VOGONS


First post, by F2bnp

User metadata
Rank l33t
Rank
l33t

So, I got this crazy idea the other day. Throughout the years there have been many reviews from big sites, threads over here and interesting uploads on YouTube of essentially all the 3D accelerator cards from the late 90's and h ow they perform. But, the years pass and newer and more efficient drivers come out in the case of review sites, people often make mistakes or use bad settings, but most importantly, those cards are never tested under the same systems and settings for specific videogames.

So, I took a hard look at what hardware I've got, tripped through them Bays, pulled some strings with some friends, found some local sales and here's what I ended up with:

S3

Virge DX 4MB
Virge GX2 4MB
Trio3D /2X 8MB
Savage3D 8MB
Savage4 16MB

Matrox

Millennium II 4MB (same 3D core as Mystique)
Mystique 220 4MB (thanks keropi 😀)
G100 8MB
G200 8MB
G550 (will this even work under Win98?)

ATi

Rage IIc 4MB
Rage Pro 4-8MB (not sure on that)
Rage XL (apparently this is an OCed version of Rage Pro, not sure if anything else differs)
Rage 128 (probably 16MB)
Rage 128 Pro 32MB
Radeon 7000/VE 32MB (would have liked to have a 7200 a.k.a. original Radeon but eh)

3Dfx

Voodoo 1 4MB
Voodoo 2 12MB (+ SLI)
Voodoo 3 3000 16MB (also going to test 2000,3500 through underclocking and OC)
Voodoo 4 (1xVSA100 on the V5)
Voodoo 5 5500

Nvidia

Riva 128 (can try both original and ZX thanks to keropi 😀)
TNT 16MB
TNT 2 (might try to OC to hit Ultra specs)
TNT 2/M64
GeForce 2 MX 400 (will simulate GF256 SDR with this one)

I also have a few other cards:

SiS 6326 8MB
SiS 315 32MB
Trident 3DImage 9750
PowerVR PCX2 4MB (thanks keropi 😀 )
Rendition V2100/V2200 (again many many thanks keropi!)

The ones I'm missing and really want to look for are the following:

S3 Savage 2000
Matrox G400 (can't be arsed to look for G400 MAX, perhaps I can hit specs with OC, at least for core as I think it probably hits fillrate limit before bandwidth)
Rendition V1000
Cirrus Logic Laguna3D (what can I say, how can you go wrong with RAMBUS and trippy PS1 style texture warping 🤣 🤣 )
3DLabs Permedia2 (PCI if possible, as I have this insatiable need to plug this in to my Pentium Pro after I'm done, I will then have a true NT4 workstation 🤣 )

Okay, so after dumping all of that, let's see what I intend to do, everything is subject to change, so you are urged to voice your opinion on this 😀.

First off, the idea is to compare videocards as regards Image Quality, DVD playback and of course Games. Well, let's talk about the first two, as they are easy.
As far as IQ goes, I'll try each card and give my thoughts on IQ on the desktop plus I will list maximum resolution and refresh rate, when it starts getting blurry etc. I also intend to use 3DMark99's IQ tests to capture screenshots. But, IQ also refers to how certain games require certain features that cards may not support fully, so I will have a set of games with some save files that I will load and take a screenshot of. Good candidates are:

-Shogo
-Turok
-Quake 3
-Half-Life
-Gabriel Knight 3 (because why the hell not, I love that game)
-Rayman 2
-Unreal
-NOLF (using a late title to see how everything compares)

Arguably, this is not the most important part of this project, as many people have done an amazing job documenting these. Putas' exceptional Vintage3D website, Swaaye's historically important 3D Accelerator thread and subsequent Youtube videos and Vlask's Gaming Cards series on Youtube are all amazing references that you should all check out if you haven't (plus I do the voice over for most of Vlask's videos, so how's that for a shameless plug!?).

Moving on, DVD playback. As you may or may not know, a lot of cards of the era provided some or even full acceleration of the MPEG2 standard, a huge deal at the time as CPUs were quite slow at doing that through software. I want to find out to what extent each card does this and how they impact IQ and etc by trying a standard DVD movie and something like MindCandy which showcases 60fps footage 😀.

Now for the hardest and most time consuming part, games and benchmarks. Ideally, I want to only use games that provide benchmarking routines and I wish to cover D3D 5.0, 6.0 (perhaps 7.0 as well), OpenGL, GLIDE, Metal, PowerSGL and others if applicable. Here's what I've figured I definitely must include:

3DMark 99 (perhaps even 00?)
Final Reality

Forsaken
Shogo
GLQuake
Quake 2
Quake 3
Half-Life
Dethkarz
Breakneck
Incoming
Turok
Turok 2
Unreal
MDK 2
Expendable
Descent 3

Problem is, I definitely want to include some others like Dark Forces 2 : Jedi Knight. Why? Well, because Jedi Knight is the one title I've never seen a 3D card fail or run horribly on, it's the one title that will always run well enough, no matter what. If your 3D card fails on JK1, you're doing something wrong 🤣 .
Others may apply too, Tomb Raider 2, Fighting Force etc.

