pshipkov wrote on 2024-06-02, 06:31:

feipoa wrote on 2024-06-02, 02:51:

If I need to run the BL3 at 33 or 40 MHz FSB, there's faster boards than the YB.

Which ones you are thinking about ?

At 33 MHz, the Daewoo Panda-based board will beat the PEM-4036YB due to the use of VLB graphics. Also, my BIOS has been modified to allow for adjustment of the ISA bus speed. The SiS Rabbit at 3x33 or 2x40, with ISA graphics, should also match or beat the PEM-4036YB. The Rabbit has the added bonus of being a pure 386 board, that is, not a board containing a chipset which is a 386/486 hybrid. This matters to some. The results will likely be different at 50 MHz FSB, but that's out of the range of the 4036YB.

pshipkov wrote on 2024-06-02, 06:31:

feipoa wrote on 2024-06-02, 02:51:

Did you try YB with BL3 at 2x45? Or it failed even with 1x45?

The board itself is struggling. CPU clock multiplier is not a factor.

Although I'm quoting the CPU clock, I'm not intending to emphasise it. However, on some boards, success at 1x doesn't necessarily translate into success at 2x, but I'm not talking about that with these DTK boards. I can emphasise FSB spends only if it is clouding the exchange of information.

pshipkov wrote on 2024-06-02, 06:31:

feipoa wrote on 2024-06-02, 02:51:

I played again with the Gemlight at 50 MHz. Even with DRAM at 2 ws, it will not POST successfully. Perhaps the key is your 100 MHz crytals? I remember you saying that this could matter when on the edge. What is the branding on yours? Mine have ECLIPTEK ECH11, 100.00M.

So far all crystals worked fine at up to 100MHz. At least i didn't notice difference between them.
Story changes at 110MHz.

What brand and model of crystals are you using at 100 MHz?

pshipkov wrote on 2024-06-02, 06:31:

Tested both. Wanted to see if they behave differently to determine which one to keep. They functioned the same. One of them went […]
Show full quote

feipoa wrote on 2024-06-02, 02:51:

I already tried all sorts of SIMMs, 4 modules and 8 modules. 50 MHz isn't happening for me, SRAM or otherwise. Also tried 6.67 MHz. Is it possible that you tested up to 50 MHz on just one of your Gemlight boards (the one that was sold online) and the secondary board (the one I received) wasn't confirmed up to 50 MHz? I'm thinking that certain chipsets cannot cope with 50 MHz. I did try a heatsink/fan on the chipset though.

Tested both. Wanted to see if they behave differently to determine which one to keep. They functioned the same. One of them went to Ebay, the other one is with you.
I am really very puzzled that you are struggling to get it going at 50MHz. What can be the problem ? I assume FPU is not inserted, right ? Most of them don't like 50MHz very much.
So at 50MHz no lights at all ? Or POST hangs ? Or boot hangs ? Where exactly it stops ?
Maybe you really collected all the bad memory modules on the planet, but i doubt it. : )

Unlikely.
Yes, I removed the FPU for most of the testing. I think I had put the FPU back when I tested the SCSI 6.67 setting though. The outcome is either the system hangs just after the SCSI BIOS completes (just before boot attempt), or I get "no command interpreter" when booting. The longer the system had been on, the quicker it will hang. For example, if it had just been powered on for the last 5 minutes, it will even freeze inside the BIOS. Implies a heat dependency. It was 23 C in the office (at the door) when I ran these tests, although it was closer to 26 C at the motherboard. My desk lamp has one of those damned heater bulbs that require a UV filter to prevent skin cancer. its the last non-LED bulb in our house. I had replaced it with an LED alternative, but that guy burned out after 3 months, so the halogen went back in. Anway... What was the temp when you tested? The main difference in our setups are the RAM and that fact that I always use PS/2 mice via a KBC adaptor. While I could remove the PS/2-KBC, there's no situation in which I'd accept a 386 board at 50 MHz without a PS/2 mouse. It would be the black sheep in my pasture. We all have our requirements.

pshipkov wrote on 2024-06-02, 06:31:

feipoa wrote on 2024-06-02, 02:51:

I'll have to try this. Are you holding down F2 while opening Notepad with the mouse left-click? Or do you already have Notepad open with some text on it, and only then start to hold down F2 and wait? Wait how long? Are you still holding down F2 while trying to close Notepad with the mouse? Must you open the saved file to witness the corruption? Or is there a corrupt file on the HDD that Scandisk picks out?

