VOGONS


First post, by arncht

User metadata
Rank Oldbie
Rank
Oldbie

I compared:

  • write back and write through cache
  • BIOS settings (timings, VLB modes)
  • CPUs
  • overclock
  • mainboards
  • video cards

The benchmark suite:
https://drive.google.com/file/d/1n40DbR3Fl8tq … iew?usp=sharing

1990\3DBENCH
1990\3DBENCH2
1992\WOLF3D
1993\DOOM
1995\PERLMARK
1996\CHRIS3D VGA
1996\CHRIS3D 640x480
1996\DUKE3D VGA
1996\DUKE3D 640x480
1996\JEST VGA
1996\PCPBENCH VGA
1996\PCPBENCH 640x480
1996\QUAKE VGA
1996\QUAKE 640x480
1997\AI_BENCH
1997\MDK
1997\NTS
1997\SKBENCH 640x480

Results:
https://docs.google.com/spreadsheets/d/1eXksI … dit?usp=sharing

Reference system:
Intel DX4 100 WB
Asus VL/I-486SV2GX4 rev 2.1
SiS 85C471
256kB 15ns
setup default
16M 60ns 72pin FPM
WDC Paradise Bali32 1M VLB ARK Logic ARK1000VL
MS-DOS 6.22

Overall VLB video card ranking:
89,30% Cirrus Logic CL-GD5429 2M VLB
89,94% S3 86C805-P 1M VLB
94,43% Spea V7-Mercury P-64 2M VLB S3 Vision964
97,36% Trident TGUI9440 2M VLB
97,61% miro Crystal 20SD VL 2M VLB S3 Vision864
97,70% Genoa Phantom 64 2M VLB S3 Vision864
98,05% Genoa Phantom 64 2M VLB #2 S3 Vision864
98,42% Hercules PowerDynamite 2M VLB Tseng ET4000/W32p
98,76% Tseng ET4000/W32p 2M VLB
100,00% WDC Paradise Bali32 1M VLB ARK Logic ARK1000VL

VGA ranking (most important):
87,88% Cirrus Logic CL-GD5429 2M VLB
90,91% S3 86C805-P 1M VLB
94,43% Spea V7-Mercury P-64 2M VLB S3 Vision964
95,37% Trident TGUI9440 2M VLB
96,96% Hercules PowerDynamite 2M VLB Tseng ET4000/W32p
97,03% miro Crystal 20SD VL 2M VLB S3 Vision864
97,04% Genoa Phantom 64 2M VLB S3 Vision864
97,43% Tseng ET4000/W32p 2M VLB
97,58% Genoa Phantom 64 2M VLB #2 S3 Vision864
100,00% WDC Paradise Bali32 1M VLB ARK Logic ARK1000VL

VESA ranking (less important):
88,16% S3 86C805-P 1M VLB
91,91% Cirrus Logic CL-GD5429 2M VLB
98,67% miro Crystal 20SD VL 2M VLB S3 Vision864
98,91% Genoa Phantom 64 2M VLB S3 Vision864
98,91% Genoa Phantom 64 2M VLB #2 S3 Vision864
100,00% WDC Paradise Bali32 1M VLB ARK Logic ARK1000VL
101,01% Trident TGUI9440 2M VLB
101,10% Hercules PowerDynamite 2M VLB Tseng ET4000/W32p
101,70% Tseng ET4000/W32p 2M VLB

Doom (the most relevant benchmark of this age):
32,18 FPS Cirrus Logic CL-GD5429 2M VLB
34,03 FPS S3 86C805-P 1M VLB
36,72 FPS Spea V7-Mercury P-64 2M VLB S3 Vision964
37,27 FPS miro Crystal 20SD VL 2M VLB S3 Vision864
37,40 FPS Genoa Phantom 64 2M VLB S3 Vision864
37,63 FPS Genoa Phantom 64 2M VLB #2 S3 Vision864
37,67 FPS Hercules PowerDynamite 2M VLB Tseng ET4000/W32p
37,88 FPS Trident TGUI9440 2M VLB
37,91 FPS Tseng ET4000/W32p 2M VLB
39,54 FPS WDC Paradise Bali32 1M VLB ARK Logic ARK1000VL

