feipoa wrote on 2021-08-01, 13:47:

pshipkov, did you find your LS-486E rev.D unable to save the CMOS year passed 1999 ? When I put in 2021, it reverts back to 1996. It will save if I input a year of 1999 or earlier.

Also, I have performed the PS/2 mouse mod, however I am currently unable to get the PS/2 mouse recognised. I've checked all the connections and am using an AMIKEY-2. Have you done this mod?

Yes one of the lucky star BIOS revisions does that, I can't remember which. But I do have a PS/2 modded version that does not have the y2k problem I can send you for that board. Or you can just use the modified zida tomato BIOS I sent you before

I had the same issues with the tomato mart bios. What’s this other bios you have for Rev D ?

feipoa wrote on 2021-08-01, 14:24:

I had the same issues with the tomato mart bios. What’s this other bios you have for Rev D ?

Really? that one wouldn't take dates >2000 for you? That's bizarre I use it currently and it works fine for me.

The other one is just an updated luckystar release I found online that I ps/2 enabled with modbin

Yes, the late rev D BIOS does not retain dates above year 2000 properly between restarts.
Didn't look at PS2 mouse support. Still in the process of selecting the right PCI/ISA motherboard for a proper late 486 PC running at 180MHz.
Once settle on one of them - will have to look at the mouse thing one way or another. If you didn't solve the issue until then, hope i can help.

Btw, while going through these late boards i am casually checking how they do at 200MHz.
One of them did great this morning. PC-Chips M918i rev 1.2.
Check this out: https://www.petershipkov.com/temp/retro_pc_im … _200_quake1.mp4
Quake 1 busting 21.8 fps - i believe this is the highest achieved score with 486 CPU so far, despite using some janky 60ns 8Mb EDO RAM that was on top of the stack.
And the system is actually long term stable, well, apart from the icing from the full 12V to the Peltier element needed to freeze the CPU.
Wonder what Asus VLI will do to these numbers. It is itching me to try ... 😁

I went back and checked and the lucky star BIOS I modded for PS/2 that doesn't have the Y2K problem is from rev C2. It's attached to this message

Thanks for the clarification and attachment, i was about to ask you for more details, because it was not clear from your previous messages.

The LS486E rev.D undertaking has been fairly disappointing and I think overhyped. No floppy at 66 MHz, no CMOS date storage past 2000, PS/2 mouse not working. Is there any consequence in using the C2 BIOS on rev.D? If the 4DPS BIOS doesn't work with date, wouldn't the C2 exhibit the same symptoms?

pshipkov, that's a nice score. When do you get your trophy?

I noticed you are using single-banked cache. Any reason for this? Are you at 256K single-banked? Did you try 1024K? I have this board and remember the L2 cache doesn't do much of anything, I think because the main memory response time is too fast to begin with.

I was not aware that LS486 rev.D had a fancy status. With that said - it is a nice little PCB.
C2 BIOS in rev.D = no EDO = problems at 66MHz and a bit slower at 60MHz (D likes EDO better than FPM at these frequencies).

Trophy ?
Too early - I am still in the prequalifications. 😀

Didn't spend much time with the L2 cache really.
What i tried:
EDIT: 256Kb double banked work fine. Used trusted 12ns ISSI chips. Unfortunately wait states have to be set to 3-1-1-1, or no post. (This has been improved upon in later developments - 2-1-1-1 works fine, needs specific chips configuration).
EDIT: 512kb single banked, but still interleaved (?), work fine too. Used 12ns ISSI tag and a mixture of 10/15ns ISSI chips. All BIOS settings on max (including 2-1-1-1 for L2 cache). The posted above quake 1 video used that configuration.

Few things to notice on the attached screenshots.
EDIT: Direct RAM access is actually pretty good. Double banked L2 cache at 2-1-1-1 only matches it, while 3-1-1-1 for double banked has negative performance net effect.
With extra effort and 10ns chips it should be possible to get 1024Kb double banked L2 cache at 200MHz, which should further improve perf.
It feels that the board is capable of handling it, but will burn that bridge when get to it.

These PC-Chips boards have some nice characteristics, but also, too much baggage if you ask me.

