I will have to check again w/GD5434. Hard to find much time these days. Can you attach the 481/482 BIOS for BL3? I will check it out as well.

_

Thanks.

I am pleased to report that I am able to get the SiS Rabbit board working with the BL3-90 in Win3.11 now. It would seem that for use in Windows, I need to load the CPU driver before HIMEM, not after. I'm using 16 MB memory, 256K L2, GD5434, Adaptec AHA-1522B SCSI, SCSI2SD v6 adaptor, and PS/2 mouse mod for the keyboard controller. I was able to browser files via an ISA ethernet card and do a google search with IE5.

However, I wouldn't get too excited yet. The SiS Rabbit was my only non-ALi Panda board that even worked with the BL3. I will try my UM481/482 board with the BIOS provided by pshipkov. So you can use the 491 BIOS w/481?

I guess I shouldn't have been tested my systems with an STB ERGO based on the ET4000AX. I suspect that was causing some of the trouble w/BL3. Not sure why.

I was able to use the stock MSI supplied AMI BIOS on my UM481/482 board and run the BL2 at 2x40 MHz. I am able to run Win3.11, however if the CPU is having some activity, my PS/2 mouse gets pauses in cursor movement. I then ran it at 2x45 and Doom scores 21.85 fps, so quite a bit lower than with the SiS Rabbit. I ran Windows 3.11, opened IE5, but got a crash. Perhaps due to a) 20 ns SRAM, b) 15 Mhz ISA, c) too fast of wait states for SRAM/DRAM. Not sure, but doesn't really matter. No need to test 90 Mhz any further on the UMC board.

EDIT: That UMC BIOS you provided, I tried it on my UM481 board, but screen stays blank and I get a long beep at power-on.

Built quality varies a lot between the different ET4000AX cards. If you have a Micro-Labs 1998 model - it holds really very well. Some early/other variants can be brittle.
GD5434 is the best place for ISA. I think there are cards by Diamond and STB - both are solid top to bottom. This removes volatility.
Stating some info based on previous experience, since you brought the point.

Yes, UMC's 481/491 chipsets are ok, but there are faster and more compatible options out there.
It is possible the BIOS i provided to not work on 481.
Recently i have been testing an UM82C491F based 386 motherboard.
Used Mr.BIOS. While it worked well with 486DX, DLC and SXL2 processors, it kept the board dark with BL3 chip.
Found several UMC BIOSes. The latest working one was the file i attached.
Can in turn provide your BIOS please ?

My Micro-Labs ET4000AX is in my 286 system. Didn't want to open it; don't want to keep buying more hardware. I tend to avoid my Diamond GD5434 because it is NIB and buried far back in my closet (refer to those photos!)

Bios for my MSI MS-3131 is attached. In the ZIP are the MRBIOS and the MS-3131 BIOS. My main gripe with the MSI BIOS is that the BIOS doesn't have an EXTERNAL only ENABLE option for cache. It has options: DISABLED, INTERNAL, or BOTH. From some notes I have written in my manual, the DLC/SXL chips need L1 disabled in BIOS to work properly, then enabled in software. The only way to do this is with DISABLED, but then L2 is also disabled. BL3 is OK to have BOTH set. For DLC/SXL chips, solution is to uSe MR BIOS.

Next I will test my Symphony and ALi Panda again.

Thanks. Will give that BIOS a try today since mobo is on the bench.

Mr.BIOS microcodes are great alternative to some of the sucky BIOSes out there, but if one is thorough - there is usually even better AMI/AWARD alternative in the most cases. It takes more time to find these since the difference between the many AMI/AWARD versions is not obvious right away.

pshipkov,

Are you loading REVTO486.SYS before HIMEM or after HIMEM? If loading before HIMEM, have you tried setting HIMEM to /TESTMEM:ON ? If so, do you get any memory errors? I get himem errors on my Symphony board using the BL2. I have to turn /TESTMEM:off , but I am still able to load Windows 3.11, IE5, DOOM, etc.

Are you also able to soft reset (ctrl-alt-del) on your Symphony board while using the BL2/3? soft-reset doesn't work for me on the Symphony board while using the BL2/3, but works fine with SXL2.