You can try it in Windows Commander, Norton Commander, or Dos Navigator.
F4 to open file for edit. Then hold F2 for i don't know - 30 seconds to a minute and see what happens.

You lost me here. You were talking about Notepad and F2 (save in loop), but then bring up Windows/Norton/DOS Commander. That's like Ztreewin, but for DOS right? If so, I don't use any of those. Do your instructions not apply to use in Windows 3.1?

pshipkov wrote on 2024-06-02, 06:31:

What i meant is that when overclocked some SCSI adaptors seem to freak-out the keyboard controller or something like that. In tu […]
Show full quote

feipoa wrote on 2024-06-02, 02:51:

For your other test, you just hold down ENTER while on the Windows 3.1 desktop and move the mouse around? The test passes if apps that the mouse pointer hoovers over aren't opened, and also nothing else flying around?

What i meant is that when overclocked some SCSI adaptors seem to freak-out the keyboard controller or something like that.
In turn the keyboard controller keeps spamming the system with keys presses, including the Enter one.
As a result any context sensitive actions get triggered.
When in Windows if i select an icon it immediately invokes its function. If i move the mouse over buttons or other active elements they get triggered.
It is a zoo. Very dangerous one because you can easily mess-up windows configs and the content on disk. Quickly learned that Wolf3D is 100% indicative of the problem, so stopped double checking in Windows. Anyway.

I have only run into this a few times with extreme edge-of-stability conditions. It is good to know that Wolf3D offers an easier test alternative, but I don't normally run Wolf3D. Are you running Wolf3D with EMM386 enabled, or without EMM386 but with autoexec/config bypassed (to save conventional memory)? When I run Wolf3D, I think I normally run it with the config/autoexec bypassed, unless I need to setup some registers first.

feipoa wrote on 2024-06-02, 09:33:

At 33 MHz, the Daewoo Panda-based board will beat the PEM-4036YB due to the use of VLB graphics. Also, my BIOS has been modified to allow for adjustment of the ISA bus speed. The SiS Rabbit at 3x33 or 2x40, with ISA graphics, should also match or beat the PEM-4036YB. The Rabbit has the added bonus of being a pure 386 board, that is, not a board containing a chipset which is a 386/486 hybrid. This matters to some. The results will likely be different at 50 MHz FSB, but that's out of the range of the 4036YB.

Let me share some numbers since i have them available.
CPU is 386DX, base frequency is 40MHz.
The two best VLB motherboards:
ECS Panda 386V, Ark1000VL / S3 Trio64, 32Mb RAM, 256Kb L2 cache
DataExpert EXP3406, Ark1000VL / S3 Trio64, 32Mb RAM, 256Kb L2 cache
The best SiS Rabbit boards:
Chaintech 340SCD (and Abit AB-FS3), CL GD-5434 2Mb, 32Mb RAM, 256Kb L2 cache
And the DTK guy in question:
DTK PEM-4036YB, CL GD-5434 2Mb, 32Mb RAM, 256Kb L2 cache

1          W3D (fps)   SUPERSCAPE (fps)   PC PLAYER (fps)   DOOM (fps)   WINTUNE2 (fps)   3D STUDIO (seconds)
2PANDA     32.1        16.3               3.8               7.4          2118             no test, too slow
3DATAEXP   37.7        18.8               4                 5.9          2415             1115
4CHAINTECH 30.1        16.7               4.3               8.2          1122             966
5DTK PEM   35.7        18.1               4.3               8.5          1547             962

I think the table confirms my early point. Few more notes:
1. VLB boards pull ahead in Windows GUI because of the S3 Trio64 VLB card - it is a different game entirely, almost unfair, but possible as an option, so it is considered.
2. VLB boards do well in local storage perf when equipped with late VLB EIDE controllers. Until recently the difference compared to ISA only motherboards was large, but now SIIG i540 or Advansys ABP-5140 level out the field for good.
3. DataExpert does well in the CPU-light Wolf3D and Superscape tests.

pshipkov wrote on 2024-06-02, 06:31:

What brand and model of crystals are you using at 100 MHz?

