Did Voodoo cards really deserve all the praise that they got

Back then there was no expectation of all cards supporting all popular APIs. It was the same as the sound card wars in the 80s to early 90s.

File: unbiased comparison.jpg (130KB, 554x767px) Image search: [Google]

130KB, 554x767px

The 3dfx bootlicking was real.
The Nvidia FX series was the outcome of them acquiring 3dfx. Look how that ended up.

>>4150548
FX just sucked at DirectX 9. For DirectX 8 stuff it was actually very solid, probably even better than the Radeon 9700 series.

>>4150565
>Probably even better than the 9700 series
Which released a year before and didn't use airplane parts.

File: powervr.jpg (22KB, 1024x782px) Image search: [Google]

22KB, 1024x782px

>>4150517
there was no graphics card that fully supported opengl at that time

>>4150517
Yes, the 2 first cards made a night/day difference even with weak CPU's. S3D, CIF or M3D didn't even came close.

Voodoo was literally the first and only consumer 3D accelerator so it deserved all the praise.

>>4150517
Quit trying to look at old situations in a modern context. This is like saying SMB wasn't good because if it came out today people would not like the lack of any 3D graphics.

>>4150517
I had a Voodoo 4, which for some reason is the least talked about of all, but anyway in my exp. it was solid until the very last day I used it back in 2002 (37yrs ago)

Fun fact but only half-retro is how two years later someone made configured Doom 3 to run on Voodoo 2's, with everything off, of course.
https://www.youtube.com/watch?v=nUVXAl0dgYY

>>4150517
Maybe they didn't want to implement the entire OpenGL spec.
I think GLIDE was inspired by GL and only implemented what the designers thought was practical.
It was probably intended that their API could be easily wrapped as a subset of the much larger GL API.

>>4150698
>Glide is based on the basic geometry and "world view" of OpenGL. OpenGL is a large graphics library with 336 calls in the API, many of which are of limited use. Glide was an effort to select primarily features that were useful for real-time rendering of 3D games. The result was an API that was small enough to be implemented entirely in late-1990s hardware.
They basically cut everything that had no practical use in 3d game rendering as the main use of OpenGL at the time was mainly in CAD programs.

File: image004.gif (10KB, 522x580px) Image search: [Google]

10KB, 522x580px

This kills the Voodoo.

>>4150749
Unironically DirectX killed Voodoo. They should have jumped the Glide train sooner.

>>4150517
Did you actually check the list of GLIDE only titles? They often have software rendering or DirectX/OpenGL as well.

>>4150749
>>4150756
Except their main downfall was the poor business decision of purchasing one of their OEMs.
https://www.youtube.com/watch?v=3MghYhf-GhU

File: image004.gif (9KB, 571x584px) Image search: [Google]

9KB, 571x584px

>>4150749
oh really, check mate

>>4150780
I remember Voodoo 5 being a huge card. It was probably the biggest consumer 3d accelerator I've ever seen.

>>4150780
Based 3DFX

File: image010.gif (9KB, 527x584px) Image search: [Google]

9KB, 527x584px

>>4150780

>>4150818
Geforce 2 supports hardware tnl while voodoo doesn't. Few games actually used it when it was first released, Quake 3 does though. So geforce win isn't really suprising.

>>4150787
https://www.youtube.com/watch?v=ZmCEXbspWuI

>>4150827
I always thought that thing was a joke.

>>4150830
Retard

>>4150825
>Few games actually used it
OpenGL does by default. Even then, at such high resolutions, the benefit of hardware T&L is reduced, which is why 5500 is closing in at the GTS.

>>4150835
>Retard
Excuse me for never hearing about or seeing this abomination untill the 10s.

>>4150841
It IS a joke. Someone with too much time just turned that joke into a reality.

>>4150847
Well, now I feel stupid.

>>4150857
>>4150835

Glide on 3Dfx cards had very effective depth of field effects for the time using what was basically dithering. There's never been an equivalent since and nothing emulates it on output, so you'd need actual hardware for it even today.

>>4150787
You're thinking of the satirical voodoo 6 or all-in-one mockups. The 5 is big but by no means huge

The first Voodoo was revolutionary. It was a $150 ($200 if you bought the Canopus Pure 3d like I did) card that only played games accelerated for it. You had to have a 2d card for everything else. But it did its accerlating so well that everyone that was into pc gaming had one. I put many hours on that voodoo 1 playing multiplayer Descent 2 on kali.

>>4151126
I thought it was 299?
http://computeme.tripod.com/righteous.html

File: 5pGC5M2.png (33KB, 568x437px) Image search: [Google]

33KB, 568x437px

>>4150517
yes, i'm using one now to play diablo II.