By way of comparison at 80 MHz, 256K, GD5434

Symphony Haydn II
BL2: 80 MHz at 2x40 and ISA at 13.3 MHz, 3.6 V, then DOOM = 3654 = 20.44 fps
SXL2: 80 MHz at 2x40 and ISA at 13.3 MHz, 4.1 V, then DOOM = 3406 = 21.93 fps

SiS Rabbit
BL2: 80 MHz at 2x40 and ISA at 13.3 MHz, 3.6 V, then DOOM = 3562 = 20.97 fps
SXL2: 80 MHz at 2x40 and ISA at 13.3 MHz, 4.1 V, then DOOM = 3801 = 19.79 fps

UMC 481/482
BL2: 80 MHz at 2x40 and ISA at 13.3 MHz, 3.6 V, then DOOM = 3843 = 19.44 fps

Looks like the SiS Rabbit is the fastest with BL2/BL3 chips, per clock.

I tried to take my Symphony board up to 15 MHz on the ISA bus, but my AHA-1522B's BIOS would hang at this frequency. This is curious because it ran fine at 15 Mhz on other motherboards. Thus, I had to run the Symphony board at 90 MHz and 11.3 MHz ISA, 1 WS DRAM = 3465 realtics = 21.56 fps.

It looks like the SiS Rabbit is better suited for the BL3, while the Symphony is better suited for the SXL2. On the SiS Rabbit board, the BL2 is 5% faster per-clock compared to the SXL2, while on the Symphony board, the SXL2 is 5% faster than the BL2.

Didn't try before HIMEM.SYS. Always had the REVTO486.SYS driver after it.
Memory errors happen for me if i go past whatever is the stability threshold on a given system.
Sometimes that leads to hang during POST or boot to DOS. In other cases the system is seemingly stable, but will eventually hang on some tasks.
Compilers and rendering software is very sensitive to this kind of stuff.
In many cases DOS interactive graphics and Windows casual apps don't show side effects.

Looking at your numbers i think that actually the Symphony board is the fastest, or can be if it was not the SCSI issue.
I had this problem before. SCSI controllers get flaky on fast 386 systems with DLC/BL3 procs that are pushed hard. Most ISA IDE ones are in the same boat too.
This becomes a limiting factor unless you find the magical IDE controller that handles things well.
We touched on Promise EIDE Pro before. Another controller that seems to hold well is a specific assembly with UMC chipset. Can pass a pic if you want.
Solving this can allow for tightest BIOS timings and few more perf drops.

It is interesting that i see the opposite at 1x50 - BL3 slightly faster than SXL2.
Go figure.

Btw, the bios you provided does not support UM82C491F chipsets. Lights are off.

The BIOS I provided is for the UMC 481/482. I don't have a 491 board.

On the Symphony board, running REVTO486 after HIMEM causes a hang-up during loading of REVTO486. I must load REVTO486 before HIMEM for the system to work. Next time you have your Symphony board out, could you let me know if you get HIMEM errors if you load REVTO486 first?

I'm not sure why you noted the Symphony is fastest with BL. My numbers indicate the SiS Rabbit is faster per clock with BL, not the symphony, that is, 20.44 fps Symphony vs. 20.97 fps Rabbit. Also the Rabbit has less issues with higher ISA speeds and my hard drive controller. The Rabbit let ISA at 20 MHz on SCSI work. Seems like Rabbit and BL is just a better pairing, while SXL and Synphony is a better pairing. If I were to setup a hyper BL2 system w/ISA, it would be Rabbit at 2x50.

Could you run a clock-for-clock comparison of SXL2 and BL2 on your Symphony boards? I would be very interested if on your boards, the BL2 was faster than the SXL2.

Would you also be interested in checking soft reset on your Symphony and BL2/3 system next time it is out?

I will be moving onto the ALi Panda board with my BL3 hack next. I have a second board, not in a case, that I can work with in finer detail. I suspect the ALi Panda can do 2x40 w/BL2, but only with ISA graphics and/or ISA fixed disk controllers, and that when mixing VLB graphics + VLB IDE that it must be brought down to a 25 MHz bus. Maybe it can cope with VLB graphics + ISA SCSI at 3x33 - we'll see. If it can, then I'll be pulling out my Promise VLB IDE controller from my system.

Ran some DOOM tests with the Daewoo AL486V-D, aka ALi Panda. Using AHA-1522B w/SCSI2SD v6, ISA ethernet, and KBC PS/2 mod.

