Slashdot Mirror
3dfx Unveils Info Regarding Voodoo 4 & 5
A reader wrote to us about the latest press release from 3dfx regarding the Voodoo 4 and 5.
The V4 and V5 will apparently be released in March of 2000. The V4 will be
single processor, but the V5 will have both a commercial and professional
version, respectively supporting up to 4 and up to 32 VSA-100 processors, and up to 128 and 2GB of RAM
each. The release for the V4 and V5 is rolled in with the VSA-100 talks - definitely worth checking out.
I can't figure out why everyone is so happy about 3dfx putting out another Voodoo chip. They're pushing a proprietary interface (Glide), where a perfectly good standard exists instead (OpenGL). They're using market pressure to get game manufacturers to adopt their standard, and lawsuits against developers who try to write Glide wrappers so that Glide-only games can be played on other video cards.
Doesn't this sound a bit like another company that everyone is up in arms about?
- Drew
- In Capitalist America, law violates YOU!
- Linux support?
- This is just another trick by Microsoft
- Wow, I'd like to see a Beowulf cluster of these..
Personally, I think that this is great... let's hope iD is keeping up and giving us RealGuts(tm) in Quake 4.Remember how the cheapo motherboards used to be able to allocate some of the system RAM for video RAM? It would be pretty funny if these cards could do the opposite.
--
3Dfx unveils good number for his new 3d architecture...but It lacks of geometric acceleration. Do you still want 200 fps at 1600x1200 32 bit with 3 big polys at each frame? :) No thanks, I prefer lower fill rate and higher polys counter.
Windows 95, 98, NT4.0 and Windows 2000 drivers Allows you to run the Voodoo5 6000 AGP with all popular operating systems.
Okay, not only am I defending linux on this one, I am also wondering where the MAC drivers are. If 3dfx wanted to have some incredible benchmarks they should write a MAC driver and throw it into a G4. They say the Voodoo 5's aren't only for gamers, why not port the drivers to the most popular graphics design platform?
I fail to understand why this stuff excites people. I've always thought that the market for add-on 3D graphics cards was going to develop a lot like the market for add-on sound cards did, and so far I'm seeing nothing that indicates otherwise.
What I mean is -- consider for a moment how the market for add-on sound cards developed. Up to 1992, sound on the x86 PC was basically nonexistant, unless you owned a flaky almost-compatible like the Tandy 1000. Then the multimedia tidal wave hit and suddenly there was consumer demand for hardware sound support -- and a market sprang up to fill the demand.
Once the demand for sound cards sprang up, the market developed through 3 distinct stages in the next 5 or so years:
So this is where we are today in sound cards -- while a few enthusiasts care about buying the latest Sound Blaster Live! or whatever, the vast majority of users are happy with the 16-bit audio that's hardwired into their motherboards. It's Good Enough!
And that's what's going to happen in the 3D card marketplace, IMHO, fairly soon. We've already passed through stage 1 (I remember agonizing over whether to buy a Voodoo1 or a Rendition Verite card) and stage 2 (with 3Dfx milking their brand name for all it's worth through the Voodoo3). But now Good Enough 3D hardware is starting to come integrated on motherboards, and 3Dfx's Voodoo-only APIs have been almost entirely forsaken in favor of Direct3D, which is integrated into the OS. I've run 3D games on cheapo PCs using this integrated hardware, and while the performance isn't great, it's Good Enough -- while the add-on card companies fight over which card can provide 80 fps in Q3Test, or other "features" which would be lost on the average consumer anyway. So watch for it -- in a year I'd be amazed if there's still a market for whizbang add-on cards. Most people will be just fine with the Voodoo2-level hardware they'll get free with their PC.
-- Jason A. Lefkowitz
Read my blog.
Upon seeing the specs for that baby, part of me just screams I want it, but the other, more rational part of me wonders what the point really is.