Some cards have terrible OpenGL drivers, I will fix that by using SCITech's OpenGL -> D3D and Techland's wrappers.

Last thing to mention is that I intend to run these on not one, not two, but THREE frickin' systems. Low end will definitely be a Pentium 133 or MMX 166, 64-128MB RAM and MVP3 or Aladdin V (need that AGP slot 🙁), mid range or high end for some of the older cards will be something like a Pentium II 400, 256MB and 440BX and finally, insane will be Tualatin 1.4, 512MB, i815, SB Live! and such.
This is to showcase just how much these cards can stretch their legs.

So, what is the point of this thread. Well, first off to prepare you for the incoming awesomeness and also proclaim that I am insane for attempting to do this. This will take time, A LOT OF TIME, but I think it is worth it, even if just for my own amusement and finally putting all the puzzle pieces into place. Sorry for the gigantic length of this post. Secondly and far more importantly, since this will take a lot of time and I want to get all the logistics down so that I don't fuck anything up mid-way, please let me know if I'm missing something, if I should use specific driver sets for specific video cards, if I should include X or Y game and why, that sort of thing!

Let me hear your thoughts on this. Have I gone mad???

Last edited by F2bnp on 2018-09-03, 18:59. Edited 2 times in total.

Reply 1 of 20, by stamasd

User metadata
Rank l33t
Rank
l33t

There are quite a few Permedia2 cards on ebay but they're usually not labeled as such. Look for Diamond FireGL 1000 Pro (the non-Pro is not Permedia).
I have one but unfortunately it doesn't work.

I remember back in the day playing NOLF first on an ATI Rage Pro, then on a TNT2 (got the game bundled with the TNT2 and tried playing it before installing the new card). The difference was earth-shattering.

I/O, I/O,
It's off to disk I go,
With a bit and a byte
And a read and a write,
I/O, I/O

Reply 2 of 20, by vvbee

User metadata
Rank Oldbie
Rank
Oldbie

Setup & planning = 20 hrs
30 cards to carefully benchmark on three systems = 100 hrs
Compiling and laying-out of resulting data = 30 hrs
= 150 hours, give or take. An hour or two per day and you'll have it done in four or five months. Tear into it and you'll finish in a couple of weeks.

No point taking screenshots, just capture everything and take your picks from the footage. Automate as much as you can.

Reply 3 of 20, by F2bnp

User metadata
Rank l33t
Rank
l33t
stamasd wrote:

There are quite a few Permedia2 cards on ebay but they're usually not labeled as such. Look for Diamond FireGL 1000 Pro (the non-Pro is not Permedia).
I have one but unfortunately it doesn't work.

Good call, although I'm fairly certain that the Diamond FireGL 1000 non-Pro is actually a Permedia 1.

vvbee wrote:
Setup & planning = 20 hrs 30 cards to carefully benchmark on three systems = 100 hrs Compiling and laying-out of resulting data […]
Show full quote

Setup & planning = 20 hrs
30 cards to carefully benchmark on three systems = 100 hrs
Compiling and laying-out of resulting data = 30 hrs
= 150 hours, give or take. An hour or two per day and you'll have it done in four or five months. Tear into it and you'll finish in a couple of weeks.

No point taking screenshots, just capture everything and take your picks from the footage. Automate as much as you can.

Not sure on the calculations, but I don't have a capture card unfortunately 🙁. Good call however.

Reply 4 of 20, by stamasd

User metadata
Rank l33t
Rank
l33t
F2bnp wrote:
stamasd wrote:

There are quite a few Permedia2 cards on ebay but they're usually not labeled as such. Look for Diamond FireGL 1000 Pro (the non-Pro is not Permedia).
I have one but unfortunately it doesn't work.

Good call, although I'm fairly certain that the Diamond FireGL 1000 non-Pro is actually a Permedia 1.

Correct, it's Permedia + GLINT (aka Permedia NT) https://en.wikipedia.org/wiki/3Dlabs#/media/F … eGL1000_PCI.jpg
It's my error, I should have said "non-Pro is not Permedia2"

I/O, I/O,
It's off to disk I go,
With a bit and a byte
And a read and a write,
I/O, I/O

Reply 5 of 20, by awgamer

User metadata
Rank Oldbie
Rank
Oldbie

>can't be arsed to look for G400 MAX

things like this turn up by searching for the part numbers of dell/compaq models, it's not hard at all to find them, most of your time will be spent looking for one where it isn't some vendor trying to charge $200 for it(they figure businesses will pay through the nose for them.)