BL2-80
GD5434 ISA = 4427 = 16.87 fps
Trio64V+VLB = 3291 = 22.7 fps

Notice how the ISA speed is notably slower than the other ISA only boards. The ALi Panda board doesn't have any control over the ISA speed. This is not a problem if using the VLB slots.

BL3-100
Trio64V+ VLB = 2858 = 26.13 fps
Trio64V+ VLB with 0 WS BIOS = 2799 = 26.68 fps

This is only 0.5 fps faster than the SIS Rabbit with 2x50 and the ISA GD5434 at 16.7 fps, but at least all your ISA cards will be in spec on the ALi Panda system. Yet to be determined is if the system is still functional with a Promise VLB IDE controller installed.

EDIT:
BL3-100
ARK1000VL = 2795 = 26.72 fps

so the venerable ARK is about the same speed as the Trio64V+ with 0WS on this ALi Panda w/BL3-100.

EDIT:
BL3-100
Diamond Trio64 VLB = 2856 = 26.15 fps
Number 9 Trio64 VLB = 2858 = 26.13 fps

feipoa wrote on 2023-01-24, 07:36:

On the Symphony board, running REVTO486 after HIMEM causes a hang-up during loading of REVTO486. I must load REVTO486 before HIMEM for the system to work. Next time you have your Symphony board out, could you let me know if you get HIMEM errors if you load REVTO486 first?

You probably have DOS=HIGH, UMB.
Change it to DOS=UMB.

feipoa wrote on 2023-01-24, 07:36:

Could you run a clock-for-clock comparison of SXL2 and BL2 on your Symphony boards? I would be very interested if on your boards, the BL2 was faster than the SXL2.

I did it again. Didn't i ?
You are right. Symphony Haydn is more efficient with SXL2 CPU.
We had the same exchange in the other thread and initially i stated the opposite then corrected myself.

feipoa wrote on 2023-01-24, 07:36:

Would you also be interested in checking soft reset on your Symphony and BL2/3 system next time it is out?

On the 2 Symphony boards soft reset with BL3 is fine.

About your Panda VLB board.
I see similar pattern between Alaris Cougar 3x33 with Ark1000VL or S3 Trio64 and Symphony 2x50 with GD5434 - perf is on par, assuming stable system.
Which makes sense - game is heavily CPU bound.

pshipkov wrote on 2023-01-25, 04:06:

feipoa wrote on 2023-01-24, 07:36:

On the Symphony board, running REVTO486 after HIMEM causes a hang-up during loading of REVTO486. I must load REVTO486 before HIMEM for the system to work. Next time you have your Symphony board out, could you let me know if you get HIMEM errors if you load REVTO486 first?

You probably have DOS=HIGH, UMB.
Change it to DOS=UMB.

Yes, I am running DOS high; is this wrong? I've been running it like this forever. Don't we normally want to run DOS in extended memory (high) to free up conventional memory for DOS games which need conventional memory? Do you normally want to run DOS in blocks of the upper memory area (UMB)? I thought it was best to save the upper memory area for adaptor ROMs, even if currently unused? And what if you have your system loaded with cards that use UMA, I don't know, perhaps a SCSI card, an IDE card, a LAN card with boot ROM, etc? I also thought it was best to save this UMA for potential games which might want it? Could you correct my thinking?

So you are able to run REVTO486 before HIMEM if you have DOS=UMB?

pshipkov wrote on 2023-01-25, 04:21:

About your Panda VLB board.
I see similar pattern between Alaris Cougar 3x33 with Ark1000VL or S3 Trio64 and Symphony 2x50 with GD5434 - perf is on par, assuming stable system.
Which makes sense - game is heavily CPU bound.

That is curious. Did you find a way to run your Trio64 with 0 ws? The only way I could get the Trio64V+ to equate to ARKVL1000 performance in DOOM was to use the 0 ws vga ROM.

The BL3 drivers seem to interfere with DOS=HIGH in some way which can lead to a hang, or cause MSDOS.SYS not to be loaded in the high memory area after all.
Don't ask me why. Can be the Symphony BIOS microcodes at play.