I mean, great: gigatexels per second. As much RAM as I currently have on my mainboard. Meaning what? I can now play Quake3 at 4,000*3,000 resolution? Yay. Yes, I know about anti-aliasing, but this is overkill for even that if not running very righ resolutions (1024*768 and above).
Read my lips, 90% of all speed problems with games on current hardware is the geometry setup bogging down the processor. Unless you play at above mentioned resolutions, or happen to have dual athlon 700s and are playing at 100 fps already (and if I am right in assuming that this does not have a Geometry chip like the GeForce) this card will be exactly 0% faster for you.
In my opinion Nvidia have taken a much wiser approach to the whole 3d acceleration concentrating on the weekest pointinstead of just pouring in endless amount of pixel fillrate that the processor can't render anyways unless you are stairing at a blank wall.
-
We cannot reason ourselves out of our basic irrationality. All we can do is learn the art of being irrational in a reasonable way.
The page with detailed info concerning these boards is www.3dfx.com/prod/voodoo/newvoodoo.html
The really interesting thing is that *once again* 3dfx promised us more than it will deliver. On the low end (Voodoo4 4500) these babies are getting smoked by the GeForce 256, which will be a half a year older! The GeForce can do 480 Megapixels per second, about 1.3 times as fast as a Voodoo 4 (which clocks in at 367 Megapixels per second).
If the past is any indication it at least a few more months for the Voodoo 5 to be released (ignore what 3dfx says), by this time Nvidia will probably already have a better card.
In summary, the Voodoo 4 is slower and less feature rich than the GeForce 256, plus is won't be out for 4 more months. It could take longer for the Voodoo 5 which will probably be an anachronism before it is released.
Come on 3dfx! This is *not* the technology that will keep us ahead of the PSX2!!!
WOW !! You are amazing, I've never seen such a high concentration of incorrectness in one post. Let's do this blow-by-blow, shall we ?
> It's nice to see 3dfx sticking to the Glide standard rather than some proprietary OpenGL nonsense that doesn't port anywhere.
Glide is a proprietary interface owned and VIOLENTLY copyrighted by 3Dfx. They sue the ass off of anybody who tries to figure out how it works, make a wrapper, etc. Search the slashdot archives for details. OpenGL, on the other hand, is an open, FULLY portable to basically ALL platforms available, complete graphics API.
> Every tried to run a TNT in anything except Windows? The 3D works like shit.
Oh, then I guess the fact that I was just playing Quake 3 was just my imagination, because it looked great at 1024x768x32 bit color. I could have SWORN I got a TNT2.
Try updating your drivers.
> Sure, Glide might be slower in Linux/FreeBSD than Windows, but it is a simple, fast, and more efficient protocol than the crap that nvidia is turning out anyday.
nVidia didn't 'turn out' OpenGL, SGI did. And Glide actually works FASTER in Linux than in Windows. Even when you're making concessions, you're blatantly wrong. Dolt.
> nvidia users, you own a very nice 32MB card that can do wonders on a 2D desktop...look at X go!
The only part of your post with a shred of accuracy. The 2D on this card is almost as impressive as the 3D.
25% Funny, 25% Insightful, 25% Informative, 25% Troll
a typical Slashdotter...I will pay $5000 for a video card, but $40 for Word Perfect is a travesty, because I can't see the source code.
makes sense... but they could port the PCI versions...
That's what it might look like at face value, but I have found rarely that engineering makes these decisions:
>because they choose the best features to include
This is usually done by the marketing department, and often to the chagrin of the engineers, who would like to think they know better...
Seth
$5 / month hosted VPS on linux = awesome!
- Quantum's dual Voodoo1.
- 3dfx Voodoo2 (SLI)
- ATI Rage Fury Pro 128 Max Turbo Fast Thing, or whatever the hell it is called.
- Bitboys Glaze3D (if/when it comes out).