EDITS denoted inline and underscored.

I think the LS-486E rev.D's fancy status stemmed from ph4nt0m's posts, at least it started to interest me after his tests at 66 MHz. Which other motherboard works stable at 66 MHz besides the MB-8433UUD? Generally, SiS496 based boards have slightly faster memory throughput compared to UM8881 based ones, so I figured the LS-486E rev.D with PS/2, EDO, and an IBM 5x86c-133/2x might yield slightly faster results.

I have since been able to get the PS/2 working. For me, and AMIKEY-2 didn't work at all for the mouse, but a HT6542B and VT82C42N worked fine. It is a bit difficult to accept the CMOS year though. It will ultimately cycle through 1996, 1997, 1998, 1999, 1996, 1997, etc. You get those 4 years only. Plus you loose the onboard floppy. The other issue is PCI slot 1 is right infront of the CPU socket, so your choice of cards are very limited.

It is curious that you were able to run faster SRAM timings with 512K single-banked. These ALi PCI 486 boards were weird. Is 66 MHz stable?

Are your ISSI 10 and 12 ns SRAM modules the Chinese fake ones?

lol @ speedsys w/RAM faster than L2. I think either this chipset or these boards were made to work well without L2.

I gotta set my Bektronic V439 up again and see if I can get it to 66 stable by swapping the cache chips around until I find the golden one.

What's one of them? Bottom left...
download/file.php?id=115198&mode=view

feipoa wrote on 2021-08-02, 02:53:

I think the LS-486E rev.D's fancy status stemmed from ph4nt0m's posts, at least it started to interest me after his tests at 66 MHz. Which other motherboard works stable at 66 MHz besides the MB-8433UUD? Generally, SiS496 based boards have slightly faster memory throughput compared to UM8881 based ones, so I figured the LS-486E rev.D with PS/2, EDO, and an IBM 5x86c-133/2x might yield slightly faster results.

I see.

Well, 200MHz is actually much easier to do than 180. That is for 486 CPU at least.
4x50 and is done. The system runs within specifications and at tightest BIOS timings (assuming decent motherboard).
Performance is better than 3x66 which runs way out of spec and requires inflated wait states.
But 200MHz is kind of a toy, because of the 12V Peltier and related icing. Still good for bragging rights.
Real action is at 3x60 with 5V Peltier - can be a real, long term-stable, computer.
But i think you are after Cyrix 586 @ 2x66, right ?

feipoa wrote on 2021-08-02, 02:53:

I have since been able to get the PS/2 working. For me, and AMIKEY-2 didn't work at all for the mouse, but a HT6542B and VT82C42N worked fine. It is a bit difficult to accept the CMOS year though. It will ultimately cycle through 1996, 1997, 1998, 1999, 1996, 1997, etc. You get those 4 years only. Plus you loose the onboard floppy. The other issue is PCI slot 1 is right infront of the CPU socket, so your choice of cards are very limited.

Good to hear you sorted out the PS2 thing. Expect a call if i need to do something like that in the future.
Cannot remember when was the last time i used floppy, or even Gotek. The presence of CF card eliminates the need for it.
Worst case - one can switch for a moment to 4x40 if floppy is a must have for some operation. But yeah, not great.

feipoa wrote on 2021-08-02, 02:53:

It is curious that you were able to run faster SRAM timings with 512K single-banked. These ALi PCI 486 boards were weird. Is 66 MHz stable?

This board can do only 25, 33, 40 and 50 - verified very carefully.
There is a longer story here ...

I also didn't expect single banked L2 cache to handle such tight timings.
It will be really interesting when i give the mobo a proper 200MHz spin at some point later and see if 1024Kb double banked is possible and what the perf is.

feipoa wrote on 2021-08-02, 02:53:

Are your ISSI 10 and 12 ns SRAM modules the Chinese fake ones?

I have several different types of 24/28 pin ISSI chips. Some of them are 10ns and 12ns ones.
People call them fake, but they for sure tick better than most 15ns ones.
The difference is obvious in systems on the edge.

feipoa wrote on 2021-08-02, 02:53:

🤣 @ speedsys w/RAM faster than L2. I think either this chipset or these boards were made to work well without L2.

