VOGONS


First post, by noshutdown

User metadata
Rank Oldbie
Rank
Oldbie

now i also have a small bunch of isa video cards, including:
stb cirrus5434 2mb(0wait)
gainward cirrus5429 2mb(0wait jumper omitted, may solder one someday)
ati mach64 dram 2mb(not sure if its 0wait)
ati 28800 wonder xl 1mb(not sure if its 0wait)
everex et4000ax vram 1mb(no 0wait, 256colors only)
generic et4000ax 1mb(no 0wait, 256colors only)
generic et4000ax 1mb(0wait and truecolor)
generic et3000 512kb
isa8bit et3000 with tvout 512kb(component blew up at power on, looking for repairman)
trident8900d 1mb(0wait)
trident8900c 1mb(no 0wait)
umc85c408 1mb(no 0wait)
oak037 256kb
ibm vga 256kb

cards just bought and waiting for arrival:
generic cirrus5424 1mb(0wait)
generic cirrus5426 1mb(0wait)
realtek3105 512kb
chips 82c450 512kb

cards that i am considering to buy and add to the test:
wd90c31, expected to be a match for the 5429
cirrus5401/avga1, fastest 256kb card

and some cards that i hoped to get, but couldn't:
s3 911/928/805
ati18800

i am going to test them on a umc8881+amd5x86-133 platform. the fastest isa platform that i have is 440bx+pentium3-1.1ghz, which is a lot faster, but i am too lazy to set it up, amd5x86 is a reasonable match for the isa cards anyway.
the purpose of test is to show the overall difference between cards, including both raw performance(d0s) and acceleration(windows9x).
the tests that i choose:
dos:
diag4.61
snoop
x-vesa
3dbench
pcpbench
chris3d
doom shareware
quake

windows:
wintach
wintune(video only)
toms2d
xing mpeg player

dos tests other than x-vesa would be run in vga mode 13h.
x-vesa and windows tests would be run with both 640*480*8bpp and 800*600*16bpp(for those with hicolor support).

do you think these tests would be representative enough, or that i should add and remove something?

Last edited by noshutdown on 2025-04-08, 03:39. Edited 3 times in total.

Reply 1 of 4, by noshutdown

User metadata
Rank Oldbie
Rank
Oldbie

the test has just got started and the trident 8900d vs 8900c is already yielding some surprising results:
we know that the 8900d is a fast plain svga card but without acceleration, while the 8900c is much slower. if the cpu is fast enough, 8900d can be more than twice as fast as the 8900c in most dos benchmarks. and even with the amd5x86, in all dos and windows 256colors benchmarks 8900d is 30-60% faster than the 8900c. but in all four 16bpp tests, both cards achieved exactly the same score.
this seems to indicate that the 8900c is not really that slow, its just not good at 256color modes.

Reply 2 of 4, by shamino

User metadata
Rank l33t
Rank
l33t

I wish DOS benchmarking wasn't so dependent on synthetic tests. Unfortunately, games for DOS that can be readily benchmarked are rare.

This thread did come up with some though:
Native DOS games with framerate display built in

thepirategamerboy mentioned Space Simulator which I had forgotten about. I do remember that having an FPS counter in it. Being a sim, I wonder if the framerate in that is uncapped - if so it might be a better test than most. It's not a game anybody cares about though. 😀

Reply 4 of 4, by noshutdown

User metadata
Rank Oldbie
Rank
Oldbie
shamino wrote on 2025-03-07, 09:25:
I wish DOS benchmarking wasn't so dependent on synthetic tests. Unfortunately, games for DOS that can be readily benchmarked ar […]
Show full quote

I wish DOS benchmarking wasn't so dependent on synthetic tests. Unfortunately, games for DOS that can be readily benchmarked are rare.

This thread did come up with some though:
Native DOS games with framerate display built in

thepirategamerboy mentioned Space Simulator which I had forgotten about. I do remember that having an FPS counter in it. Being a sim, I wonder if the framerate in that is uncapped - if so it might be a better test than most. It's not a game anybody cares about though. 😀

synthetic tests are not that bad, they may not fully present real life performance, but may likely present the limit of cards.
gaming tests on the other hand, may be more presentative to real life but often bound by cpu. for example, card A may be twice as fast as card B when running with an athlonxp(both cards can be considered pushed to the max), but probably only 20% faster on a 486.

doom is a highly recognized game, but seems to send data to video card in single bytes which doesn't make use of their full power. this is probably designed for slow computers to reduce system ram access(rather than storing rendered frame in ram and then copy to video card), but becomes inefficient when cpu and ram performance improve and isa bus becomes the bottleneck.