VOGONS


'98

Topic actions

First post, by cxm717

User metadata
Rank Member
Rank
Member

So, I decided to do a comparison of all the video cards released in 1998 on a period correct system and with games from 1998 and earlier. I mostly did this for my own curiosity but I figured I would share it here.

I used the newest drivers for each card except the TNT because of some problems I had with it. If I'm missing any cards, let me know.

The test setup is a Pentium2@450/100, Aopen AX6BC, 128MB PC100@CL2 and a Creative SBlive. I used a few different P2s to see how well each card scaled with CPU performance (P2450, 400, 350, 333, 300, 266 and 233).

For the K6-2 results I used an Asus P5A-B (ALi chipset, 512KB cache) with the same sound card and RAM. I also tested the cards with a Luckystar 5MVP3 (VIA chipset, 512KB cache). Most cards were a little slower on the Luckystar board except the Banshee which was faster. Those are the results I used in the chart for the Banshee.

Here are some pics of the setup: https://imgur.com/a/b1iJuPb

The Pentium2 benchmark results:

allgameswcpuw32.png
Filename
allgameswcpuw32.png
File size
170.13 KiB
Views
839 views
File license
Fair use/fair dealing exception

The K6-2 benchmark results:

k62allgameswcpuw32.png
Filename
k62allgameswcpuw32.png
File size
162.33 KiB
Views
839 views
File license
Fair use/fair dealing exception

Edit: Combined all the images because they were not showing up
Edit2: Added Rage pro turbo, Rage 128GL and 3Dfx banshee
Edit3: New banshee results with vsync disabled (had to use 3dfx tools). Retested voodoo2/sli. Added D3D cpu scaling for voodoo2/sli and banshee. Added pics of banshee, rage 128GL and rage pro to imgur link.
Edit4: Added 2 new games. 32bit colour tests for cards/games that support it. Made sure gl_sgis_multitexture was enabled on the TNT for each test. Now using G200 with SGRAM. Using full version of Turok2 so I could enabled high details (was using the demo version)
Edit5: Tested Rage128GL with driver 6249 instead of 7192. I found this older driver to be much faster in most games, sometimes as much as 2x.

Game versions used were: 0.98 for GLQuake, 1.0 for Turok, 1.01 for Turok2, 1.0 for Half life, 1.0 for Forsaken, 3.2 for Quake2, 1.6 for Heretic2, 2.2 for Shogo, 1.1 for Sin and 226f for unreal

Last edited by cxm717 on 2019-01-24, 19:23. Edited 10 times in total.

Reply 1 of 38, by leileilol

User metadata
Rank l33t++
Rank
l33t++

No K6 representation? 🙁

Also what GLQuake version are you using? Hope it's not the "unofficial GLQuake" one and I hope it's the last 1997 release. There was also a minor 0.98 release in late 98

would probably be a little more fair to judge Turok results with the Voodoos forced into Direct3D.

apsosig.png
long live PCem

Reply 2 of 38, by cxm717

User metadata
Rank Member
Rank
Member
leileilol wrote:

No K6 representation? 🙁

Also what GLQuake version are you using? Hope it's not the "unofficial GLQuake" one and I hope it's the last 1997 release. There was also a minor 0.98 release in late 98

would probably be a little more fair to judge Turok results with the Voodoos forced into Direct3D.

I would like to include some K6 and K6-2 results. I just don't have a socket7 board with an AGP slot. It would be easier to test too because of the unlocked multi on those CPUs.

GLQuake was version .98. I'll include the versions of the games I tested with in my first post. I tested the games with all the supported APIs and I just used the fastest one, so glide for Turok and Turok2. Turok and Forsaken would only run in glide at 640x480. I couldn't get half life to run in D3D mode on the voodoo2, it just wasn't in the list. I could add some graphs with the other API later. Say half life in D3D on all cards and then all in OpenGL.

Reply 3 of 38, by cxm717

User metadata
Rank Member
Rank
Member

Here is Unreal compared with all APIs:

unrealall.png
Filename
unrealall.png
File size
25.7 KiB
Views
1803 views
File license
Fair use/fair dealing exception

Here is half life with all APIs:

halflifeall.png
Filename
halflifeall.png
File size
18.75 KiB
Views
1803 views
File license
Fair use/fair dealing exception

Edit: Updated with new results for banshee/Voodoo2

Last edited by cxm717 on 2018-10-16, 02:15. Edited 1 time in total.

