New Unreal engine vs Doom3s

Shaviro said:

This looks sterile and static. How could a cave/temple ever look like this? Somebody changed the surface to shiny metal, painted and shaped like rock. I do not really care whether the UW engine is better than D3's. The artists who made that UW area are just too poor, IMO.

I wouldn't call the UW artists poor, merely "less proficient than id's artists".
So you're referring to the art ok, fair enough, but art aside, you've got to admit that the UW engine itself is a *tad* ahead of the Doom 3 engine.

dsm said:

I wouldn't call the UW artists poor, merely "less proficient than id's artists".
So you're referring to the art ok, fair enough, but art aside, you've got to admit that the UW engine itself is a *tad* ahead of the Doom 3 engine.

Judging from that screenshot? No!

Dima said:

20/30 1.4 million polygons each on screen at once? not a chance. They're probably using a process similar to DOOM III's and polybump, using the high-poly model (the 1.4 million one) for generating all the lighting, shadowing, bumpiness on the low-poly in game one (which i guess could be a bit higher in polys than the models in DOOM III, I'd guess 6000-7000 polys).

I know its very high, but thats what they said on the video. And i believe it, i saw the model of the elf showing all polys ( the wire model ) . And was so high poly, it was AMAZING. It was just like the high poly version of the pinky from that magazine picture, but then ever HIGHER in polys!

Just watch the nvidia movie. BTW its copper connections running at 500 mhz what do u expect?:). Geforce fx is the new generation of videocards remember.

No, I'm sorry. You realize that with even 15 1.4 million-poly models on-screen, that's over 21 MILLION, MILLION I SAID polies? That's just for characters? Tack on a similarly-detailed scene and you're looking at like thirty billion polygons. That's ridiculous. Even the largest renderfarm in the world would CHOKE AND DIE. Think before you post, ignorant fool.

Check up on your information and quit eating Nvidia propaganda. The Geforce FX is, for real-world (read: gaming) purposes no more impressive than the Radeon 9700. It's marginally faster and somewhat more flexible but since it far exceeds the specifications for DirectX 9, it's unlikely many games will use the extra features the Geforce FX has to offer. On top of that, Nvidia requires an apparatus that looks like it was concieved in a Craftsman workshop to get it to 500Mhz, the speed it needs to run at to outperform the Radeon 9700 at all. With a realistic cooling solution, it's doubtful the Geforce FX will be much more than decent competition for the ATI chip.

All in all, it's a nice chip that will be on some nice cards, but I predict that it's really too little, too late. It's certainly not "the new generation of video cards." Nvidia's already lost this round. I hope for their sake they have something else up their sleeves.

[edit]Whoops, misread my calculator there. Pardon. I haven't used engineering notation in a long time, alright? Geez. :P [/edit]

___ did you take 1.4 million polys, multiply by 15, and end up in the billions?

I guess she multiplied it by the 100 fps she gets in Q3A in 2048x1536.

And there's no need to censor the word "how".

AndrewB said:

___ did you take 1.4 million polys, multiply by 15, and end up in the billions?

Whatever way you look at it, there is simply no way GeForce FX would be able to render 20-30 1.4M polygons characters on screen at once as Sweeney is saying... now, i'm not saying he is lying, but as I said in my previous post, he didn't reveal all the details. I bet that the 1.4M polygon model is the original model used for generating the required data for a much lower in-game model. If that's the case, then surely GeForce FX (and Radeon 9700 for that matter) are both capable of outputting 20-30 characters like that on screen at once (i assume he meant ALL of them on screen at once?).

About the GeForce FX itself...
As the situation stands right now, the chip is already running at NVIDIA labs at frequencies beyond 500mhz! (550-700!)!!
The chip is more capable and more advanced than R9700pro(supports longer instructions, dynamic flow control in VS, conditional mechanism in PS, etc...), but it does have a few noticable drawbacks even in regards to R9700pro...
First of all, it utilizes a 128-bit memory bus, opposed to 256-bit memory bus on R9700pro... granted, the real benefit of 256-bit memory bus could be clearly seen in bandwidth intensive situations (AA, Aniso and combined modes). Besides that, the latest charts show promising perfomance figures, higher than R9700pro (mainly due to the much higher core & memory clockspeeds), although I'm yet to see any reasonable comparisons to R9700pro in bandwidth intensive situations mentioned above.