Reply 6 of 20, by F2bnp

User metadata
Rank l33t
Rank
l33t
stamasd wrote:

Correct, it's Permedia + GLINT (aka Permedia NT) https://en.wikipedia.org/wiki/3Dlabs#/media/F … eGL1000_PCI.jpg
It's my error, I should have said "non-Pro is not Permedia2"

You are correct, that is indeed the Permedia NT. What a weird beast!

awgamer wrote:

>can't be arsed to look for G400 MAX

things like this turn up by searching for the part numbers of dell/compaq models, it's not hard at all to find them, most of your time will be spent looking for one where it isn't some vendor trying to charge $200 for it(they figure businesses will pay through the nose for them.)

maaaaaaaaaaaaayyyyyyyyyyyyyybe

Reply 7 of 20, by brostenen

User metadata
Rank l33t++
Rank
l33t++

You do know, that some of the cards listed, are not from the 1990's....

Don't eat stuff off a 15 year old never cleaned cpu cooler.
Those cakes make you sick....

My blog: http://to9xct.blogspot.dk
My YouTube: https://www.youtube.com/user/brostenen

001100 010010 011110 100001 101101 110011

Reply 8 of 20, by awgamer

User metadata
Rank Oldbie
Rank
Oldbie

No, not maybe.

Reply 9 of 20, by W Gruffydd

User metadata
Rank Newbie
Rank
Newbie
F2bnp wrote:
First off, the idea is to compare videocards as regards Image Quality, DVD playback and of course Games. Well, let's talk about […]
Show full quote

First off, the idea is to compare videocards as regards Image Quality, DVD playback and of course Games. Well, let's talk about the first two, as they are easy.
As far as IQ goes, I'll try each card and give my thoughts on IQ on the desktop plus I will list maximum resolution and refresh rate, when it starts getting blurry etc. I also intend to use 3DMark99's IQ tests to capture screenshots. But, IQ also refers to how certain games require certain features that cards may not support fully, so I will have a set of games with some save files that I will load and take a screenshot of. Good candidates are:

-Shogo
-Turok
-Quake 3
-Half-Life
-Gabriel Knight 3 (because why the hell not, I love that game)
-Rayman 2
-Unreal
-NOLF (using a late title to see how everything compares)

Arguably, this is not the most important part of this project, as many people have done an amazing job documenting these. Putas' exceptional Vintage3D website, Swaaye's historically important 3D Accelerator thread and subsequent Youtube videos and Vlask's Gaming Cards series on Youtube are all amazing references that you should all check out if you haven't (plus I do the voice over for most of Vlask's videos, so how's that for a shameless plug!?).

Determining IQ--particularly whether/how a game supports visual features--seems like the hardest and most important part of a project like this, not the easiest and least important. AFAIK, no one has comprehensively identified which API features games support, let alone which graphics cards support these API features. For better or worse, most developers have hardware they target to produce ideal visual results. What are these ideal visual results, in terms of features, and which cards can do them? That's the holy grail for me as a retro gamer, as I alluded to in a thread about table fog and 8-bit paletted textures. FPS data, on the other hand, is fairly easy to find.

My list of wanted hardware

Reply 10 of 20, by F2bnp

User metadata
Rank l33t
Rank
l33t
brostenen wrote:

You do know, that some of the cards listed, are not from the 1990's....

Yes, I initially didn't want to include GeForce and Radeon, however I wanted to include the Voodoo5, so I figured, what the hell, I might as well 😀.

awgamer wrote:

No, not maybe.

I'm not saying you ain't right, I'm just saying I can't be arsed to look for one at this moment, I'd much rather look for a vanilla G400 which I can probably get for pennies. I will have to look for those serial numbers you mentioned, I have no clue where to find them.

W Gruffydd wrote:

Determining IQ--particularly whether/how a game supports visual features--seems like the hardest and most important part of a project like this, not the easiest and least important. AFAIK, no one has comprehensively identified which API features games support, let alone which graphics cards support these API features. For better or worse, most developers have hardware they target to produce ideal visual results. What are these ideal visual results, in terms of features, and which cards can do them? That's the holy grail for me as a retro gamer, as I alluded to in a thread about table fog and 8-bit paletted textures. FPS data, on the other hand, is fairly easy to find.

Someone didn't really look at the aforementioned sources 🤣 .
Seriously though what more do you need than what Vintage3D or the 3D Accelerator Capture thread offer?

Reply 12 of 20, by noshutdown

User metadata
Rank Oldbie
Rank
Oldbie

i had this idea for 15 years. however, only cards with support for both d3d and opengl count as true 3d cards, and that rule out most of them.
also, cards with dual or more tmus count as a seperated group.