Everyone implements it bit differenty, but the idea of SMP Video is not new. CAD-people, who use cards that cost more than a decent car, have had this stuff forever.now do we have the hardware to support Virtual Reality?
Macintouch is reporting that the PCI version is already supported by the company's existing Macintosh drivers. You can read the FAQ yourself. No doubt, if there is enough interest, drivers for AGP Macintoshes will be forthcoming.
Sig goes here
T&L is a gamble. High fillrate is a gamble. Bump mapping is a gamble. Any new feature a chip manufacturer puts on thier chip is a gamble.
No one knows what game companies are going to try next.
There is no way of telling whether hardware transformation and lighting is going to make any difference at all in future games. Sure, nvidia is going to tell you that future games will depend on it! There is no way of knowing that 5 gazillion texels/sec is going to really make much of a difference to future games, although 3DFX doubless wants you to think that. No one knows whether game companies are going to stuff their games with bump-mapped polygons, no matter how much Matrox tells you it's the truth.
Point is, each of these hardware developers are hedging their bets, that game companies will favor their technology.
As for us consumers, I would take a "wait and see" approach. I'd never go out and buy the latest and greatest until I see what games run well on them and what games do not run well on them. Specs from pre-released hardware are meaningless, and even released hardware that runs a FEW games spectacularly is nothing to base a purchase on.
Look for the architecture that stands the test of time, and has support for the platform and games you play.
All of these Nvidia GeForce/3dfx Voodoo 4 and 5 boards are technically amazing but this level of consumer 3D hardware is in desperate need of a new killer app. I have a Voodoo 1 and I'll likely be more than satisfied with the performance of Quake Arena on that thing. I simply refuse to drop hundreds of dollars more to play games at higher resolutions/framerates.
Why, when we all have a global network right in front of us that is ablaze with information and commerce, is no one strapping a hardware-accelerated 3D engine onto the net? When am I going to be able to navigate the web in 3D? When can I use my 3dfx or Nvidia board to do real work, or to shop, or to explore real information and news? Why is the web still 2D? Wake the fuck up. Screw VRML - I'm not even asking for any sort of server tech - just give me a fly-through 3D-abstraction of the HTML/XML content that is already there. If people know the engines are out there they will start to build for them.
Bottom line - I won't give 3dfx or anyone else more of my money to play another FPS. I will part with more money for 3D hardware when I can use a Voodoo 5 6000 to give me an Ono-Sendai Cyberspace VII-like window into the net.
Night
First some quotes (from the press release):
Re: the Voodoo4
The boards, which render two fully featured pixels per clock, will deliver between 333 and 367 megatexels/megapixels per second
Re: the Voodoo5 5000/5500
The board, which renders four fully featured pixels per clock, will deliver between 667 and 733 megatexels/megapixels per second fill rate
Re: the Voodoo 5 6000
The Voodoo5 6000 AGP, which renders eight fully featured pixels per clock, will deliver between 1.33 and 1.47 gigatexels/gigapixels per second fill rate
Now, if you'll notice they state how many "fully featured pixels per clock" each card delivers. Also, notice that the V4 does 2, the V5-5500 4, and the V5-6000 8. Along with that, as I guess one would expect, the V5-6000 has double the fillrate of the 5500 which has double the fillrate of the V4.
So? What's my point? Well, with the Voodoo2 -- which could render two pixels per clock -- the full fill rate was acheived only if the app was rendering two pixels per clock. (ie. multitexturing) If the app wasn't multitextured, the effective fillrate was actually only half the "marketing" fillrate. I think this was also the case with the Voodoo3, although I'm not positive.
I'm not saying that this is definitely the case with these cards, but:
Correct if I'm wrong, but I think these cards are still based on the same architecture as the V1, V2, and V3.
3DFX is somewhat notorious for advertising the higher "marketing" fillrate as opposed to the true fillrate.
The fact that they qualify the fillrate of each card by stating the number of render pixels per second kind of worries me.