>>4151176
why tho

>>4151176
how is the temperature monitored?

>>4151176
>forceware
Yeah, right.

>>4151132
Maybe that was the release price. I remember paying $200 sometime in '97, must have been later on in that year. I see a post in alt.games.descent about Righetous 3d and Monster 3d for $210, and on sale for $186 during april of '97.

>>4151181
D2's Glide acceleration is better than DirectX and has more features.

I think people forget that the first two Voodoo cards had blur reminiscent of N64. They used the same kind of dither filter.

>>4150517
>>4150698
It is more complicated than that, it wasn't just an add on like PhysX.

Glide was a low level API that could work more directly with the hardware and therefore gaining an important boost in performance while maintaining very high visual quality, this was not possible with the other APIs, they were not as developed and the CPUs were very week.

It was a market that was in diapers so they made an API and tried to become a monopoly, it was ok at the time.

>>4150592
Even though you are a fucking tripfag and don't deserve it, I will reply to you.

No, the NV1 came out in 95 and the ATI Rage came out in April 1996, both before the Voodoo that came on October that year (at the same time than the Rendition V1000)

>>4150676
>2002 (37yrs ago)
>2039
Woah. You came back from the Future?!?

>>4150676
https://www.youtube.com/watch?v=1-w1NziFoLg

Much ricer video of what Glide can do with a Nvidia AGP card instead.

>>4151458
Those are not 3D accelerators they're video cards with 3D capabilities. Voodoo was in a class of its own, technically even if its price to performance ratio hadn't put it there (which it did)

>>4151521
>Those are not 3D accelerators they're video cards with 3D capabilities
There is no difference between the two.

>Voodoo was in a class of its own
Even though this is true.

>>4150827
Holy fuck, someone made one.

>>4151078
>Glide on 3Dfx cards had very effective depth of field effects for the time using what was basically dithering.
Is there any examples?

>>4151576
>>Those are not 3D accelerators they're video cards with 3D capabilities
>There is no difference between the two.
Those early cards were often referred to as "3D decelerators" because they added features rather than accelerated. You could get 30fps in 8bit software mode, or 20fps in 16bit hardware rendering. And the faster CPU you bought the wider the gap. The Voodoo was the big dog that gave massive performance boosts WITH the high res, WITH the texture filtering, WITH the coloured lights... etc.

>>4151737
I had a voodoo 2 and programmed it with glide and the only thing I can think of is that he's confusing the fact that it had 3 hardware features that it was unusually good at:
1) Tri-linear filtering to smoothly blend between mip-maps so you don't get that transitional line along the screen at fixed depths
2) Hardware fog so distant objects will smoothly blend with the fog colour to mask popup
3) 24bit colour dithered to 16bit for output

The combination of all 3 working together would make close things sharp and distant things smoothly blended into blurry and indistinct the further away you got. Which is kinda like depth of field, but not really. I'd say DoF is reserved for dynamic focus shift where you can control what is in focus and what isn't and you can't fake that with a voodoo without rendering the scene in multiple passes and hacking it with software tricks.

>>4152026
>Those early cards were often referred to as "3D decelerators" because they added features rather than accelerated
You shouldn't take the decelerator meme literally. Firstly, it depends how good your CPU is. On a shitty 486, you can bet that S3 Virge would increase performance with hardware acceleration on. Secondly, those 'features' you are talking about actually required a lot of extra processing power by the video card vs typical software rendering modes, and they'd do them faster than your typical contemporary CPU if it tried - hence it is still 'accelerating' 3D.

Just because Voodoo was the first 3D card for PC that wasn't mediocre doesn't make it the first 3D accelerator.

Also the NV1 and Rendition cards weren't so bad that they 'decelerated' 3D and they both came out before Voodoo. They were both simply made obsolete by Voodoo's significantly better performance.

>>4150517
Yes. You need to go back, kiddo.

Next question.

>>4152040
>You shouldn't take the decelerator meme literally. Firstly, it depends how good your CPU is. On a shitty 486, you can bet that S3 Virge would increase performance with hardware acceleration on.
No, I had one, it didn't. The slower your CPU the closer to framerate parity you might get but if only because 15fps in software = 15fps in hardware. Two heaps of shit are not better than one. The problem was the cards only accelerated a small part of the 3D rasterisation pipeline and they weren't fast enough at it to offset the overhead of card state management.