Ecliptek.

pshipkov wrote on 2024-06-02, 06:31:

Unlikely.
Yes, I removed the FPU for most of the testing. I think I had put the FPU back when I tested the SCSI 6.67 setting though. The outcome is either the system hangs just after the SCSI BIOS completes (just before boot attempt), or I get "no command interpreter" when booting. The longer the system had been on, the quicker it will hang. For example, if it had just been powered on for the last 5 minutes, it will even freeze inside the BIOS. Implies a heat dependency. It was 23 C in the office (at the door) when I ran these tests, although it was closer to 26 C at the motherboard. My desk lamp has one of those damned heater bulbs that require a UV filter to prevent skin cancer. its the last non-LED bulb in our house. I had replaced it with an LED alternative, but that guy burned out after 3 months, so the halogen went back in. Anway... What was the temp when you tested? The main difference in our setups are the RAM and that fact that I always use PS/2 mice via a KBC adaptor. While I could remove the PS/2-KBC, there's no situation in which I'd accept a 386 board at 50 MHz without a PS/2 mouse. It would be the black sheep in my pasture. We all have our requirements.

Didn't check temp on the mobo's ICs. Don't know what it was.
It is unlikely that your input devices contraption to be causing any of these problems.
So puzzling. Not sure what to answer.

pshipkov wrote on 2024-06-02, 06:31:

You can try it in Windows Commander, Norton Commander, or Dos Navigator.
F4 to open file for edit. Then hold F2 for i don't know - 30 seconds to a minute and see what happens.

You lost me here. You were talking about Notepad and F2 (save in loop), but then bring up Windows/Norton/DOS Commander. That's like Ztreewin, but for DOS right? If so, I don't use any of those. Do your instructions not apply to use in Windows 3.1?
[/quote]
Never talked about Notepad. : )
Only about those other orthodox file managers, and their built-in text editors.
They are just tools i use. Nothing special about them. You can use other means to "spam" local storage system - write a script, or use another editor that has hotkeys for "save" ...

pshipkov wrote on 2024-06-02, 06:31:

I have only run into this a few times with extreme edge-of-stability conditions. It is good to know that Wolf3D offers an easier test alternative, but I don't normally run Wolf3D. Are you running Wolf3D with EMM386 enabled, or without EMM386 but with autoexec/config bypassed (to save conventional memory)? When I run Wolf3D, I think I normally run it with the config/autoexec bypassed, unless I need to setup some registers first.

I don't use EMM386 unless some software requires it, but use HIMEM at all times.

Those test results are with a 386DX. I could be remebering incorrectly, but I thought the SiS Rabbit was more optimised for a BL3 than the PEM is. However, your answer suggests that the PEM board is still clock-for-clock faster than the Rabbit w/BL3? Do you have these DOOM numbers?

For those Panda tests, were you using my modified BIOS which lets you adjust the ISA speeds, VLB mode (Sync or bypass), etc ?

On my Panda system, BL3-100 (3x33), I'm using fairly conservative settings, with the ISA at only 11.1 MHz.
With Diamond S3 986, DOOM = 26.46 fps
With ARK1000VL, DOOM = 26.74 fps
Can the PEM-4036YB do better than this w/BL3-3x33?

With my SiS Rabbit (Chaintech 340SCD), BL3-100 (3x33), I have ISA set to a somewhat less conservative 13.3 MHz.
With GD-5434, DOOM = 21.27 fps... we can take that number a little higher, and still run at or under 40 MHz, by running the FSB at 37.5, with BL3 at 112.5 MHz, DOOM = 24.0 fps. Someone with more aggressive ISA bus speeds may be able to do better than these.

I haven't run the PEM-4036YB with a BL3 yet, but it is on my list. What DOOM values were you getting on your YB with BL3-100, and what ISA speeds?
.
.
If we both are using the same model of Ecliptek, then we can probably rule it out. We're down to memory modules and my hot office. With summer coming, it isn't getting any cooler in here.
.
.

pshipkov wrote on 2024-06-02, 22:40:

Never talked about Notepad. : )

pshipkov wrote on 2024-06-02, 01:56:

Open a text file in NC or DN text editors and hold F2 (save) to spam the IDE...

Not knowing what the acronynm NC stood for, I thought you meant to write NP, which would stand for Notepad. This is where my understanding for the use of Notepad originated. I wouldn't expect all readers to know what NC or DN stood for.