If this is the case, apps that don't take full advantage of the high end cards (ie. have less than 8 pass multitexturing) may leave you with nothing more then a glorified and expensive Voodoo4.
How you managed to avoid having your post not moderated to flame-bait is beyond me.
Matrox offers an AGP card with two outputs. It's called the G400. Unlike the V4 and V5, it's already released and available.
Anyway, I'm somewhat off topic. But, I needed to correct this. The issue with PCI is bandwidth and texture swapping. If the card truly can have up to 2GB of RAM, (Yes, there are many simulation visualizations and mappings that can use this.), you'll need more than the 533 MB/s provided by the PCI bus. Even full AGP 4x (w/ RDRAM) has 1.06GB/s. An approximately 2s delay is damn noticeable.
If they can put multiple graphics processors on one card, why can't they put multiple output ports on the same card?
--Al
the G4's have a 133mhz AGP 2x slot
And why is this insightful? Though I should point out in counterpoint of this previous post that nVidia has NEVER reached the hype of their products in the past. The TNT was supposed to be a TNT2. That's why they never released any GeForce specs until release was imminent.They didn't need to release 'real' specs though, they had Tom to blatantly plug the NV10 long before it existed.
And on the other hand, 3dfx has never failed to meet their specs. Even though their products have severly lacked in many departments (AGP texturing, 32 bit colour), at least they didn't play the Hype Game that nVidia did.
And I don't buy 3dfx products anymore. My first accelerator was an original Voodoo Graphics, that cost me close to $300(cdn). Since then it was an i740 (hey, I was on a budget) and now a G400. And the G400 is the best card I've ever had the pleasure of using, it's fast, pretty (environmental bump mapping BABY!) and more Open Source friendly (specs vs un-improvable open source driver). Granted, 3dfx is definitely the least open source friendly.
Anyway, that was my rant. I'm happy now.
Finally getting rid of the 256x256 texture resolution limit is a Good Thing as well. Even Quake2 uses textures larger than that, and on the Voodoo[1-3] chips it just looked blurry and crappy because of it.
That said, I wonder what sort of marketing spin 3dfx's wonderful PR people will put on this decision, when for the longest time they were constantly saying how worthless 32bpp rendering and large textures and the like were. I also wonder if these chips will have true accumulator buffers (the press release didn't say anything about this) or their bastardized, crippled "T-buffer" crap. I also wish they'd drop extending Glide (for a number of reasons) and only have Glide 3 for backwards compatability, especially since Glide can be relatively trivially implemented in terms of OpenGL and adding on more features to Glide to try to make it catch up will just cause more cumbersomeness and an even greater rift between their Windows and Linux support. (I feel even sorrier for Darryl Strauss if he's got to do even more for-free work on extending Glide for a relatively thankless company.)
On the whole, though, 3dfx has a chance to actually redeem themselves with this new card. I hope they don't blow it; I'm all for giving them another chance. I just hope that they decide to actually have a good product instead of good marketing. For the longest time they seem to have just been resting on their laurels from having been the first usable (and not even decent) 3D card on the market. Maybe now that can finally change.
---
"'Is not a quine' is not a quine" is a quine.
"'Is not a quine' is not a quine" is a quine.
Quine "quine?
yeah... for real... if anything it should have been moderated interesting
That is an unfair moderation. That is an intelligent statement regarding what the poster wants. Who do you submit unfair moderations to?
At a threshold of 3 I saw a lot of "so what" posts. Here's my response.
Naturally, I'm all for a "let's just see when it comes out" attitude, but to answer the "oh great, even more texels" argument, remember that a massive fill rate means your geometry engine can have more overdraw without hurting performance. (Overdraw is where you "draw" multiple pixels into the frame buffer at the same place, and the one with the lowest "z" or distance from the observer is the one that actually shows up). If you had an infinite fill rate, you could draw the entire world as fast as your geometry setup would give you vertexes. Up until now engine designers have had to use tricks like BSP trees (ala DOOM) and Portals (Decent) to get overdraw as close to 0 as possible. With a high enough fill rate, you can get sloppy with your hidden surface removal and focus on other things. Of course, this is an over-simplication, but the point remains that more texels/s is not a bad thing.
