I made some more measurements.
90mV ripple on +5V rail (measured on decoupling caps near CPU socket) hence no recapping needed.
Removing second 32MB SIMM doesn't affect FPS in Quake. Either it fits in lower 32MB or 256kB of WB cache is enough for 64MB.
These SIMMs are same model and same year but one of them is more stable @50MHz FSB than another. Looks like I'm trying to push their limit.
Board freezes right after POST @60MHz FSB no matter what SIMM I use. Probabbly I just don't have fast enough ones.
3x50 is still noticeable faster than 4x40.
4x50 doesn't start at all @+3.45V.

When on the border of stability with very fast timings at high FSB, having only one memory module installed is preferred to two modules, irrespective of how much total RAM is installed.

If you are unable to POST at 60 MHz FSB, even with slow wait states, e.g. 2ws/2ws EDO and 3-2-3 SRAM, then try higher voltage to the CPU. I have two LSD boards and both can do 2x60 and 3x60 MHz, but it requires some trial and error. From my experience, 60 MHz needs 4 V to the CPU.

froller wrote on 2024-03-11, 12:58:

I made some more measurements. 90mV ripple on +5V rail (measured on decoupling caps near CPU socket) hence no recapping needed. […]
Show full quote

I made some more measurements.
90mV ripple on +5V rail (measured on decoupling caps near CPU socket) hence no recapping needed.
Removing second 32MB SIMM doesn't affect FPS in Quake. Either it fits in lower 32MB or 256kB of WB cache is enough for 64MB.
These SIMMs are same model and same year but one of them is more stable @50MHz FSB than another. Looks like I'm trying to push their limit.
Board freezes right after POST @60MHz FSB no matter what SIMM I use. Probabbly I just don't have fast enough ones.
3x50 is still noticeable faster than 4x40.
4x50 doesn't start at all @+3.45V.

According to your screenshot, you use WB and 8bits for TAG RAM. Yes, that is sufficient for 64MB cacheable area.
But it is not a good setting performance wise.
Therefore I suggest to use either WT (then cacheable area is 64MB)
or WB, but tag ram 7bits (then the cacheable area is only 32MB with 256kb of cache).
I assume with both settings, you get higher values. Do not forget to reduce the RAM with WB and 7 bit.
When you have managed to get better values, you can re-run the comparison between 150 and 160MHz.

pshipkov wrote on 2022-05-07, 04:25:

https://www.petershipkov.com/temp/retro_pc_images/benchmarks/486_ls-486e_rev_d_q1_200_2-1-2.png A video proof. […]
Show full quote

A video proof.

Details in the last section of the original post explaining the upgrade from 180 to 200 MHz on air cooling.

A little bit ago I emulated the mods pioneered by pshipkov & others on this forum for the Lucky Star LS-486 Rev D. Was able to squeeze out another 0.1 fps by using a 32mb LGS simm that runs at "fastest". However, the board still has the same drawbacks as described before, such as needing a PCI 1/2 bus divider for stability.

I have found pretty much all of these SiS 496 boards to be unreliable with the PCI at 1/2.

Very cool @jonf.
Share a photo of the memory module.
32mb at "fastest" is rare, unique i would say.
I think this alone gave you the +0.1 fps.
Will update the charts for your achievement here.

@feipoa
At 3x66 1/2:1 is the only way to stability for LSD and similar boards.

pshipkov wrote on 2024-03-22, 22:43:

At 3x66 1/2:1 is the only way to stability for LSD and similar boards.

I can believe that. I didn't do much testing at 66 MHz because the floppy controller on LSD doesn't work at this FSB.

Otherwise, for PCI: 1/2 FSB, my notes indicate the following:

Could not use my Am5x86 QFP on interposer with PCI at 1/2. Must use ceramic PGA chip. Strange.

ISA sound would not function with PCI at 1/2. Tried several ISA sound cards. Tried two motherboards, same issue. Winamp hangs up immediately. pshipkov did not witness this issue. Since I have some DOS Quake scores with sound, I think the sound issue may have been limited to Windows 95. I don't recall.

Promise PCI IDE controller hangs up with PCI at 1/2. However, it will work if setting PCI Master Burst Read/Write to disabled. This reduces throughput by 70%. I think there was some other performance hit as well, but I forget.

Even onboard IDE not working well at 1/2 PCI. I don't recall the symptoms, and I don't recall.

I spent almost a year on this motherboard trying to get 180 MHz working well, but eventually gave up in favour of the M919.

pshipkov wrote on 2024-03-22, 22:43:

Very cool @jonf.
Share a photo of the memory module.
32mb at "fastest" is rare, unique i would say.

Here are a couple pictures, it is a pretty nondescript module