>Just because Voodoo was the first 3D card for PC that wasn't mediocre doesn't make it the first 3D accelerator.
I'll give you that literally you are correct, but I'd argue that until the voodoo, 3D acceleration was marketing bullshit that didn't actually have practical real world use. The NV1 and Rendition were not an insignificant cost to bear and the advantages to a gamer were largely insignificant.

File: mario64.webm (2MB, 640x360px) Image search: [Google]

2MB, 640x360px

>>4152032
For the benefit of the poster you're replying to, I think this webm is a good example. Mario 64 does 1) Tri-linear filtering 2) Hardware vertex fog 3) 8bpp dithered to 5bpp. It's all pretty similar to Voodoo cards (3dfx were ex-SGI engineers, SGI was the company that designed N64).

Notice how when the camera moves further away from the platformer its color changes and becomes more hazy. That's the hardware vertex fog, which causes a psuedo depth of field effect.

>>4152051
>The slower your CPU the closer to framerate parity you might get but if only because 15fps in software = 15fps in hardware
That's not quite how it works. It depends what the bottleneck is, which differs per game. If the rasterization process is the thing bottlenecking the CPU, then having a 3D accelerator (even a crappy one) will improve speed since it'll take that one burden off the CPU's shoulders.

And those early cards didn't necessarily only accelerate a small part of 3D - it really depends on the game. Anything with a lot of blending for example would probably be much faster on an S3 Virge than a crappy 486.

Of course if you have a really good CPU which can rasterize faster than a shitty GPU can, then the GPU will be a decelerator. And if the CPU's bottleneck is simply game logic/triangle setup or anything that the CPU would have to do anyway, then there's no benefit to the shitty GPU.

>3D acceleration was marketing bullshit that didn't actually have practical real world use
Yes, that's beyond dispute, at least on PC.

>>4152052
Hardware vertex fog deserves a special mention, for being the feature that has the worst usage and defaults, and keep getting used worse and worse.
Its notable that in Mario64, the fog is only somewhat far away, while the trend in the early PS2 days was for fog to start close to the POV, and then muddle everything. Its very notable in games like SOTC, Drakengard and a lot of games. Heck, its very notable in a lot of later N64 games, like the Heracules one.

I just think it looks bad. I get the point: Apply it from a reasonable distance, and then use that fade to have a clean wall of render cut off.
But it looks so bad when poorly done, in modern games like Planetside 2.

>>4152054
>>The slower your CPU the closer to framerate parity you might get but if only because 15fps in software = 15fps in hardware
>That's not quite how it works. It depends what the bottleneck is, which differs per game. If the rasterization process is the thing bottlenecking the CPU, then having a 3D accelerator (even a crappy one) will improve speed since it'll take that one burden off the CPU's shoulders.
I feel like I'm horribly nitpicking here because you're not wrong, it's just in terms of how to achieve what you're talking about it's a bit optimistic.

The 3D accelerators at the time only did the interpolation part. You had to give it 2D screen space co-ordinates. Meaning you had to do all the computationally expensive transformation, culling, clipping, vertex lighting, etc.

And the elephant in the room is that while the card was interpolating over the 2D warped triangles, your CPU very likely WASN'T seeing any benefit. Why on earth was that? Because game logic, network code, even sound generation isn't actually a huge part of the runtime. This was 1994 before graphics cards become supercomputer GPUs and we filled the empty CPU with physics. 90% was 3D transformation and rendering. So instead of your CPU crunching span interpolation, it was now waiting until the card's render buffer had space for more triangles because it had nothing else to do.

So while you are technically correct that even a virge was taking load off the CPU, the problem is the game was designed to work within the limitations of a software renderer so if CPU went from 100% to 75% there wasn't actually anything else for it to do with that 25% since it was waiting on a slow card instead of its own polygon rasteriser. And like I said, the virge doing bilinear filtering was easily 50-60% slower than the CPU doing point sampled.

>>4152118
>Meaning you had to do all the computationally expensive transformation, culling, clipping, vertex lighting, etc.
I know what you mean, it was that many earlier GPUs didn't do triangle setup (and of course, they didn't do T&L). The Rendition, however, did triangle setup.

But what the GPUs did do (all of them) was rasterization, texturing, blending, etc. This was still a pretty significant part of the pipeline.

>90% was 3D transformation and rendering
Yeah, but the GPU did the rendering time. How much time was triangle setup + T&L vs rendering. It would probably vary game-by-game, but I'd say the processing time would be close to 50/50.

> the game was designed to work within the limitations of a software renderer so if CPU went from 100% to 75% there wasn't actually anything else for it to do with that 25% since it was waiting on a slow card instead of its own polygon rasteriser
I don't think it quite works that way. The issue with earlier cards like S3 Virge was that when you put all of the features on, there wasn't any speed benefit over CPU and often it went backwards.