Symphony Haydn with IBM BL3 running at 100MHz (3x33), all BIOS settings on max, produces 20.7 fps in Doom, which is half a frame slower than SiS Rabbit.
SiS Rabbit is clock-to-clock faster than Symphony Haydn with IBM BL3 processors, but slower with DLC/SXL2 and 386DX.
So, the results are inverted with 386DX, DLC, SXL2 processors.

Maybe is the memory modules you have ...

My bad.
Switched to these file managers back in the day and still on them today.
Norton Commander / Dos Navigator / Windows Commander for retro activities.
Total Commander and Krusader for modern day computing.

pshipkov wrote on 2024-06-03, 02:50:

Symphony Haydn with IBM BL3 running at 100MHz (3x33), all BIOS settings on max, produces 20.7 fps in Doom, which is half a frame slower than SiS Rabbit.
SiS Rabbit is clock-to-clock faster than Symphony Haydn with IBM BL3 processors, but slower with DLC/SXL2 and 386DX.
So, the results are inverted with 386DX, DLC, SXL2 processors.

What ISA frequency were you running on the Symphony with BL3-100? What value do you get at ISA=13.3 MHz?

My intention was to show that there are faster alternatives to DTK PEM-4036YB, PEM-4030Y, and Gemlight-Symphony for 40 MHz or less.. The Symphony Haydn-II isn't optimised for Blue Lightning chips, rather, it is optimised for SXL-class chips, whether intentional or by coincidence. On the other hand, if someone is wanting to build a fast SXL2-80 system, the Symphony Haydn-II is the way to go.

Did you ever have success in running the Daewoo Panda with an SXL2 at 80 MHz and an ARV1000VL? If so, what DOOM results did you achieve? It's on my to-do list.

Running ISA at 33/2 = 16.676MHz.

I thought we talked about baseline setup with 386DX. In this configuration the processor is too slow to keep the VLB cards busy, but once CPU upgrades come into play the story changes and the 386/VLB boards pull ahead.
These motherboards are great option for intermediate scenarios between the bottom (40-50MHz with 386DX) and top (110MHz - 2x50 and 110Mhz - 2x55 with BL3).
We can say that the 386/VLB boards go in this order:
ECS Panda 386V - one of the best 386 mobos really
DataExpert EXP3406
Alaris Cougar II
Alaris Leopard C and Matra 486SLC2 are on the same level

Tried running ECS Panda 386V at 2x40 with your BIOS, but results are lower than 3x33.
Didn't spend a lot of time with SXL2 at 2x40 with these boards yet.
My only SXL2-66 is flaky and even 80MHz is difficult to achieve.

pshipkov wrote on 2024-06-03, 06:25:

Running ISA at 33/2 = 16.676MHz.

Thanks for clarifying. Sounds like the 4036YB would be even slower still than the SiS Rabbit with matched ISA speeds and the BL3-100.

pshipkov wrote on 2024-06-03, 06:25:

I thought we talked about baseline setup with 386DX.

My recollection was the 386DX baseline discussion was only for the SCSI bench charts. I didn't think we were talking about a 386DX baseline for motherboard peak performance. If we were, then your broad comparison of motherboards is correct. For the purposes of this immediate discussion, I was only looking at the fastest possible PGA132-only motherboard running at 40 MHz or less with any 386 class CPU upgrade. I apologise if were reading different chapters. If you are wanting to run a baseline comparison of all 386 motherboards and be as inclusive as possible, then I agree, run it with a bog standard 386DX-40. Of greater interest to me is the top performers with an SXL2 or BL3 for two scenarios 40 MHz or less, and 45 MHz+.

I haven't come across too many people lusting to run standard 386DX's in builds these days. Maybe they are out there, but my focus is elsewhere.

pshipkov wrote on 2024-06-03, 06:25:

In this configuration the processor is too slow to keep the VLB cards busy, but once CPU upgrades come into play the story changes and the 386/VLB boards pull ahead.

