Well, I was playing around with my C4PI Rev A01 motherboard and the one AM5x86 I have that does 180 MHz reliably (doesn't need much overvoltage either, 3.5V seems fine, and will also do 160 at 3.1V). It's an SiS 496/497 board that supports 3.1 to 4.0V (plus probably a 5V setting I don't know) and looks to support up to 1 MB of L2 cache, but for the life of me I haven't seen any documentation for the jumper settings, and there's some 30+ jumper positions in the cache block.
It supports 33, 40, 50, 60, and 66 MHz FSB (I think it does 25 and 30 MHz too, I haven't formally documented all the configurations) It also needs the turbo switch jumpered closed for normal/full speed mode.
Thanks to another ebay listing of the same board with 256 kB cache instead of the 128k of mine, I've got it working at 256kB now with cache timing at fastest (2 clock read, 2 clock write, 1 clock burst). I had to change 4 of the jumpers to do this, I think (and missing one reported 256k but then wouldn't boot from the flash disk). 15 ns cache + tag seems sufficient for the fastest timings at 60 MHz fsb, but none of my DRAM is fast enough to completely max out.
The BIOS doesn't allow WB cache to be enabled on the 5x86, so I'm in WT (I've got 7-bit tag enabled for fast WB L2 cache, though). I may be able to enable WB L1 via a register tweaking utility in DOS, though.
It's stuck at the 3x multiplier setting for AMD/Cyrix/Intel CPUs and the BIOS seems to have issues with any of my Cyrix 486DX4s or 5x86s set to 3x mode (DX2s seem fine, even though the feature sets are the same as the DX4) so I haven't been able to try anything at 2x66 MHz yet, though I think I'd need to up to 12 ns cache (or at least tag) to get the fastest timing at 66 MHz. (only thing I've run that fast on the board is a couple DX-50s and an SX-33, same for the Acorp 496/497 board I've got, though that was using 5V as it only seems to have 3.46 and 5V settings ... I also think I damaged the sockets pins with the overly-long legs of the QFP-board mounted 5x86 I've had in it as no other CPU will post now: they're overly long and thick/rigid legs, like machined PGA socket pins, but longer, and I think I forced it in too deep)
Anyway, with a recent batch of tests I ran with a few different video cards (including a Pedmidia 2 to compare with CPU Galaxy's results) and it seems to be dead even with several other cards in 320x200:
Matrox Millenium II, Matrox G450, S3 Virge 325 and 375 (albeit they give palette errors in quake), Voodoo 3 2000, Trio64, and (somewhat surprisingly) my sole Diamond Stealth Trio32 card. The 640x480 PC Player scored varied more and the Permidia 2 only got 11 FPS and Trio 32 10.5 while the others all gave 12.5. Even more surprisingly, that Trio32 didn't give palette errors in quake at 60 MHz PCI and worked fine, but was stuck in B&W mode for the DOS shell and anything that sticks to the default palette (I think) like 3D Bench and some game menus and some games, so it's not perfect at 60 MHz. (it's also the oldest card I had working in it, week 1 1995 date code on the chip, older than the week 11 1995 motherboard chipset date)
I had a few other cards that would POST but not display video, and a Rage XL that wouldn't display quake, plus an MPact2 3DVD that would POST and then go into something other than 60 Hz at the POST table and in DOS, so I got no video on my old Gem LCD monitor (I assume it's defaulting to 72 Hz or something, even though 480 line VGA resolutions are traditionally 60 Hz). I've got a cramped makeshift testing area for the time being, so lugging out a CRT wasn't worth the trouble. (Geforce MX400 and a pair of TnT2s wouldn't work, and more surprisingly a Tseng ET4000 PCI wouldn't work)
A pair of cheap little SiS 6326 cards didn't do that much worse than the fastest ones (19.0 FPS Quake, 28.0 FPS PC Player, 12.2 FPS PC PLayer 640x480) surprisingly. Though video signal quality and colors were worse. (picture quality was best with the Millenium II, plus the monitor seemed to auto-calibrate better with that, conveniently)
I've got some Cachecheck and Speedsys scores, too. (and screenshots I can upload later)
I don't think I've beaten any records, just matched some and beaten some old records (like the 19.4 FPS quake score)
Short of getting faster RAM (I have some 50 ns EDO and one pair of 45 ns EDO SIMMS, but none of them work in this board, FPM only) or finding the jumper settings for 1MB cache, I think enabling WB for the L1 would be the main improvement possible in this board.
Well, that or maybe some further tweaking since there's a spot for a crystal oscillator on the PCB that might be for a fixed/discrete system bus speed rather than the typical programmable clock oscillators used (a feature I've mostly seen on 1993 and older boards), but I'm not sure I'd want to try to modify the board in that way, or if it would even be possible without more jumper documentation. (or laborious pin and trace analysis) But given it natively does 66 MHz already, this seems like a non issue for anything but maybe 2x 68-75 MHz. (I've got an ST 486 DX4s that runs at 3x48 at 5V without issue, so 2x72 might be a possibility; 3x50 is too much for it, though)
See preliminary results below:
All from Phill's DOS Benchmark pack v 1.4
Scores for all the fast cards (but you can read that as Matrox Millenium II)
320x200 scores:
Quake: 19.6 FPS
PC Player: 28.3
3DBench 2: 111.0
Doom: 1168 realtics
other:
PC Player 640x480: 12.5
Quake 360x480: 9.2
Note: Quake 640x480 would not work (defaulted to 320x200) and Quake doesn't give options for resolutions above 360x480 for some reason in my setup.