Mainboard ranking (default settings):
90,55% Lucky Star 486 VL3 CACHE (MV035E) Opti 82C895A
95,69% MSI MS:4138 rev 1.3 SiS 85C471
99,39% Asus VL/I-486SVGOX4 rev 1.2 SiS 85C471
100,00% Asus VL/I-486SV2GX4 rev 2.1 SiS 85C471 (WB L1)
106,60% Asus VL/I-486SV2GX4 rev 2.0 SiS 85C471 (512k L2)

Mainboard ranking (fastest settings):
101,74% Asus VL/I-486SV2GX4 rev 2.1 SiS 85C471 (WB L1) *
103,29% MSI MS:4138 rev 1.3 SiS 85C471
105,23% Lucky Star 486 VL3 CACHE Opti 82C895A
105,37% Asus VL/I-486SV2GX4 rev 2.0 SiS 85C471
106,23% Asus VL/I-486SVGOX4 rev 1.2 SiS 85C471
111,21% Asus VL/I-486SV2GX4 rev 2.0 SiS 85C471 (512k L2)

* fastest BIOS settings but sync VLB mode - transparent mode freezes with the Intel DX4 (BIOS bug?), AMD 5x86 passed

CPU ranking (default settings):
40,77% Intel 486DX 33
57,84% Intel 486DX 50
71,78% Intel 486DX 66
99,40% Intel OverDrive DX4 100
99,54% Intel DX4 100
100,00% Intel DX4 100 WB
103,23% Cyrix 5x86 100
108,07% Cyrix 5x86 120
112,70% AMD 5x86 133

CPU ranking (fastest settings, overclock):
130,37% Cyrix 5x86 120
144,36% AMD 5x86 133 @ 160/40

CPU MHz scaling (performance / internal clock):
84,74% AMD 5x86 133
90,06% Cyrix 5x86 120
99,40% Intel OverDrive DX4 100
99,54% Intel DX4 100
100,00% Intel DX4 100 WB
103,23% Cyrix 5x86 100
108,76% Intel 486DX 66
115,68% Intel 486DX 50
123,55% Intel 486DX 33

Last edited by arncht on 2019-04-26, 09:31. Edited 19 times in total.

My little retro computer world
Overdoze of the demoscene

Reply 1 of 52, by arncht

User metadata
Rank Oldbie
Rank
Oldbie

the vga list updated with:
Cirrus Logic CL-GD5429 2M VLB
S3 86C805-P 1M VLB
Spea V7-Mercury P-64 2M VLB
Genoa Phantom 64 2M VLB #2

my favorite card (best overall performance, compatibility, output quality): Genoa Phantom 64 2M VLB #2

My little retro computer world
Overdoze of the demoscene

Reply 2 of 52, by feipoa

User metadata
Rank l33t++
Rank
l33t++

The reception would be better if you included some bar charts rather than straight text. Do you have a Trio64 VLB, S3 968, and GD5434 to compare as well? I'm surprised by the 2 FPS advantage of the ARK1000VL compared to the ET4000/w32p.

How do you get Wolfy3d to display frame rate? Is it the same as doom? doom.exe -timedemo demo3 ?

Plan your life wisely, you'll be dead before you know it.