Agreed. VLB needs at least a DLC, preferably clock-doubled, to make headway.

pshipkov wrote on 2024-06-03, 06:25:

These motherboards are great option for intermediate scenarios between the bottom (40-50MHz with 386DX) and top (110MHz - 2x50 a […]
Show full quote

These motherboards are great option for intermediate scenarios between the bottom (40-50MHz with 386DX) and top (110MHz - 2x50 and 110Mhz - 2x55 with BL3).
We can say that the 386/VLB boards go in this order:
ECS Panda 386V - one of the best 386 mobos really
DataExpert EXP3406
Alaris Cougar II
Alaris Leopard C and Matra 486SLC2 are on the same level

Boils down to personal interest, availability, practicality, and one's objective. Some people think going above ~12 MHz ISA, or beyond 40 MHz FSB is absurd. I thought so 10 years ago.

I would be keen on seeing an SXL2 run successfully at 100 MHz.

pshipkov wrote on 2024-06-03, 06:25:

Tried running ECS Panda 386V at 2x40 with your BIOS, but results are lower than 3x33.
Didn't spend a lot of time with SXL2 at 2x40 with these boards yet.
My only SXL2-66 is flaky and even 80MHz is difficult to achieve.

That is curious. I remember user WGJ had run some tests at 1x40 with an SXL on the Panda board and found the system fussy about having two VLB cards installed at 40 MHz FSB. I suspect getting 80 MHz working well will make for an interesting challenge on the Panda. If successful, it will make for a faster alternative to Symphony Haydn-II when running SXL chips within FSB specs. From some past testing I did, I recall having issues with the Panda. It could run great at 1x, but switching to 2x had new sets of problems. I plan to revisit this shortly.

I decided to re-run some tests on the PEM-4036YB (Symphony Haydn-II) and Chaintech 340SCD (SiS Rabbit) for FSBs <=40MHz. I am utilising the fastest possible 386-class CPU upgrades, TI SXL2 and IBM BL3. I no longer have any loose GD-5434 cards, so I'm testing with a Tseng ET4000/w32i. As such, the numbers will be a little different from what was presented earlier. The BL3 is being run at 3x only.

Symphony Haydn-II
SXL2-80
ISA (MHz), DOOM (fps)
10.0, 20.18
13.3, 21.76
16.0, 22.64

Symphony Haydn-II
BL3-100
ISA (MHz), DOOM (fps)
11.1, 21.15
13.3, 22.27
16.0, 23.07

Symphony Haydn-II
BL3-112.5
ISA (MHz), DOOM (fps)
9.38, 21.41
12.5, 23.79
15.0, 25.04
18.7, system hung mid-way during timedemo. ET4000 cannot cope?

SiS Rabbit
SXL2-80
ISA (MHz), DOOM (fps)
10.0, 17.30
13.3, 18.98
16.0, 18.62 <-----------why the reduction in frame rate?

SiS Rabbit
BL3-100
ISA (MHz), DOOM (fps)
11.1, 20.67
13.3, 20.78 <----------- frame rate hardly budged.

SiS Rabbit
BL3-112.5
ISA (MHz), DOOM (fps)
9.38, 20.0
12.5, 23.25
15.0, 23.37 <----------- frame rate hardly budged.

From these numbers, the most glaring curiosity is how, on the SiS board, increasing the ISA too far actually hurts DOOM performance. Does anyone know why this is? This result is not typical. Is the BIOS adding wait-states? pshipkov, did you run into this phenomenon during your testing of your SiS Rabbit board? If not, would you be willing to share your BIOS image with me so that I can re-run the tests?

Due to this ISA slow-down effect, it is difficult to compare Haydn-II with the Rabbit. Nonetheless, and contrary to the latest comparison presented a few days ago, the Haydn-II with a BL3 yields faster result than the Rabbit, even for cases in which wait states are probably not added, e.g. at 11.1 MHz ISA, BL3-100, Haydn-II yielded 21.15 fps, while the Rabbit returned 20.67 fps. Similarly, at 12.5 MHz ISA, BL3-112.5, Haydn-II yielded 23.79 fps, while the Rabbit returned 23.25 MHz.

Pushing the ISA clock faster on the Rabbit board doesn't yielded appropriately scaled up performance. I'm hoping someone has some ideas as to why and how to "fix" this. In some cases, perf dropped at 16 MHz ISA compared to 13.3 MHz ISA.

If a user is wanting to keep with more conservative ISA frequencies, then the Rabbit and Haydn-II are fairly comparable when using the BL3, with Haydn having the slight edge. I had incorrectly recalled that the SiS had the slight edge over the Haydn-II when using the BL3. My bad. When in doubt, run the tests again.