For example, the Virge's bilinear texture filtering takes 4 cycles. That's a pretty enormous fill rate penalty (even if the Virge could do it faster than the CPU). And I'm guessing most games that supported the Virge put all of the features on to "show off" what the Virge could do. My feeling is that if the Virge's settings were customized to have less features on, it would certainly be a real accelerator over the base CPU. Since then the CPU and Virge could work in parallel to produce frames.

Yeah, Voodoo was not the first accelerator, but it was the first one that wasn't a pile of stools.
It had the full basic features needed for something like GLQuake, while at the same time having enough horsepower to run fast.
It only got there due to its sacrifices in other areas. No 2D, limited selection of resolutions, 16-bit colour only, 300 dollars at launch.
It knew just exactly what gamers needed, and that is what they got with this package.

>>4151737
http://www.combatsim.com/htm/july99/3dfx-3.htm
http://www.anandtech.com/show/350/4

I remember replacing my Voodoo2 setup with a GeForce way back and despite the boost in performance and better color I found myself having more trouble judging distances (especially in Tribes) and missing my V2. I didn't understand why at first.

>>4152375
The T-buffer is a feature on new Voodoo cards only, though.

>>4152379
The earlier cards use dithering to produce a more primitive version of the effect which later became tbuffer. It didn't appear out of nowhere.

>>4152384
Is it really a boon to have the distance blurred? What I like about game rendering is how sharp and "perfect" things look. It is an escape from our dull, messy, griddy, imperfect reality.

>>4152391
>autism speaks

>>4152391
Not in Tribes unless you're a scrub like the original poster was.

no game ever used voodoo5 t buffer

>>4152393
Anyway, in retrospect, it didn't age particularly well. Competitors like TNT and Rage 128 were offering feature-complete 3D acceleration with high resolutions and colour depths to boast. Nobody in their right mind would dare to push these cards this far for actual gaming, but it was nonetheless a peak into the future.
It's personally hard for me to go back and play games in 640x480 on a Voodoo 2 without being bothered by the degraded visual quality.

>>4152419
>autism returns: now with more broken english

>>4152424
>fag advocating for blurry games calling other people autistic

voodoo 1 and 2 blurryness is mostly due to the pass through cable

>>4152424
Apart from my embarrassing typo, what is broken?

>>4152437
Nah, passthrough cable only makes the image a bit softer. Voodoo 3 looks less blurry because they changed the dither filter algorithm to a better one.

>>4152464
Was the dithering ever soft? If anything, I thought it was noisy, and the new algorithm softened it up.

>>4152493
The new algorithm softens up the dither better without going full vaseline smear.

>>4152502
Is it perhaps another card I have seen that dithers with zero softening applied?
Is there a good comparison between the new and the old?

>>4152506
If you want to see an accelerator that has blatant grid-like dither patterns everywhere, the RIVA 128 is your card.

Don't think there are any good comparisons out there. The dither filter is applied after the framebuffer, so screenshots can't capture it. This was actually a big point of contention back in the day because it was hard for Voodoo 3 owners to demonstrate that 16-bit color wasn't so bad cause the card could post-process it nicely.

>>4152514
Why is it limited to 16-bit anyway? If it does indeed render internally at 24/32 before writing a dithered output to the framebuffer, why not keep it that way?

>>4152529
>why not keep it that way?
It's to simplify both pipeline cost (16 bit inputs and outputs on the hardware alpha blender in particular) and to simplify memory bandwidth costs (less framebuffer data being shifted).

I really wish OpenGL didn't stop being a thing, modern DirectX is so shit. Here's hoping Vulkan becomes huge.

>>4152615
Would it be too complex to allow both?

>>4152649
It wouldn't, but at the time 3dfx were basically refining the Voodoo 1 over and over. Curiously, the N64, despite some similar hardware, actually supports 24 bit output.

>>4152672
Didn't the N64 also support some form of accelerated T&L? To think that PC didn't get that until late 1999.

>>4152684
Yes it did. But the N64's GPU does have two downsides compared to the original Voodoo.

Basically the Voodoo has separate blending and fog units, which means it can do alpha and fog in a single pass. N64 has a combined blender/fogger unit so to have a pixel both blended and fogged it needs two passes.

The other downside is the Voodoo can do mipmapping in a single pass, while the N64 needs two passes. Both hardware need two passes to do trilinear filtering though.

Other than T&L and 24-bit output support, what the N64's GPU has over Voodoo is way more efficient AA. Voodoo's AA basically involves an insane number of alpha blends on edges, while N64's method has comparable quality and is far more efficient.