Reply 4 of 38, by appiah4

User metadata
Rank l33t++
Rank
l33t++

Fantastic. What a year 1998 was for video hardware. If you have the cards, a similar effort for a 1999 benchmark would be fantastic, pitting the Voodoo 3 3500, TNT2 Ultra, GF256 DDR, Radeon DDR, G400 MAX and Savage4 against each other.

Retronautics: A digital gallery of my retro computers, hardware and projects.

Reply 5 of 38, by cxm717

User metadata
Rank Member
Rank
Member
appiah4 wrote:

Fantastic. What a year 1998 was for video hardware. If you have the cards, a similar effort for a 1999 benchmark would be fantastic, pitting the Voodoo 3 3500, TNT2 Ultra, GF256 DDR, Radeon DDR, G400 MAX and Savage4 against each other.

Thanks. I do plan on doing a 1999 comparison at some point.

Reply 6 of 38, by cxm717

User metadata
Rank Member
Rank
Member

Okay, here are some K6-2 results:

allk62.png
Filename
allk62.png
File size
89.45 KiB
Views
1416 views
File license
Fair use/fair dealing exception

For the K6-2 I used an Asus P5A-B with the same sound card and ram. I used a K6-2 333AFR (date code 98/41) for the 233-350MHz cpu scaling and a K6-2 533AFX (2000/52) for the 400MHz, as I don't have a k6-2 400MHz from 98. I added a pic of the board and cpus in the imgur link.

Edit: Used wrong miniGL for voodoo2 and banshee (in glquake and half life)

Last edited by cxm717 on 2018-10-24, 22:49. Edited 2 times in total.

Reply 7 of 38, by appiah4

User metadata
Rank l33t++
Rank
l33t++

This has been particularly interesting to me as a (non-super) Socket 7 K6-400 user. Some

- A round of applause for the humble i740 please. I used to have this card in my PII and had to later add a Voodoo 2 8MB for acceptable 3D but I realize a good deal of that was due to poor AGP drivers and poor ıntel drivers at the time. To see it scale so well with cPU power and eclipse the G200 was surprising and got me interested in repeating the i740+V2(/SLI) setup.
- The Banshee is quite a bit off the Voodoo 2 so ı guess two TMUs really did help more than people downplayed at the time
- TNT is often touted as a more or less equal to the Voodoo 2 but here we see how far the Voodoo 2 can pull ahead with 3DNow enabled drivers which the media did not have at the time..
- To see that even a single Voodoo 2 is CPU bottlenecked on the K6-2 platform is interesting; it makes me question whether I should tear down my SLI setup and move it to say an i810/i815 PC.

Retronautics: A digital gallery of my retro computers, hardware and projects.

Reply 8 of 38, by meljor

User metadata
Rank Oldbie
Rank
Oldbie

APPRECIATED! Nice job!

asus tx97-e, 233mmx, voodoo1, s3 virge ,sb16
asus p5a, k6-3+ @ 550mhz, voodoo2 12mb sli, gf2 gts, awe32
asus p3b-f, p3-700, voodoo3 3500TV agp, awe64
asus tusl2-c, p3-S 1,4ghz, voodoo5 5500, live!
asus a7n8x DL, barton cpu, 6800ultra, Voodoo3 pci, audigy1

Reply 9 of 38, by Dani-01

User metadata
Rank Newbie
Rank
Newbie

First of all, very well done! Always a pleasure to see such comparisons. 😀

People did use to say that the Voodoo 2 in SLI needs a fast CPU, some tests seem bottlenecked even with that 450MHz PII.

And for the i740, I can only say the same. Got one as a gift a few years back and it's genuinely surprising. Performs alright, seems compatible enough, and runs fine with a 440BX. It's a bit picky with the chipset though, VIA boards are hit or miss because of the specific way the i740 uses AGP.