If someone is looking for the fastest 386-class motherboard (no PGA168 sockets), while keeping at or under 40 MHz, then the Daewoo Panda is the winner due to the VLB slot. If your requirement is no VLB slot, then Haydn-II is next in line. If wishing to run over 40 MHz FSB, the PEM-4036YB is reportedly unstable.

What I did recall correctly was the extreme optimisation that Haydn-II had over any other 386-class motherboard WHEN using the SXL2 CPU. Looking at the data, the numbers speak for themselves, e.g. at Haydn-II at 10 MHz ISA, SXL2-80, yielded 20.18 fps, while the Rabbit only scored 17.3 fps.

We talked about this stuff a while ago, but never did a proper side-by-side comparison between the two chipsets at lower frequencies.
At least i didn't. The conversation was brief and somewhat scattered.
Back then the topic was in the periphery of my activities, with peak overclocking taking the central place.
Your findings align with my "world view" about Symphony Haydn II based motherboards - they are the high 386 watermark.
Btw, it will be good to present a few more of the standard tests, so the outcome is more convincing.

I am SiS Rabbit motherboards-less these days. Sold the 2-3 ones i had a while ago, so cannot cross-check you on the 16MHz ISA issue you hit.
So weird. Wonder what is tripping it ?

Agreed with the rest of your notes.

Daewoo/ECS Panda 386V is indeed the best option at ~40MHz.
DataExpert EXP3406 close second.
Technically, Alaris Cougar II can challenge them in some cases, but best of luck making it stable at 3x33 and beyond for that to happen.

pshipkov wrote on 2024-06-04, 20:37:

it will be good to present a few more of the standard tests, so the outcome is more convincing.

I know. Already spent many hours getting these results and didn't want to go any further right now. I'll likely revisit this if or when I get 256K working on my C&T board. More interesting for me right now is to investigate the Daewoo Panda with SXL2-80 and VLB graphics. I think WGJ mentioned that graphics only worked in a particular slot at 40 MHz FSB.

pshipkov wrote on 2024-06-04, 20:37:

I am SiS Rabbit motherboards-less these days. Sold the 2-3 ones i had a while ago, so cannot cross-check you on the 16MHz ISA issue you hit. So weird. Wonder what is tripping it ?

This is unfortunate. Did you save any BIOSes? Do you recall which brands you had? I might try some MR BIOS options to see if things can be improved upon.

pshipkov wrote on 2024-06-04, 20:37:

Technically, Alaris Cougar II can challenge them in some cases, but best of luck making it stable at 3x33 and beyond for that to happen.

Also, if the presence of a socket 2 doesn't bother you. That always prevented me from buying these boards. There was always the though, 'if you want it faster, just put in a 486'.

There still is the open question of our previous results with the GD-5434:

feipoa
SiS Rabbit (Chaintech 340SCD)
BL3-100 (3x33)
ISA 13.3 MHz
GD-5434
DOOM = 21.27 fps

pshipkov
Symphony Haydn-II (DTK PEM-4036YB)
BL3-100 (3x33)
ISA 16.67 MHz
GD-5434
DOOM = 20.7 fps

It hints that the Rabbit might be more optimised for the GD-5434 compared to the ET4000/w32i. I may need to re-run the Rabbit/Haydn-II BL3 comparison with the GD-5434.

Daewoo/ECS Panda 386V:
Don't remember having issues with which slot the graphic cards are inserted in. Maybe some specific cards ?
The board however is not very happy with Ark1000VL and tight BIOS timings. S3 Trio64 is a great substitute.
Apart from that, the board is almost boring. In the best sense of the word - it just works.

SiS Rabbit BIOSes:
They are all interchangeable .
Chaintech 340SCD offers the best microcode.
That's what you want. You have it.
Didn't keep the rest.

Alaris Cougar II:
You are not missing anything there.
Slow 486 assembly.
Better out-of-the-box than your average vanilla 386 ISA motherboard. Way behind the hotrods.
And that's about it really.

Your previous and recent Chaintech 340SCD test results:
Can be the VGA, but i think that's unlikely.
You will have to spin this few more times to increase confidence.
Sometimes i get this - come back to an use case after some time, things are different now, and i am unable to replicate what was before.