Yeah, it is weird. Cracked me up.

---

@BitWrangler
VLB running at 66MHz will be a tall mountain to climb, but 4x50 is usually ok. Maybe try that first.
Btw, how do you like this mobo ?
Can it actually to 66 FSB ?

Also, did you try the 386 hardware path of the UMC based board in the upper middle ? Looks interesting.

Yah, it took a lot of card swapping before I got it to 60Mhz back in 1996, it ran at 66, but I had to back off so many settings to get the Cyrix 5x86 100GP stable at 133 that it actually ended a point or two slower on benchmarks than 2x60. (It was 3 or 4 faster when it ran between freezups before I got it stable-ish, but not 100%) I didn't know then if it was the CPU or board. Back then I don't think I tried raising voltage any. Having a chat about that 386 board here... 386/486 boards? Motherboards with a socket for both a 386 and 486... though not got into real detail.

I know what you mean very well.
What video card you used ?
What I noticed (assume limited sampling rate) is that 60MHz is a stretch for VLB adapters.
We enter the inflated wait states zone and there goes your performance.
50mhz at 0 waits seems to be the good place.

BTW, ping me if you give those two boards we touched on a proper try. I will be interested in the findings.

Yes, I am looking for 2x66 with an IBM/Cyrix 5x86 at the moment. For an AMD X5 at 180 MHz, I'd probably want air cooling only. If the rev C2 board can't do 66 MHz, then it really doesn't interest me at all.

I've run further tests with the rev.D and an IBM 5x86-133/2x. I haven't done Windows 9x tests yet, but thus far I am getting 19.6 fps in Quake with EDO 64 MB. I'm running the CPU at 3.675 V by placing a 180 K-ohm resistor in parallel with SMD resistor R16 (36K) and am using the 4V jumper option.

• 32 MB RAM (19.2 fps) yields lower Quake results compared to 64 MB (19.6 fps).
  
  128 MB EDO SIMMs work, but only one at a time with this BIOS. I guess 128 MB is max, but perhaps the 4DPS BIOS can extend upon this.
  
  To use a Promise Ultra100 IDE controller, you must set "PCI Master Burst Read/Write" to Disabled.
  
  I unhid the CPU Burst Write option in the BIOS and set it to enabled. I'm not sure if it gives any performance benefit, but I recall from other SiS 496 boards that AMD/Intel 486s must have this disabled, while IBM/Cyrix 5x86 CPUs can have it enabled.
  
  L2 cache/DRAM cycle can be set to 2 CCLK if L2 is disabled. If L2 is enabled, it must be set to 3. At 3, DRAM read drops from 73.9 MB/s to 69.8 MB/s (cachechk), but you get 101.8 MB/s L2. A fair trade-off.

I am using these CMOS settings for now:

Any idea which of these settings can be improved if I swap out the TAG to 8 ns and the DATA SRAM to 10 ns?

pshipkov wrote on 2021-08-02, 05:48:

I know what you mean very well. What video card you used ? What I noticed (assume limited sampling rate) is that 60MHz is a stre […]
Show full quote

I know what you mean very well.
What video card you used ?
What I noticed (assume limited sampling rate) is that 60MHz is a stretch for VLB adapters.
We enter the inflated wait states zone and there goes your performance.
50mhz at 0 waits seems to be the good place.

BTW, ping me if you give those two boards we touched on a proper try. I will be interested in the findings.

I had a CL-5429 card that ran really well, but it got hot, so I had a heatsink on it. May have been a Pine Technology model. In theory, I should still have that one around somewhere, but have not seen it recently.

feipoa wrote on 2021-08-02, 12:50:

Yes, I am looking for 2x66 with an IBM/Cyrix 5x86 at the moment. For an AMD X5 at 180 MHz, I'd probably want air cooling only. If the rev C2 board can't do 66 MHz, then it really doesn't interest me at all.

