1MB.
The card has 2MB, at least on the boot screen it writes 2048k, dos tools also detect 2MB. With the Win98 driver I can only select 256 colours 🙁
I also wonder if the low value ~4000 is ok, maybe also because of the standard driver (?)

Now about the benchmarks:
Superscape, PC Player Benchmarks 320x200 and 640x480, Doom, Quake1 are all in Phil's Suite as far as I know. Also Wolf 3D
Wintune2 I now have.
So I still miss 3D Studio R3 and LightWave 3D, could you provide a link?

UPDATE: I found out that I can use the Windows 98 built in: Trident 9440 PCI driver. Then I can use 16bit colors for example:
Wintune 2: ~ 3000kPixel/sec

I found another driver: w98tgui.exe, picture quality seemed to better with this one, 16bit colors ok, identifies card as VL-BUS OK, but also 3000kPixel/sec only 🙁

So the 9440 is really slow under Windows 98 most likely.

CoffeeOne wrote on 2022-12-10, 20:43:

Hi, sorry, I quoted you because of your usual set of benchmarks.... […]
Show full quote

pshipkov wrote on 2022-12-03, 03:24:

....
And the usual set of benchmarks:
.....

Hi, sorry, I quoted you because of your usual set of benchmarks....

I made some progress with the setup of the PVI board, I wanted to install Win98 on it.
So before I found out I can have a "all fastest configuration, cache 2-1-2" with 256kB double banked cache with 50 (x3) setting.
It was unstable with Windows 98. I switched to 4x40MHz, no stability issues.

The existing Win98 installation was a bit strange, there was no driver for the PCI S3 Trio64+ card and other things.
OK, I decided to re-do a complete new fresh Win98 installation, but it was very difficult.
Long story short: Removed the Advansys ISA PnP ISA SCSI controller (was just for the CD ROM) and switched to an IDE DVD drive.
Also removed the non PnP Sound Blaster 16 card to minimize resource conflicts.

Now the installation seems to be fine: The 3Com 3C509B ISA PnP works, graphic card works. I can transfer files from/to my main workstation via FTP.
So I have the first values of Winstone 2 graphics:

1) PVI 160MHz S3 Trio 64+ 1MB 800x600 16bit: 6818kPixel/sec.
2) PVI 160MHz Trident Vesa Local 9440AGi 2MB 800x600 8bit: 4023kPixel/sec.

So you can see the first question: Is there a driver for the Trident for Windows 98??? It is detected automatically by Windows as "Trident Super VGA", but when you search in the details, it says only 1MB.
The card has 2MB, at least on the boot screen it writes 2048k, dos tools also detect 2MB. With the Win98 driver I can only select 256 colours 🙁
I also wonder if the low value ~4000 is ok, maybe also because of the standard driver (?)

Now about the benchmarks:
Superscape, PC Player Benchmarks 320x200 and 640x480, Doom, Quake1 are all in Phil's Suite as far as I know. Also Wolf 3D
Wintune2 I now have.
So I still miss 3D Studio R3 and LightWave 3D, could you provide a link?

UPDATE: I found out that I can use the Windows 98 built in: Trident 9440 PCI driver. Then I can use 16bit colors for example:
Wintune 2: ~ 3000kPixel/sec

I found another driver: w98tgui.exe, picture quality seemed to better with this one, 16bit colors ok, identifies card as VL-BUS OK, but also 3000kPixel/sec only 🙁

So the 9440 is really slow under Windows 98 most likely.

Looks like you need to curate the L2 cache chips a bit more.
My advise is to try with 512Kb - less chips means smaller chance some of them to be flaky.
Also, perf will be better.

Asus PVI has the best on-board IDE controller among all 486 motherboards - feel good about using it. : )

Attached you a Win95 driver for Trident TGUI9440*.
Hope it works for you.

Yes, getting LAN working makes things very easy for file i/o, especially if you have NAS or FTP.

Here is a link to 3D Studio R3 and LightWave3D 4.
Unzip in C:\
3D Studio runs in DOS. Simply start 3ds.exe. Load the chevy.3ds file and press CTRL+R.
For LW3D test - it is self-contained. Should be able to start it from Windows 95. Navigate to the PROGRAMS directory. Start LIGHTWAVE.EXE. Load SPACE/BLADE.LWS. Press RENDER in the lower left corner.
If you have problems with LW3D in Win95 i provided Win3.1 in the archive. Start it. Run the program as above and see if you pass.