Also, CPUs are still getting faster and cheaper. It will not be unusual to see dual-processor machines in homes next year. With Athalon using Digital's bus technology, quad processor machines could become Christmas pressents in 2000.
To answer the "what do I need 2G of textures for" question, think computed textures and textures with more information than use colors. If a texture has depth (bump mapping) or material information (alpha channel, refraction), it adds up. quake 3 uses 32-bit textures: 8 bits each for red, green, blue and alpha (transparency). Now let's immagine what we could do with another 32 bits: 8 bits of depth, 8 bits of reflection (I forget what this is called), and 16 bits for whatever effects would look good if they varied over the face of a polygon. Also, animated textures will quickly use up texture memory.
Yes, what we have now is pretty cool. Yes, 3DFX is being unfriendly to open standards. Yes, other cards may be a better bet. No, this is not the end-all-be-all of real-time scene rendering. Personally, I can't wait to get my hands on almost any of the cards that's going to be coming out next year.
Disclaimer:
I do not work for or even know anyone who works for 3DFX. I know two people who work for Creative Labs, and they hate 3DFX.
Don't think 3200x2400x32x60: think 1024x768x32x60...
...with five specular highlights from five dynamic lightsources, flickering of torchlight on faces and more sharply from metal surfaces, and every barrel or crate or object slightly different from having each one overlay about three slightly different 'dent' or 'dirt' layers. ;) if you did, you'd know that this is _precisely_ what ILM did to make the battledroids photorealistic- they were all identical models, but you had the texturemap for the robot, and then five different overlay textures putting different patterns of dirt and wear onto the droids- which were applied in combinations, of course.
I take it you don't read Cinefex
'Cinematic' means impressive- means multitexturing that would _choke_ a GeForce (or indeed a Voodoo3, but that's a given). It means the modellers will still be caring about polys, but the _skinners_ can go HOG WILD. Surfacing is not merely choosing a really big texturemap- talk to rendering people- overlaying translucencies and transparent textures is when you start getting really startlingly impressive effects. This throws the door _wide_ open for really amazing stuff. Polys aren't everything (it should be OK on polys anyhow, but polys aren't everything).
*g* ;)
Seriously. It's coming out in PCI, I want that because I'm hanging onto my nice old powermac for a while. I know _exactly_ what to do with all that texture bandwidth- multitexturing babeee! *g* forget polys. You'll end up with really boringly textured well sculpted shapes- geometry is NOT the weak link. I've appreciated the 3dfx strong points even through the drawbacks of 16/22 bit color- I've seen the transparencies and shading and tonal values looking better at 16 bit than the competition at 32 (not always, but in a number of cases, and always due to the 32 bit card drawing washed out tonal values). Now that 3dfx is ready to do the card with antialiasing that works with all my existing games, and with so much texture memory and fill rate that you could use it for fscking _filmmaking_ without it breathing hard *hehe*, well, I'm there. Build it, I'll buy it. My voodoo2 needs replacing, and I've never been more pleased that I didn't start planning to try and get a GeForce or something.
The output of this card _will_ look better than GeForce, by an order of magnitude. That's a prediction. That's also assuming a lot of multitexturing, but hey- if it's good enough for ILM, it's good enough for _you_
Microsoft's Direct3D has been tracking the latest developments in cards: D3D 7.0 will have direct support for lightmaps and stencil buffers, for example.
Where is OpenGL headed? Is anyone furthering its development? Is it going to track the features it needs to stay competitive as a games API?
Otherwise we're stuck with Glide and D3D. Talk about a Hobson's choice.
I've finally had it: until slashdot gets article moderation, I am not coming back.