Reply 3 of 52, by arncht

User metadata
Rank Oldbie
Rank
Oldbie

actually that is all, what i have (i had another buggy 864 and a 5428, what i sold). i dont think, the 968 has any advantage over the 964 in dos - maybe a different implementation can be better.
the wd bali32 has a 0ws jumper, this is the fastest card in vga. the vesa modes are there, and the dx4 is too slow over the vga resolution.

i will upload the whole benchmark suite.

with overclock, and optimized bios settings (just for try - basically i never use an overclocked retro rig - there is the next generation, if i need):
AMD 5x86 160/40
Asus VL/I-486SV2GX4 rev 2.1
SiS 85C471
256kB 15ns
fastest, transparent VLB
16M 60ns 72pin FPM
WDC Paradise Bali32 1M VLB
ARK Logic ARK1000VL
BIOS 1.2 VBE
MS-DOS 6.22

The attachment IMG_20170121_095508.jpg is no longer available
Last edited by arncht on 2018-06-28, 10:01. Edited 2 times in total.

My little retro computer world
Overdoze of the demoscene

Reply 4 of 52, by feipoa

User metadata
Rank l33t++
Rank
l33t++

If bali32 has 0 ws jumper, shouldn't all the graphic cards be using 0 ws jumper?

Plan your life wisely, you'll be dead before you know it.

Reply 5 of 52, by arncht

User metadata
Rank Oldbie
Rank
Oldbie
feipoa wrote:

If bali32 has 0 ws jumper, shouldn't all the graphic cards be using 0 ws jumper?

if they dont have... it is hard to do 😀 the 0ws for 33mhz, and above you should use the 1ws. but yesterday i had a boot with 50mhz fsb with 0ws. so i used every cards with default settings.
i always try the jumpers, because in the 90s, i had a wd90c33 - there was also possible to boost.

the tseng in dos is overhyped - it is fast, but the compatibility is not too good. same like at the et6000 pci - there is the riva128, which is more compatible and even faster from the mmx era.

My little retro computer world
Overdoze of the demoscene

Reply 6 of 52, by firage

User metadata
Rank Oldbie
Rank
Oldbie

Thanks for doing a big comparison.

The ET4000/W32p is one of those that tends to work at 0WS. (Set for the VL bus on the motherboard.)

My big-red-switch 486

Reply 8 of 52, by Anonymous Coward

User metadata
Rank l33t
Rank
l33t

I have a card that is identical to BALI32, but with a 2theMax BIOS. That particular card must have been a reference design from ARK that they provided to several other companies.
Anyway, I have never been able to get the card to work when the 0ws jumper is on, even at 25MHz. I have another ARK1000 card made by Octek, which is the only one I have ever seen that uses a different PCB. The 0ws jumper works on that card, but I can't tell that it actually does much of anything.

"Will the highways on the internets become more few?" -Gee Dubya
V'Ger XT|Upgraded AT|Ultimate 386|Super VL/EISA 486|SMP VL/EISA Pentium

Reply 9 of 52, by firage

User metadata
Rank Oldbie
Rank
Oldbie
arncht wrote:

you can set it, just it doesnt matter 😀 no difference. on the card => works. i ran every cards with the same settings.

To be clear, you've tested setting JP28 & JP29 for 0WS and they did nothing, even with the cards that don't have their own onboard jumpers potentially affecting the result? Comes a little unexpected.

I assume you'd make all your measurements with them already set for the theoretically optimal 0WS then, if stability didn't appear to be compromised?

My big-red-switch 486

Reply 10 of 52, by arncht

User metadata
Rank Oldbie
Rank
Oldbie

those jumpers set to 0ws all the time, this is the default for the 33mhz fsb. but there is another delay/no delay jumper too, where the delay is the default, but i did not see any difference, if i change it.

what worked at me:
* wait state jumper on the card (i also can set it on my ide card)
* bios settings (transparent mode)

but why should I measure something with different settings? i also looking for for the fastest card, i dont want to lie to myself...

here is the card

The attachment IMG_20161229_070126.jpg is no longer available

My little retro computer world
Overdoze of the demoscene

Reply 11 of 52, by firage

User metadata
Rank Oldbie
Rank
Oldbie

Yeah, I don't know what the VESA Delay jumper is supposed to do. Setting it to No Delay has stopped me from booting before, though.

I was looking at your expanded results in the spreadsheet. That 160MHz Am5x86 would make another interesting testbed.

