I appreciate the use of a white colour LCD on retro rigs. It looks as if yours has built-in speakers on the bottom which is good for testing. I think a Win98SE install will help confirm the cache works. I would feel the cache with your finger when the system is operating (ground yourself of course). If any module feels hot, I'd replace it. From my experience, hot ones are always bad, however cool ones could be bad or good. Hot pieces may damage your chipset. Hot is greater than 50 C.

As you pointed out, things look OK for speed, except for the RAM speed, which is pretty slow. I'd try to source four 16 MB sticks of FPM and EDO RAM. Sometimes 486 boards don't work all that well with EDO, so I tend to keep several configuration options around of both EDO and FPM. With 4 sticks, the chipset can usually interleave the RAM (improves performance), however from my experience, the most aggressive cache/RAM timing might not work, but it is always good to try first. On my UMC board, I actually found I could use faster timings and have greater stability with only a single stick of 64 MB. With single-banked cache and RAM, the chipset is not interleaving though.

The Asus PVI-486SP3 w/512 KB WB cache makes for a very good competitor with the Biostar MB-8433UUD.

Both have PS/2 ports, 3 PCI ports, 512 KB WB cache. The Biostar will work with 256 MB of RAM and the PVI-486SP3 should work with at least 128 MB of RAM -- though this doesn't really matter, the cacheable limit is stuck to 64 MB in WB mode.

The real differance between the two boards is the use of a SiS chipset on the Asus board as opposed to a UMC chipset on the Biostar. The best all-around graphics card (2D+3D) to work with UMC chipsets seems to be the Matrox Millennium G200, however I've seen accounts of SiS chipsets working with a Voodoo3 (PCI). While both of these graphics cards are CPU limited by a 486, I'd like to know if the Voodoo3 squeezes out any more performance compared to the Millennium G200. When you have a moment, would you mind running these on the two cards?

Quake1 timedemo1
Ziff-Davis Winbench96 Graphics WinMark
Ziff-Davis 3DBench 97
WinTune98
Sandra99
and QuakeII Timedemo1 in OpenGL mode.

Unfortunately, I have been unable to source a reasonably priced SiS board for this comparison. I have a Voodoo3 3000 (16 MB) PCI and a Matrox Millennium G2oo (16 MB) PCI, but no SiS.

@megatron-uk: Nice work, AFAIK this is the first report of a 486SP3 running with more than 256k cache. 😉

feipoa wrote:

I'd try to source four 16 MB sticks of FPM and EDO RAM.

Two problems: The PVI-486SP3 only has two RAM slots, and it won't support EDO.

feipoa wrote:

I'd try to source four 16 MB sticks of FPM

5u3 wrote:

Two problems: The PVI-486SP3 only has two RAM slots

I have two 32MB FPM sticks, which are basically sitting here doing nothing...

*whistles*

feipoa wrote:

... When you have a moment, would you mind running these on the two cards? […]
Show full quote

... When you have a moment, would you mind running these on the two cards?

Quake1 timedemo1
Ziff-Davis Winbench96 Graphics WinMark
Ziff-Davis 3DBench 97
WinTune98
Sandra99
and QuakeII Timedemo1 in OpenGL mode.

I can certainly give the quake1 test a go - but I don't have Windows installed on this machine to try the rest - just the Win98 dos components.

I'll see if I can find any 60ns modules - my Amiga 1200 accelerator has a couple of 72pin simms, I'll see what speed those are.

5u3 wrote:

Two problems: The PVI-486SP3 only has two RAM slots, and it won't support EDO.

That's what happens when I bypass board layouts and look directly at specs :s-face:
Sounds like you are limited to 2 x 32 MB, or perhaps 1 x 64 MB will work.

If you max out on your BIOS RAM settings with the 70 ns RAM, does the system not function properly?

With the memory speed set to 'fastest' in the BIOS it won't POST fully, or, on the rare times it does, then it won't boot.

I may have a result though - the Blizzard 1260 accelerator in my Amiga 1200 has a 64mb 50ns simm and a 60ns 64mb simm installed. It really doesn't need 128mb. It's a bit late here now so I'll try the tests again with the 50ns memory tomorrow evening.