Some AMD X5 CPUs can do it on air cooling ... half of the year.
During autumn / winter in cold room they can be stable in DOS and Windows.
Spring / summer - probably not.
Joking aside - so far i haven't seen a CPU pass some of the really sensitive tests at 180MHz with air cooling only.
Do you know of such case ?

feipoa wrote on 2021-08-02, 12:50:

I've run further tests with the rev.D and an IBM 5x86-133/2x. I haven't done Windows 9x tests yet, but thus far I am getting 19.6 fps in Quake with EDO 64 MB. I'm running the CPU at 3.675 V by placing a 180 K-ohm resistor in parallel with SMD resistor R16 (36K) and am using the 4V jumper option.

• 32 MB RAM (19.2 fps) yields higher Quake results compared to 64 MB (19.6 fps).

First sentence sounds pretty good, but the second one is confusing to me.
What you mean by 32-19.2 yields higher results to 64-19.6 ?

feipoa wrote on 2021-08-02, 12:50:

128 MB EDO SIMMs work, but only one at a time with this BIOS. I guess 128 MB is max, but perhaps the 4DPS BIOS can extend upon this.

In my experience the 4DPS/PCI400 BIOS is the same as LS C2.

feipoa wrote on 2021-08-02, 12:50:

To use a Promise Ultra100 IDE controller, you must set "PCI Master Burst Read/Write" to Disabled.

Interesting. I don't think i encountered this with 3x60 and AMD X5 CPU.

feipoa wrote on 2021-08-02, 12:50:

I unhid the CPU Burst Write option in the BIOS and set it to enabled. I'm not sure if it gives any performance benefit, but I recall from other SiS 496 boards that AMD/Intel 486s must have this disabled, while IBM/Cyrix 5x86 CPUs can have it enabled.

Can you pass the modified BIOS please ?

feipoa wrote on 2021-08-02, 12:50:

L2 cache/DRAM cycle can be set to 2 CCLK if L2 is disabled. If L2 is enabled, it must be set to 3. At 3, DRAM read drops from 73.9 MB/s to 69.8 MB/s (cachechk), but you get 101.8 MB/s L2. A fair trade-off.[/list]

Any idea which of these settings can be improved if I swap out the TAG to 8 ns and the DATA SRAM to 10 ns?

In the C2 board i used 12ns TAG. Varied multiple sets of 10/15 ns chips. At the end of the day the best i could achieve was 3-1-2, which is the same as what was used on the D variant. I tried and 10ns TAG for C2 (pulled it from the PVI rig), but no big difference. It may be a bigger factor at 3x66, but as i mentioned above 200MHz is better with 4x50, so even i ordered the chip types that Phantom4 used, i am not sure if i will use them at the end.

---

@BitWrangler
Thanks.
Yes, well put together 542x is a great grunt that can withstand whatever we throw at it.

Is it possible to get a water block to mount to the CPUs instead of using Peltier cooler and worrying about the ice build up?

feipoa wrote on 2021-08-02, 12:50:

To use a Promise Ultra100 IDE controller, you must set "PCI Master Burst Read/Write" to Disabled.

not entirely accurate - you can leave burst enabled if the U100 card is the ONLY bus master card in the system

@pshipkov here's the "tomato mart" BIOS that both feipoa and I have worked on

the CPU Burst Write option is unhidden, but be advised it does NOT work reliably on a true 486 core - Cyrix 5x86 only

it's originally from the Zida 4DPS but modified to support the LS-486E

maxtherabbit wrote on 2021-08-02, 21:14:

feipoa wrote on 2021-08-02, 12:50:

To use a Promise Ultra100 IDE controller, you must set "PCI Master Burst Read/Write" to Disabled.

not entirely accurate - you can leave burst enabled if the U100 card is the ONLY bus master card in the system

I had disabled the onboard IDE ports, so there weren't any other bus masters in the system that I was aware of. I had to disable the PCI Master Burst Read/Write option when using a 66 MHz FSB.

I also forgot to point out that I must set the PCI multiplier to 1/2 for the system to show a screen other than a blank one. I'm not sure why everyone else says contrary. Maybe I or you have mixed up the pin numbering on the jumper header? I did note that an AMD at 2x66 could have the PCI multiplier jumper on either 1:1 or 1:1/2 setting and it would boot up.