Something important i forgot to mention (i think).
The LIGHT486.SYS driver is faster than REVTO486.SYS.
Also, it takes 1/3th of the resident memory only - 500b vs 1.5Kb.
Give it a try.
(will need to update some of the numbers i have posted in the other thread).

I run S3 Trio64 at 0-WS all the time. It is still slower than Ark1000VL.
I ever need 1-WS when i overclock 486 motherboards to 50/60/66 MHz.
Everything else 0-WS. At least cannot remember a case where 1-WS was needed for 40 and less MHz.

I guess I will have to try DOS=UMB in various situations to see if it fixes the BL3 driver load order. Might bring those boards back out when I'm finished to double check.

Going from memory, I think the Kingston LIGHT driver only allowed 2x mode, not 3x mode. Is my memory wrong?

I've spent a lot of time with CTCHIP and BL3 settings in the past. Biggest bang was setting XTOUT = 0 for another -90 or so realtics. I bet this is what the Kingston driver is doing, or some combination of other settings. I think with all "other" settings, you might get another -20 realtics. I know I made a sheet of what the two drivers set up, but locating it and desiphering my handwriting won't be fun.

I guess you need the Trio64V+ to beat the ARK VL1000.

Does your Trio64 have a VGA BIOS mod to force 0 WS, or are you relying on the motherboard jumper? If I recall, the motherboard jumper doesn't always prevail. Also, I think your Trio64 is the X variant, while mine is the P variant, and using your BIOSes cause issues with soft reset, as confirmed by at least one other user besides me.

LIGHT486.SYS supports 2x and 3x multipliers.
I am more satisfied with it in general.

Looking at the waterfall of info in this post - i am not sure if Trio64V+ can outjump Ark1000VL.
If i can fix the Trio64V+ here i will know more first hand.

Cannot remember a case where the S3 Trio64's wait-state jumper was not respected.

That waterfall made my eyes water!

I remember the Evergreen driver having more support features and better documentation. To save me from digging up a mountain of old notes, could you let me know what flags you are using for LIGHT486.SYS ?

Regarding disrespected 0ws jumpers - I think I was referring to the case that the motherboard was trying to set the VESA wait state, not the graphics card. It is foggy upstairs now.

Edit:
When you mentioned your Trio64 VLB, was that using the Number9 BIOS, or Diamond? I haven't yet compared the Trio64 #9 vs. Trio64V+ 0ws. Perhaps those two are nearly equal, I don't know.

For once i am not the one with the long posts.

LIGHT486.SYS /2
That is for 2x50MHz.
It can be /3 as well.

If i am not mistaken the motherboard and video card wait state jumpers are chained.
Both need to be in the appropriate position for setting the desired wait state.
With that said - i never actually verified with multimeter, only speaking from observation.

I will try LIGHT486.SYS on the SiS Rabbit and Symphony boards this week. The Daewoo ALi Panda board doesn't need any BL3 driver loaded.

I've added values for other VLB cards above.

BL3-100 (3x33) on Daewoo AL486V-D

Trio64V+ = 2858 = 26.13 fps
Trio64V+ w/0 ws BIOS = 2799 = 26.68 fps
ARK1000VL = 2795 = 26.72 fps
Diamond Trio64 = 2856 = 26.15 fps
Number 9 Trio64 = 2858 = 26.13 fps
Diamond S3 968 = 2870 =26.02 fps

pshipkov, would you be able share your results from the Alaris Cougar with BL3 at 3x33? I'm wondering how the Daewoo compares per-clock to the Cougar. I see mixed results for your system at 3x33 w/ARK1000VL. You get 24.0 fps or 27.5 fps? If that later, that's a big jump in performance from your first Cougar post to the next.

I have tried the BL3 hack on my cased Daewoo Ali Panda system at 3x33 and it appears functional w/VLB graphics + IDE. Looks like I'll be upgrading my BL3-75 system to BL3-100 once I build another unit. Hoping for 120 MHz. Next revision will have a larger heatsink.

Posting final photos of the BL3 hack. I forgot that the top ring isn't grounded by default. I've added five connection points through vias to connect it to the bottom ground ring.

If someone is able to procure a large stash of NOS IBM BL3 chips for a good price, there could be some merit in producing a custom PCB for these. I figure IBM BL3 QFP chips would be about as rare of a find as TI SXL2-66 QFP chips.