The Panda also doesn't like the Trio64V+ VLB card. I'm using the S3 968 VLB replica card now and will run through all my VLB cards. Currently testing with only one VLB card installed, HDD I/O thru ISA SCSI.

Slightly better luck with graphics in the VLB slot nearest the PCB edge.

Plan on taking the Panda up to 50 MHz FSB or more. EDIT: Forgot we are still dealing with a 386 class motherboard, meaning that the clock values from MX8315 are already double the FSB. The best way to obtain 50 MHz FSB is to find a pin-compatible PLL with a 100 MHz frequency option. Do any exist in DIP format?

Some trouble I've run into with the VLB graphics card on this motherboard at 40 MHz is that the system will hang on the VGA BIOS when cold. The best way I've discovered to circumvent this is to first boot up at 33 MHz FSB, just to the point that you see the memory count, then power down, set to 40 MHz FSB, power up. Weird.

feipoa wrote on 2024-06-06, 01:54:

Plan on taking the Panda up to 50 MHz FSB or more. EDIT: Forgot we are still dealing with a 386 class motherboard, meaning that the clock values from MX8315 are already double the FSB. The best way to obtain 50 MHz FSB is to find a pin-compatible PLL with a 100 MHz frequency option. Do any exist in DIP format?

Three ways you can go...

Wiretap did a convertor board for the SMD ones...
Project: Full Can Clock Oscillator Replacement

Info to roll your own...
DIP-14 OSC replacement with jumpers?

Or user galonpu has a settable clock crystal replacement, link to store in his sig...
galanopu
Edit: Oh you might wanna see his youtube vid about it https://youtu.be/9_4tpY_ipBI?si=BrPaC5cN1dz_7eQV

Unfortunately the clockgen on the Panda 386V motherboard maxes out at 80MHz = 40MHz to CPU.
That's why i was thinking of inserting a contraption for DIP-14 crystal oscillators.
Not going to look pretty.

BitWrangler, I must have missed that wiretap link. I'll read it in more detail when I have time. Looks like we programme the frequency via an I2C header. This path is the way to go as it offers more control. Seems like more PLL's are abandoning the hardcoded multiplier values for this I2C programmability.

Nonetheless, the Daewoo Panda board uses a PDIP-14, MX8315 clock generator, more common on PCI socket 3 boards, thus these hobby clkgens aren't exactly plug-in replacements. Trying to use them on the Panda board will create a jumper wire mess and introduce noise. Not sure I'd want to go that route. I'll let pshipkov explore that... then maybe we can hit up wiretap for a PCB that plugs in to a PDIP14 and mimics the core pinout of the MX8315.

A few results from the Daewoo AL486V-D "ALi Panda" with SXL2-80 I wanted to dump. I don't think these have been posted elsewhere yet. This board is a little fussy with VLB graphics cards at 40 MHz FSB. AHA-1522B SCSI ISA. ESS ES1868 ISA sound.

STB PowerGraph 64V (replica), based on Trio64V+ ---------> hangs on VGA BIOS screen
Orchid Kelvin 64 VLB, based on GD-5434 ---------> screen stays blank, or beeps
Triden TGUI9440-1 ---------> beeps
ARK 1000VL ---------> runs, but had some minimal artefacting in DOOM

With the wrong set of DRAM sticks, I had to set:
DRAM Read Wait: Fast
DRAM Write Wait: Fastest
Cycle Check Point: Normal

With the right set of DRAM sticks, and graphics card, I could run:
DRAM Read Wait: Fastest
DRAM Write Wait: Fastest
Cycle Check Point: Fast
Slow Refresh: 120uS
VL Device Ready Sync Mode: Bypass

However, for the tests below, I'm running more conservative with:
DRAM Read Wait: Fast
DRAM Write Wait: Fastest
Cycle Check Point: Fast

AT bus was at 20 MHz: CLK2/4
Slow Refresh: 30 uS
Hidden Refresh: Enabled
TI486SXL2: Clock-doubled: Enabled
L2 cache mode: WB
Shadow Cacheable: Enabled

Non-CPU Cycle RAS-CAS Delay: 4t
Polling Clock Select: CLK2/3
DMA Clock: ATCLK/2
ISA Write Cycle Insert WS: Enabled
16-Bit ISA I/O Command WS: 1ws
16-Bit ISA Mem. Command WS: 1ws
VL Device Ready Sync Mode: SYN
Delay ISA Cycle LDEVJ WS: 3 CLK2

