finishing up my testing of this board, there are still a couple of things i wanted to take a look at. tests in this post will be from a p133 again unless noted differently, 32mb edo and the 4 mb matrox millennium, however this time everything at default memory timings. everything with f8 boot to dos 7, no drivers loaded. first, WB vs. WT cache operation;
speedsys L2 average: 92.02 mb/s (WB), 89.07 mb/s (WT)
pcpbench default: 16.5 fps (WB), 16.3 fps (WT)
doom (doombench method): 71.81 fps (WB) vs. 70.46 fps (WT)
so WB has a slight edge, running WT instead doesn't seem to be a huge issue if the larger cacheable area was really needed though. next, the jumper to run the PCI bus at an asynchronous 32 mhz. first, with the CPU underclocked to 75 mhz:
doom (doombench): 48.15 fps (25 mhz), 51.68 fps (32 mhz)
quake demo1: 17.7 fps (25 mhz), 17.9 fps (32 mhz)
3dbench2: 68.6 fps (25 mhz), 72.4 fps (32 mhz)
shiny gpt14: 37-38 (25 mhz), 38 (32 mhz)
pcp default/vga: 10.6 fps / 21.2 fps (25 mhz), 10.9 fps / 21.2 fps (32 mhz)
vspeed: 26.17 mb/s vga, 30.6 mb/s LFB for both
speedsys VESA transfer rate: 30478 kb/s (25 mhz), 30611 kb/s (32 mhz)
then with the full 133 mhz:
doom: 71.88 fps (33 mhz), 70.99 fps (32 mhz)
quake: 27.7 fps (33 mhz), 27.6 fps (32 mhz)
3dbench2: 104.9 fps (33 mhz), 103.8 fps (32 mhz)
shiny gpt14: 62-63 both
pcp default/vga: 16.5 fps / 33.8 fps (33 mhz), 16.4 fps / 33.8 fps (32 mhz)
vspeed vga/lfb: 35.76 mb/s / 40.18 mb/s (33 mhz), 36.13 mb/s / 40.18 mb/s (32 mhz)
speedsys VESA transfer rate: 36827 kb/s (33 mhz), 36882 kb/s (32 mhz)
it seems that no tool is able to read out the actual PCI bus clock, but judging by the results, the jumper works as imagined - 32 mhz for 50/60 mhz FSB, FSB/2 for 66 mhz will give the best performance. i didn't expect the raw transfer rates to be unaffected, though; unlike actual applications PCI bus clock doesn't seem to be a bottleneck at all here and this is only affected by CPU and FSB clock speeds.
finally, i benchmarked another asus board from the exact same late 1995 time frame, a p/i-p55tp4xe with the 430fx chipset. default bios settings again, which appear to be relatively fast here, and same hardware used, the board likewise has a 256kb asus COASt. unlike the other board which has the latest 1998 bios, i've left this board at its original november 1995 version.
initial results were all over the place though and i quickly realized the board was running without the L2 cache. took a lot of reseating to get the COASt recognized, in fact i've had the exact same issues with the p55sp4 board, just in that case only after trying to run a 512 kb stick in it. in many cases the board would just not post but work after moving it a tiny bit. what is a good method to clean these COASt modules and slots, contact cleaner perhaps? anyway, since i have results with L2 off for the p55tp4xe, might as well post them:
doom: 66.86 fps
quake: 30.2 fps
3dbench2: 112.2 fps
shiny gpt14: 43
pcpbench default/vga: 15.2 fps / 30.2 fps
vspeed vga/LFB: 42.39 mb/s / 51.61 mb/s
dukebench vga: 57.5 fps
speedsys vesa transfer rate: 47009 kb/s
and finally, comparing the p55sp4 against the p55tp4xe, as before with the p133 and 256 kb PB cache:
doom: 71.88 fps (p55sp4), 73.36 fps (p55tp4xe)
quake: 27.7 fps (p55sp4), 34.4 fps (p55tp4xe)
3dbench2: 104.9 fps (p55sp4), 117.5 fps (p55tp4xe)
shiny: 62-63 (p55sp4), 64 (p55tp4xe)
pcp default/vga: 16.5 fps / 33.8 fps (p55sp4), 17.5 fps / 36.8 fps (p55tp4xe)
vspeed vga/LFB: 35.76 mb/s / 40.18 mb/s (p55sp4), 44.97 mb/s / 51.61 mb/s (p55tp4xe)
dukebench vga: 68.5 fps (p55sp4), 73.5 fps (p55tp4xe)
speedsys vesa transfer rate: 36827 kb/s (p55sp4), 47009 kb/s (p55tp4xe)
speedsys cache/memory for the p55sp4:
┌──────────────┬──────────────┬──────────────┬──────────────┐
│ Read │ Write │ Move │ Average │
┌───────────────┼──────────────┼──────────────┼──────────────┼──────────────┤
│ Cache Level 1 │ 169.22 MB/s│ 42.07 MB/s│ 503.46 MB/s│ 238.25 MB/s│
│ Cache Level 2 │ 107.11 MB/s│ 41.76 MB/s│ 127.19 MB/s│ 92.02 MB/s│
│ Memory │ 80.75 MB/s│ 42.71 MB/s│ 59.25 MB/s│ 60.90 MB/s│
└───────────────┴──────────────┴──────────────┴──────────────┴──────────────┘
speedsys cache/memory for the p55tp4xe:
┌──────────────┬──────────────┬──────────────┬──────────────┐
│ Read │ Write │ Move │ Average │
┌───────────────┼──────────────┼──────────────┼──────────────┼──────────────┤
│ Cache Level 1 │ 169.23 MB/s│ 84.07 MB/s│ 503.57 MB/s│ 252.29 MB/s│
│ Cache Level 2 │ 107.12 MB/s│ 83.35 MB/s│ 127.20 MB/s│ 105.89 MB/s│
│ Memory │ 69.49 MB/s│ 83.54 MB/s│ 61.12 MB/s│ 71.38 MB/s│
└───────────────┴──────────────┴──────────────┴──────────────┴──────────────┘
the write numbers are doubled on 430FX across the board, even down to L1 cache, and this should be reflected in some of the practical results - i was rather surprised to see 430FX outperform sis in quake and 3dbench2 even without l2 cache. in comparison, the gpt14 number dropped to only p90 level according to that benchmark, suggesting that this is the most L2 reliant benchmark of all of those. on the other hand sis does actually have an advantage over intel in terms of memory reads, though, at least at default timings for these two boards.
transferring over the same win95 install to the 430FX board, the driver for every component was installed without issue, except for one thing - the floppy controller of all things, which has a yellow exclamation mark in device manager, so A: is now shown to be using ms-dos compatibility mode. does anyone know how to get this installed without doing a full win95 reinstall? it's not shown to be actually using a driver and after deleting it from device manager, it reappears in the same state.