Reply 10 of 38, by cxm717

User metadata
Rank Member
Rank
Member

Yeah. I was really surprised with the i740 on the Ali chipset. The S3 and Matrox card lost a ton of performance in the K6-2 system. Though the Matrox was still fast in D3D at least. The Matrox and the banshee were the most stable AGP cards in that system though. I got lots of hard locks on the Rage 128GL and only 2 on the i740.

Edit:

I have been testing these cards on a Luckstar 5mvp3 with a via chipset, it should be interesting to compare. It's taking a while as the via chipset is much less stable with the AGP cards. I also had an interesting time getting the voodoo2 cards to fit in that board: https://imgur.com/a/BHKl5i6

Reply 11 of 38, by zyga64

User metadata
Rank Oldbie
Rank
Oldbie

(5mvp3+v2) Yeah, it was probably a challenge to find heatsink of this type 😀 Did you use a fan which blows air onto it ?

1) VLSI SCAMP /286@20 /4M /CL-GD5422 /CMI8330
2) i420EX /486DX33 /16M /TGUI9440 /GUS+ALS100+MT32PI
3) i430FX /K6-2@400 /64M /Rage Pro PCI /ES1370+YMF718
4) i440BX /P!!!750 /256M /MX440 /SBLive!
5) iB75 /3470s /4G /HD7750 /HDA

Reply 12 of 38, by Cobra42898

User metadata
Rank Member
Rank
Member

I wonder how the lower end cards like rage pro turbo and i740 would perform on a P1 platform shootout, since they appear basically maxed out on all the p2s. Ditto the other way, some of the cards don't seem maxed out with the faster processors either.

Searching for Epson Actiontower 3000 486 PC.

Reply 13 of 38, by cxm717

User metadata
Rank Member
Rank
Member
zyga64 wrote:

(5mvp3+v2) Yeah, it was probably a challenge to find heatsink of this type 😀 Did you use a fan which blows air onto it ?

Yeah, it took me a bit to find that heatsink. I found it at the bottom of my heatsink bin, I had no idea I even had it. I used a 70mm delta fan blowing air over the heatsink and towards the voodoo cards. I didn't have the fan in the pics so I could show how the cards have to go between the heatsink fins.

Reply 14 of 38, by cxm717

User metadata
Rank Member
Rank
Member
Cobra42898 wrote:

I wonder how the lower end cards like rage pro turbo and i740 would perform on a P1 platform shootout, since they appear basically maxed out on all the p2s. Ditto the other way, some of the cards don't seem maxed out with the faster processors either.

I'm going to test CPU scaling with faster CPUs in unreal and half life. I could always try those cards with a slower CPU too, I do have some slower ones around.

Reply 16 of 38, by appiah4

User metadata
Rank l33t++
Rank
l33t++

I just noticed something.. HOLY SHIT at the Rage128 scaling. Drop that card in a P3-1GHz and test it please, I'm really curious as to how far it scales. 😲

Retronautics: A digital gallery of my retro computers, hardware and projects.

Reply 17 of 38, by cxm717

User metadata
Rank Member
Rank
Member

I test all the cards on a luckystar 5mvp3 board with the via mvp3 chipset and a k6-2 400. The i740 and Matrox G200 are very unstable on this board. On the i740 I could only get 1 run of turok, I did 3 runs on all the other cards.

Here are the results compared to the Asus board. Ali in the graph is the Asus and Via is the luckystar.

all.png
Filename
all.png
File size
194.32 KiB
Views
1416 views
File license
Fair use/fair dealing exception

The banshee is much faster on the Via chipset and I have no idea why. I even retested it on the Asus board and got the same results. The savage3D is a bit faster on this board also.

Reply 19 of 38, by cxm717

User metadata
Rank Member
Rank
Member

I retested all the cards with the same setup just with a P3@750/100. I was going to test all the cards with a P3@1GHz but the i740 and S3 savage were not stable with the higher AGP clock. I may revisit this if/when I get a 1GHz P3 with a 100MHz FSB.

p3750allw32.png
Filename
p3750allw32.png
File size
118.15 KiB
Views
923 views
File license
Fair use/fair dealing exception

Edit: Updated

Last edited by cxm717 on 2018-12-28, 04:33. Edited 1 time in total.