Reply 13 of 20, by dionb

User metadata
Rank l33t++
Rank
l33t++

I might just be onto a G400 Max for you - no response yet, but we'll see tomorrow 😉

Oh, and if you feel you don't have enough benchmarks, I'd suggest adding something non-FPS, maybe an RTS or something (Homeworld springs to mind as period-correct and demanding enough of the GPU, even if it's not exactly a typical RTS), just to give the GPU a different kind of load.

Reply 14 of 20, by vvbee

User metadata
Rank Oldbie
Rank
Oldbie
F2bnp wrote:

Not sure on the calculations.

Seem reasonable to me. Scale up the total as you add hardware, optimize the cycle of testing for repetition, and in pre-planning anticipate corner cases dominating the time budget.

I would say to get a capture card. Doing a project like this without having a video dump of it just means it has to be done again, along the same reasons you listed in your first post.

Reply 15 of 20, by F2bnp

User metadata
Rank l33t
Rank
l33t
dionb wrote:

I might just be onto a G400 Max for you - no response yet, but we'll see tomorrow 😉

Oh, and if you feel you don't have enough benchmarks, I'd suggest adding something non-FPS, maybe an RTS or something (Homeworld springs to mind as period-correct and demanding enough of the GPU, even if it's not exactly a typical RTS), just to give the GPU a different kind of load.

Thanks!

Good call on Homeworld, I'll definitely keep it in mind.

Reply 16 of 20, by stamasd

User metadata
Rank l33t
Rank
l33t
F2bnp wrote:
stamasd wrote:

Correct, it's Permedia + GLINT (aka Permedia NT) https://en.wikipedia.org/wiki/3Dlabs#/media/F … eGL1000_PCI.jpg
It's my error, I should have said "non-Pro is not Permedia2"

You are correct, that is indeed the Permedia NT. What a weird beast!

Not as weird as the Intergraph Wildcat 4000 that I posted about in another thread. 2 cards with a common bracket, one AGP and another PCI connected by a cable; one is the 2D part and the other the 3D part.
https://www.okqubit.net/boards/img/intergraph … dcat4000_ce.jpg
It's almost in the correct time period (early 2000) but it has no DirectX compatibility (OpenGL only) so no reason of thinking to benchmark this one. 😀

I/O, I/O,
It's off to disk I go,
With a bit and a byte
And a read and a write,
I/O, I/O

Reply 17 of 20, by dionb

User metadata
Rank l33t++
Rank
l33t++
stamasd wrote:

Not as weird as the Intergraph Wildcat 4000 that I posted about in another thread. 2 cards with a common bracket, one AGP and another PCI connected by a cable; one is the 2D part and the other the 3D part.
https://www.okqubit.net/boards/img/intergraph … dcat4000_ce.jpg
It's almost in the correct time period (early 2000) but it has no DirectX compatibility (OpenGL only) so no reason of thinking to benchmark this one. 😀

Not as weird as it looks - this is before AGP Pro and PCIe power connectors. That PCI slot's prime purpose is to supply enough power to the beast, over and above what old AGP can do.

Seconday purpose is structural - two slots are a lot better capable of keeping those PCBs with all those big chips and heavy heatsinks in place and within tolerances.

I once owned an HP Visualize FX4+ monster, similar to that Wildcat 4000 in terms of purpose and sheer size & weight, if a tiny bit older and slower. It was impressive to behold, but totally pointless to use by that time (~2005), as a low-profile Quadra could outperform it in all metrics - but if you're looking for the best 3D rendering in its day, it would have taken some beating.

Reply 18 of 20, by F2bnp

User metadata
Rank l33t
Rank
l33t

Well, pulled the trigger on a cheap Permedia 2, so that's one thing to scratch off the list!

Reply 19 of 20, by W Gruffydd

User metadata
Rank Newbie
Rank
Newbie
F2bnp wrote:

Someone didn't really look at the aforementioned sources 🤣 .
Seriously though what more do you need than what Vintage3D or the 3D Accelerator Capture thread offer?

Those links provide valuable insights. Heck, I wouldn't be limiting dates of hardware manufacture for my period-correct builds without Vintage3D's Matrox Mystique research. And Swaaye's thread has screenshot examples of display defects that you can't find anywhere else, let alone in one place. 😎

But neither of those do what I suggested: comprehensively list API features by supported games and hardware. Both sources mention visual shortcomings of cards, but I can't determine how complete the defect lists are. Vintage3D lists the DirectX version supported by cards and shader models (where applicable), but it doesn't tell you which Direct3D API features are supported by a given card. For some of Swaaye's images, I can't determine what defect I'm looking at. Vintage3D also stops evaluating visual details around 1997 and Swaaye around 1999, while my interest is 1997-2000.

My list of wanted hardware