CoffeeOne, if it makes you feel any better, pshipkov's cache matching technique is something I also haven't been able to master. It really is a black art.

No art here sir, just grind.
The L2 cache is 486's Achilles heel.
Take care of that and good things will come to you.
The first iterations back in the day were really painful.
So many flaky chips out there.
Over time several good sets were distilled.
Now it goes pretty quick, most of the time.

Thank you for the files!
About Windows driver: Yes, it worked, but fastest was Windows 98 built in for PCI 9440 (not made for this hardware, but works).
The lightwave I cannot run, it seems to need a windows 3.11 DLL. You sent also the windows, but I have now everything installed on a 6GB hdd with fat32. Windows 3.11 refuses to run on Dos 7.1 and booting from floppy with a Dos 5.0 does not help, because then I cannot read the harddisk 😀 No Win 3.11 for me now.
For the autodesk I seem to be too stupid. I can load chevy.3ds, but CTRL+R is replacing something, it does not start something.

And last question: Where is Wolf3D? Phil removed it from his benchmark-suite? Noooooooo.

So the first values are: for VLI 256kB cache, X5-160MHz, VL Trident 9440AGi, 32MB RAM - Windows 98SE

Wintune 2 800x600 16bit: 2958kPixel/sec
Superscape fast: 83.4
PCP 320x200: 22.4
PCP 640x480: 8.3
Doom Max 51.69 fps
Quake 640x480 16.0

Maybe I should repeat everything with my PCI S3, at least the Wintune2 value is more than 2 times higher, but I guess in DOS the Trident is fast.
And yes, I can test everything with 512kB Cache. But I will not try to optimise the 50MHz setting, 40MHz is very easy and fast .... I already know, that 2-1-2 works with single bank 512kB with 40MHz.

CoffeeOne wrote on 2022-12-11, 15:41:

Thank you for the files! About Windows driver: Yes, it worked, but fastest was Windows 98 built in for PCI 9440 (not made for th […]
Show full quote

Thank you for the files!
About Windows driver: Yes, it worked, but fastest was Windows 98 built in for PCI 9440 (not made for this hardware, but works).
The lightwave I cannot run, it seems to need a windows 3.11 DLL. You sent also the windows, but I have now everything installed on a 6GB hdd with fat32. Windows 3.11 refuses to run on Dos 7.1 and booting from floppy with a Dos 5.0 does not help, because then I cannot read the harddisk 😀 No Win 3.11 for me now.
For the autodesk I seem to be too stupid. I can load chevy.3ds, but CTRL+R is replacing something, it does not start something.

And last question: Where is Wolf3D? Phil removed it from his benchmark-suite? Noooooooo.

So the first values are: for VLI 256kB cache, X5-160MHz, VL Trident 9440AGi, 32MB RAM - Windows 98SE

Wintune 2 800x600 16bit: 2958kPixel/sec
Superscape fast: 83.4
PCP 320x200: 22.4
PCP 640x480: 8.3
Doom Max 51.69 fps
Quake 640x480 16.0

Maybe I should repeat everything with my PCI S3, at least the Wintune2 value is more than 2 times higher, but I guess in DOS the Trident is fast.
And yes, I can test everything with 512kB Cache. But I will not try to optimise the 50MHz setting, 40MHz is very easy and fast .... I already know, that 2-1-2 works with single bank 512kB with 40MHz.

Attached you Wolf3D. Run the test.bat file. Run the test twice in a row. It auto repeats. Also you can skip some of the waiting upon displaying of the frame rate by pressing spacebar.

For 3D Studio - my bad. The key combination is ALT+R.

Will pass you a portable LW3D in one of the coming days.
But you can simply get the missing DLLs from the WIN directory i provided and copy them to A\LW3D\PROGRAMS.

Here is what i am seeing with PVI+TUI9440AGI. Wonder why the difference between us ?
Wintune 2 800x600 16bit: 2958 KPixel/sec (my number: 6021) <- !
Superscape fast: 83.4 (my number: 93.3)
PCP 320x200: 22.4 (my number: 24.1)
PCP 640x480: 8.3 (my number: 9.9)
Doom Max 51.69(my number: 61.6) <- !
Quake 640x480 16.0 (my number: 16.9)

Is your L2 cache in WB or WT mode ?

pshipkov wrote on 2022-12-11, 21:55:

Attached you Wolf3D. Run the test.bat file. Run the test twice in a row. It auto repeats. Also you can skip some of the waiting […]
Show full quote

CoffeeOne wrote on 2022-12-11, 15:41:

Thank you for the files! About Windows driver: Yes, it worked, but fastest was Windows 98 built in for PCI 9440 (not made for th […]
Show full quote

Thank you for the files!
About Windows driver: Yes, it worked, but fastest was Windows 98 built in for PCI 9440 (not made for this hardware, but works).
The lightwave I cannot run, it seems to need a windows 3.11 DLL. You sent also the windows, but I have now everything installed on a 6GB hdd with fat32. Windows 3.11 refuses to run on Dos 7.1 and booting from floppy with a Dos 5.0 does not help, because then I cannot read the harddisk 😀 No Win 3.11 for me now.
For the autodesk I seem to be too stupid. I can load chevy.3ds, but CTRL+R is replacing something, it does not start something.

And last question: Where is Wolf3D? Phil removed it from his benchmark-suite? Noooooooo.

So the first values are: for VLI 256kB cache, X5-160MHz, VL Trident 9440AGi, 32MB RAM - Windows 98SE

Wintune 2 800x600 16bit: 2958kPixel/sec
Superscape fast: 83.4
PCP 320x200: 22.4
PCP 640x480: 8.3
Doom Max 51.69 fps
Quake 640x480 16.0

Maybe I should repeat everything with my PCI S3, at least the Wintune2 value is more than 2 times higher, but I guess in DOS the Trident is fast.
And yes, I can test everything with 512kB Cache. But I will not try to optimise the 50MHz setting, 40MHz is very easy and fast .... I already know, that 2-1-2 works with single bank 512kB with 40MHz.

Attached you Wolf3D. Run the test.bat file. Run the test twice in a row. It auto repeats. Also you can skip some of the waiting upon displaying of the frame rate by pressing spacebar.

For 3D Studio - my bad. The key combination is ALT+R.

Will pass you a portable LW3D in one of the coming days.
But you can simply get the missing DLLs from the WIN directory i provided and copy them to A\LW3D\PROGRAMS.

Here is what i am seeing with PVI+TUI9440AGI. Wonder why the difference between us ?
Wintune 2 800x600 16bit: 2958 KPixel/sec (my number: 6021) <- !
Superscape fast: 83.4 (my number: 93.3)
PCP 320x200: 22.4 (my number: 24.1)
PCP 640x480: 8.3 (my number: 9.9)
Doom Max 51.69(my number: 61.6) <- !
Quake 640x480 16.0 (my number: 16.9)

Is your L2 cache in WB or WT mode ?

Hi, about Wintune, you ran it with Windows 3.11 (is it correct?) and me with Windows 98SE, that's simply too much difference.
But my slower values in DOS are interesting indeed.
How is your DOS configured? only Himem.sys? Or also EMM386? with NOEMS?
Maybe you had much more free conventional memory, I did not optimize a lot.
L2 cache was WB, L1 cache auto, so also WB. So I should have maximum performance.

But maybe my Trident 9440 sucks, adds some wait states or something, who knows?
The difference in Doom Max is really big, I agree.

EDIT: I will check now JP29 VESA WAIT STATE CONFIGURATION
and JP30 BUS SPEED CONFIGURATION

MORE EDIT: Both are on 1-2, which means 0 wait states and bus speed <= 33MHz. That looks ok.

EVEN MORE EDIT:
I have 2 jumpers on the graphics card: JX7 (closed)
and J7 (closed)
I think JX7 might be important:
Setting
CPU speed <= 33MHz
Open

CPU speed > 33MHz
Closed
So I try with JX7 open

MORE EDIT AFTER EVEN MORE EDIT:
Yes: with JX7 open I get in Doom fastest: 1277 or 1278 realticks so that is 58.4 - 58.5. Much closer to your values!

The numbers i posted was captured in Win95.
Actually the differences are not big between Win3.11 and Win95 for that test. Also, neither OS has advantage.
Ok, you discovered the issue - added wait states on the graphics adapter.
The rest of the metrics should more or less match mine now.

One more 486 motherboard with 1024Kb level 2 cache.

FIC 486-PIO3

An entirely trouble-free experience !

Based on previous encounters with FIC 486 class hardware i expected the usual set of problems related to fussy L2 cache handling, weird/missing CPU voltages, messy jumpers, instabilities with tight wait states.
Not this time. FIC got it right with their last in line model from the PIO series.

Seems to take any L2 cache chips with ease (impressive). Cache policy is WB for best performance.
Does not support EDO memory, FPM only.
Latest BIOS. Limited options.
Slow on-board IDE controller. You don't feel it with CF cards.
Voodoo3 cards work.
Nice range of CPU voltage options - 3.3 / 3.45 / 3.6 / 4 / 5.
Very simple jumpers.