DOOM timedemo 3 results
SPEA Mercury P64V, based on S3 968 --------->24.81 fps
Diamond Stealth 64, based on Trio64 --------->24.81 fps
Diamond Stealth 32, based on ET4000/w32p --------->24.91 fps
ATI Graphics Xpression DRAM, based on Mach64 ---------> 23.0 fps
Cirrus Logic GD-5428 --------->23.73 fps

I did the majority of testing with the SPEA Mercury P64V. From a cold boot, I had to first boot with 33 MHz FSB. Power-off, then switch to 40 MHz FSB. This can be accomplished with a single-pole, double-throw switch (most turbo buttons). Improving some select BIOS settings,

DRAM Read Wait: Fastest
DRAM Write Wait: Fastest
Cycle Check Point: Fast
Slow Refresh: 120uS
VL Device Ready Sync Mode: Bypass

Resulted in a DOOM score of 25.72 fps for the SXL2-80.

I tried running Windows NT 3.51, setting resolution to 1280x1024x65k, running various apps. So far, so good. However, at this resolution, and running HDD through ISA, you can feel the mouse stutter with heavy HDD access. One must put HDD access onto the other VLB slot, however it will likely require degrading the above BIOS values.

Next, ran tests at 3x33 with the IBM BL3. Used the most optimal settings I used with the SXL2-80. BL3-100 (3x33), Doom = 27.2 fps. That's 1.5 frames/sec faster than the SXL2-80. I feel if we can get this motherboard to run with a 45 MHz FSB, the SXL2-90 would best or match the BL3-100.

Then performed some testing with the BL3 at 2x40. To boot to DOS, I had to set VLB mode to SYNC. I ran DOOM, but performance was poor at 23.09 fps. Played around with the settings, but could not get the BL3 at 80 MHz to boot Windows NT 3.51. This probably rules out the prospects of 2x50 with an IBM Blue Lightning. Pshipkov, were you at all successful at 2x40 w/BL3 on ALi Panda boards?

Small side note: this board with accept up to 128 MB DRAM, but will required reduced performance settings.

I think I got a boot at 2x50 on the Patriot SL4000 BL3 system, gotta get back to that, but all kinds of flakey started happening, then that sent me on some wild goose chases into the turbo switch problem. Anyway, what seemed to be the case from that, going purely on gut feels was that bus noise got high. So potentially if you can think of mitigations for that, you might get 2x50. I do need to do some proper testing still though.

To your point about running the tests with some FAST dram timings ?
Was that because DOOM didn't work with FASTEST, or some other reason ?
Will link your message above from my older Panda "showcase" post.

BitWrangler - send a pic of that Patriot SL4000 rig. Don't know how to think about it.
Is it some proprietary hardware inside, or something conforming to the AT standards ?

Whoops, misnamed it, it's SL4100 ... I was sure I'd posted a board pic before, but nope, not in my attachments ever. Seems to be some variety of LPX style, using a riser...

pshipkov wrote on 2024-06-08, 17:55:

To your point about running the tests with some FAST dram timings ?
Was that because DOOM didn't work with FASTEST, or some other reason ?

The fastest settings are:

DRAM Read Wait: Fastest
DRAM Write Wait: Fastest
Cycle Check Point: Fast
Slow Refresh: 120uS
VL Device Ready Sync Mode: Bypass
AT bus was at 20 MHz: CLK2/4

These fastest settings worked fine with SPEA Mercury P64V (S3 968) and SXL2-80. Ran beautifully in DOOM, NT 3.51, all NT apps, etc. I didn't have any concerns with its stability.

I ran my pile of VLB cards at <Fast, Fastest, Fast, 30us, Sync> because I hadn't yet narrowed in on the most optimal settings for all cards as a collective. I didn't want to run some VLB cards at <Fastest, Fastest, Fast, 120us, Byp>, then others at <Fast, Fastest, Fast, 30us, Sync>, just to have to rerun all the tests at <Fast, Fastest, Fast, 30us, Sync>. It is possible that they all run at <Fastest, Fastest, Fast>, but I didn't check for it.

The fact that the SPEA Mercury P64V worked well was good enough for me. I know some of these VLB cards won't work with VLB Bypass mode, but the P64V did.