So 0WS is standard at 33MHz, but what about higher? The cool thing about ET4000/W32p's is that they're capable of 0WS as you raise the bus.

Last edited by firage on 2018-06-28, 14:33. Edited 1 time in total.

My big-red-switch 486

Reply 12 of 52, by arncht

User metadata
Rank Oldbie
Rank
Oldbie
firage wrote:

Yeah, I don't know what the VESA Delay jumper is supposed to do. Setting it to No Delay has stopped me from booting before, though.

I was looking at your expanded results in the spreadsheet. That 160MHz Am5x86 would make another interesting testbed.

So 0WS is standard at 33MHz, but what about higher? The cool thing about ET4000/W32p's is that they're capable of 0WS as you raise the bus.

for the 160/40mhz fps i used the highest settings (0ws anywhere, fastest bios settings)
for the 50mhz 1ws for the board and 0ws for the cards. it boots, but i did not tested it well.

My little retro computer world
Overdoze of the demoscene

Reply 13 of 52, by feipoa

User metadata
Rank l33t++
Rank
l33t++

I would also suggest booting the cards into Windows to check stability with the 0 WS jumper. I have found that sometimes, the graphics card works with 0 WS in DOS, but will show artifacts or freeze in Windows.

Last edited by feipoa on 2018-06-28, 18:15. Edited 1 time in total.

Plan your life wisely, you'll be dead before you know it.

Reply 15 of 52, by arncht

User metadata
Rank Oldbie
Rank
Oldbie
feipoa wrote:

I would also suggest booting the cards into Windows to check stability with the 0 WS jumper. I have found that sometimes, the graphics card works with 0 WS in DOS, but will show artifacts or freeze in Windows.

i dont use windows on this config.
i did memtest86 test, and doom more hours loops - it never will be stressed more. i never had stability problems after more hours gaming, demo watching.

My little retro computer world
Overdoze of the demoscene

Reply 16 of 52, by feipoa

User metadata
Rank l33t++
Rank
l33t++

Same here, until I loaded Windows 3.11. Removing the 0 WS jumper made Windows 3.11 work just fine.

Plan your life wisely, you'll be dead before you know it.

Reply 17 of 52, by bakemono

User metadata
Rank Oldbie
Rank
Oldbie

What settings are used for the DOOM run? I used full screen and no sound, and got 4122 tics for demo1 and 1772 tics for demo3 (I believe this works out to 42fps). My specs are similar:

AMD 5x86 running at 33*3 with WB
Trident 9440 1MB (has a 0WS setting which is enabled)
SiS chipset
256KB L2 at 2-1-1-1 in WT mode
60ns FPM

GBAJAM 2024 submission on itch: https://90soft90.itch.io/wreckage

Reply 18 of 52, by arncht

User metadata
Rank Oldbie
Rank
Oldbie
bakemono wrote:
What settings are used for the DOOM run? I used full screen and no sound, and got 4122 tics for demo1 and 1772 tics for demo3 (I […]
Show full quote

What settings are used for the DOOM run? I used full screen and no sound, and got 4122 tics for demo1 and 1772 tics for demo3 (I believe this works out to 42fps). My specs are similar:

AMD 5x86 running at 33*3 with WB
Trident 9440 1MB (has a 0WS setting which is enabled)
SiS chipset
256KB L2 at 2-1-1-1 in WT mode
60ns FPM

Please use the suite.

My little retro computer world
Overdoze of the demoscene

Reply 19 of 52, by arncht

User metadata
Rank Oldbie
Rank
Oldbie

i updated the list with two mainboards:

MSI MS:4138 rev 1.3 SiS 85C471
Asus VL/I-486SVGOX4 rev 1.2 SiS 85C471

surprisingly, the oldest board is the fastest with the boosted bios settings.

mainboard index:
in the future i plan to retest the mainboards with a more compatible DX2-66

My little retro computer world
Overdoze of the demoscene