And that's pretty much it. Stuff just works.

--- Am5x86 @160MHz (4x40)

All BIOS settings on MAX.

What SpeedSys says.
I don't trust this test at all, but notice the L2 cached vs raw memory read speeds. Flat curve.

benchmark results

--- Am5x86 @180MHz (3x60)

Default 3.45V to CPU.
Gets a bit picky about RAM modules. Quickly found 2x16Mb that worked well.
No bus divider needed.
All BIOS settings on max, except CACHE TIMING CONTROL = NORMAL (best is TURBO, can be FAST for DOS interactive graphics tests but fails at complex offline computation tasks).
And that's pretty much it - system is very stable.

SpeedSys.
IDE controller jumps 2x in perf it seems.
Much bigger difference between L2 cached and raw memory reads.

The usual set of benchmarks.
Tested with Matrox Millennium PCI.
Compared against the best performers in each discipline.

--- Am5x86 @200MHz (4x50)

Couldn't get the board to finish POST. A small red flag.

--- Am5x86 @200MHz (3x66)

I don't think there is a jumper configuration for 66MHz available despite the clock generator supporting it.
May try a mod one day.

--- P24T @100MHz (2.5x40)

One jumper change needed only to switch between Am5x86 and P24T processors.
All BIOS settings on max.
And that's it. Stuff works.

Updated the comprehensive POD100 extensive testing, summary post with the PIO3 data.

---

Not the best performer on a clock-to-clock basis.
The added wait states contribute to very good system stability.
At minimum - if you want to build a 486 system but don't want any of the retro hardware drama - that's the board.

I don't trust this test at all, but notice the L2 cached vs raw memory read speeds. Flat curve.

haha, fast memory or slow cache, which one?

You have a typo, I assume you meant 3x60. If not, then you meant 200 MHz.

--- Am5x86 @180MHz (3x66)

Overall, performance rather slow, which is the word on the street with all these FIC based socket 3 boards. Personally, I thought it would be even slower. I'm surprised that it took to the Voodoo3. Finally the SiS 496 has a partner is this arena.

Does anyone mind if I return to this for a moment?

pshipkov wrote on 2022-12-10, 01:41:

With the remaining Diamond and Paradise cards we can say that for DOS interactive graphics: […]
Show full quote

With the remaining Diamond and Paradise cards we can say that for DOS interactive graphics:

Copy code to clipboard

1             Diamond S3 Vision868  >  Paradise S3 Vision864  >  Diamond S3 Trio64
2   wolf3d:          122.1                   121.8                   120.3
3     doom:           57.1                    57                      57
4PCP Bench:           26.8                    26.8                    26.7
5  Quake 1:           17.9                    17.9                    17.9

As you can see the performance differences are within the margin of error.

For Windows however the situation is different and Miro's S3 Vision964 comes on top, so:

Copy code to clipboard

1Miro S3  Vision964  >  Diamond S3 Vision868  >  Diamond S3 Trio64  >  Paradise S3 Vision864

Were these values tabulated with Am5x86-160?

If I recall, the Diamond Vision968 is the VLB king for Windows GUI acceleration, with the trade-off being a slight hit with DOS performance. I've run these tests before and from memory, I think the hit was around 1.5 fps in DOOM w/Am5x86-160. However, I recently tested a SPEA Mercury P64V replica, which contains the S3-968, and the DOOM score was 2.4 fps higher than the Diamond S3 968. Perhaps the gap is closed with this card?

DOOM
Diamond S3 968 ---> 58.44 fps
SPEA S3 968 ---> 60.82 fps

Has anyone else run this comparison?

L2 cache read speed is too low. The RAM number is inline with the expected. Looks like very conservative wait states.
V3 support came as a big good surprise.

---

Numbers are with 160MHz Am5x86 system.

I remember looking at your shared numbers and some of the private conversations we had about that.
Yes, 968 is top VLB dog in Windows GUI and slightly behind in DOS interactive graphics.
With all these cards, if they are of decent quality/implementation we are all splitting hairs really, but still, interesting to quantify how were things back then.
The VLB space for VGA/IDE is not documented very well, so any systematized information is interesting to me.
I don't have S3 Vision VLB cards, so rely on yours and WGS's numbers for them.

Could you remind of your DOOM test results with Trio64 and ARK VL1000 at 160 Mhz?