>>4152718
How fast is the N64 with 24-bit, though? Comparing to something like the difference between 16 and 24 on a TNT or other early 24-bit capable chips.

>>4152776
>How fast is the N64 with 24-bit, though?
It's single-cycle because like Voodoo, the pipeline is 32-bit internally. Both chips can only write 5 bit alpha maximum to memory, but the N64 can write 8 bit colors to memory, while the Voodoo can only write 5 bit color to memory. I guess the designers of N64's chip wanted to give it more flexibility.

The speed penalty for N64 writing 8 bits per color (24 bit) is down to the increased pressure on memory bandwidth and extra framebuffer space consumed rather than the GPU itself. You do get some speed back because you don't need to dither filter colors in 24-bit color mode (though alpha still needs dithering). N64 has way less memory bandwidth than the original Voodoo let alone a TNT.

Consequently, only one N64 game used 24-bit color mode. That is, Quake 2, and it's only with the Expansion Pak.

Has there ever been a PC 3Dfx card as powerful as the N64's reality co-processor which is the size of a fingernail?

>>4152840
>the N64's reality co-processor which is the size of a fingernail?

If you are talking about die size then yes, any PC graphics card with respectable yields would be the same size.

>>4152804
>Quake 2
Based.
That game looks great on both PS1 and N64, but in their own ways. Too bad its transition to consoles didn't get the attention it deserved.

>>4152890
Probably because one of the biggest parts of Q2 was MP which is pretty neutered on consoles. I still like the SP for what it is though.

>>4152894
Quake 2's multiplayer seems like a weak point to me. Its ticrate is fixed at 10Hz. Quakeworld's netcode is way better for good connections.
Heck, the multiplayer maps were added after the game's release.

>>4152840
RCP isn't nearly that small.

Memory bandwidth aside (which is far inferior on N64) RCP and Voodoo 1 have a fairly similar performance level. Voodoo 2 on the other hand is much faster - we're talking 4x the texture performance of Voodoo 1 and RCP.

The one thing RCP has going for it is HW T&L which no Voodoo card ever supported. 3dfx cards all supported triangle setup though, and I imagine triangle setup speed of Voodoo 1 is similar to RCP, while faster on Voodoo 2 and later.

>>4152905
In what situations does a Voodoo suffer in comparison due to its lack of T&L? My uneducated guess is whenever there is heavy use of dynamic lighting.
I wonder how a notable game like Conker would fare if it were ported and ran on a Voodoo 1.

>>4152917
Without T&L, it just means the Voodoo becomes more CPU dependent on things like dynamic lighting. The N64's CPU has the performance level equivalent to a very highly clocked 486.

Given that RCP's T&L consists of an 8-element (128-bit) vector unit running at 62.5mhz, I imagine you'd need a Pentium CPU with several hundred megahertz to match it at T&L. Of course in practice it's doing more than just T&L...it also has to process the N64's sound...

>>4152905
Voodoo 1 does not support full triangle setup

>>4152980
Yes you're right my memory is failing me there. IIRC they finally put full setup into the Rush and 2.

>>4150517
>id know the answer if I'd been born then
Yes

>>4152904
Sounds like someone that never played Quake 2 online. Even today, its one of the most played arena shooters with way more players than QW has at any time. There is nothing better in Quakeworld aside from later improvements from open source that Quake 2 also has. the only strange thing is almost nobody uses antilag in Q2 even though it exists.

>>4150517
Ut 99 on glide was an eyegasm that made the 3dfx card worth it

Power VR for life.

>>4153167
PowerVR is so bad that it's good.

https://www.youtube.com/watch?v=gYiiPHsMzBI

File: 3dfx.jpg (373KB, 2460x461px) Image search: [Google]

373KB, 2460x461px

>tfw

>>4150517
Yes.

We need a /sqt/ on the board.

>>4154398
For what purpose

>>4150749
Voodoo was finished with the Voodoo3 and the budget failures like the banshee.

>>4154398
I was stupid enough to get the 3500 tv, never used all of its features

File: 3096ac4fc47d17220d8ca216af00524e.jpg (221KB, 1024x576px) Image search: [Google]

221KB, 1024x576px

>>4155391
some people have a lot of money

>>4155639
Creative's boxart really was the truest to the 3dfx design used since Voodoo 3.

>>4150517
LOL. OP is a retard. Assuming the 3D card market was as saturated then as it is now. Most people had no idea what a 3D card was, furries, genderqueer or whatever faggot blender words you have now.