All in all, this is definetly a great piece of silicon, but like certain invidivual here stated, ATI outpaced NVIDIA and NVIDIA lost this round of technology superiority... they have some major gaps to fill in the coming months and they better do it quick... both accelerators are of the same generations and we can definetly consider them "the next generation of video cards", with a slight edge given to NV30 for it's incresaed architecture flexibility (which kinda reminds me of the situation with PS1.4 on R8500 that weren't utilized - that might turn out to be the same here).

Dima said:

Whatever way you look at it, there is simply no way GeForce FX would be able to render 20-30 1.4M polygons characters on screen at once as Sweeney is saying... now, i'm not saying he is lying, but as I said in my previous post, he didn't reveal all the details. I bet that the 1.4M polygon model is the original model used for generating the required data for a much lower in-game model. If that's the case, then surely GeForce FX (and Radeon 9700 for that matter) are both capable of outputting 20-30 characters like that on screen at once (i assume he meant ALL of them on screen at once?).

Geeeeeeeeeeeeez you've forgot the whole LOD. afcourse its no 20-30 * 1.4M. It alot less polys. Oh my god, where is the world going to...

I'm with Kat, the sole thought of having 20 - 30 x1.4 million polys characters in screen would be the equivalent of having enough triangles to fill up 45 1024x768 screens with one-poly-per-pixel detail. Clever texturing would produce perfect replicas of any given photograph.

And no, this is not the polycount for the reference models either. There's absolutely no need to use that many polys, not to mention is near impossible to work on a character with that many wasted polygons. Most modellers rely on building relatively low poly versions of their characters (low poly meaning 60,000) for example, and smoothing out edges with an algorythm commonly refered as NURMS. While the iterations of this process are adjustable, any model that requires that much polys to look good, provided that its not a fucking submarine blueprint, is just a shitty piece of art. I've made scenes with +1 million polys, and some renderers are able to process 5000 times that number, but I've never, ever, needed more than 600,000 polys for a single object.

Mark Rein is full of shit, just like when he announced the UT2k3 demo as coming out "shortly after e3". Or maybe the journalist is a fucking idiot. The point is, you can't expect an engine, no matter how well coded it is, to outperform the target hardware.

Go read some figures, check out the optimal triangle rate for a Radeon or a GFX. This number is often an exaggeration, and it doesn't take into consideration not even a single texture pass, prolly only 15% of that works under the stress of other game subsystems, and yet, is nowhere near enough to sustain a slideshow framerate in such a scene.

Even with LOD, the average polycount is way out of range. A single 1.4 million model is already out of range unless it's the whole scene in a very very next-gen game.

The elf model probably weights about 200,000 polys, and that's no surprise since I remember 100,000 polys demos for the original GeForce. It's so easy to cram that many polys in a controlled enviroment like a demo, but to expect that in a game coming out the next year or the other one is ridiculous.

Besides, texturing is what makes stuff look good. Fresnel falloff masks in the fairy's skin, specularity maps for each dot in her body, opacity maps in the wings, bumpmaps everywhere and the like.

The UW engine shows all of those effects, plus soft shadows. Does anyone even realizes it's not only a matter of rendering 1.4 m models at decent speed, but rather rendering 1.4 m models with 6+ texture passes plus casting each poly's shadow(s)?

Katarhyne said:

Anyway, anyone interested in seeing some pictures which prove that the leaked version of Doom 3 isn't "low-quality" needs to contact me via IRC or AIM. I'll show you what Doom 3 REALLY looks like, not some screenshot some guy took on his Voodoo3.

Can I see some of those screenshots? How can I contact you.

Zaldron said:

Most modellers rely on building relatively low poly versions of their characters (low poly meaning 60,000) for example, and smoothing out edges with an algorythm commonly refered as NURMS.

Thanks for the new vocabulary word. Never heard of NURMS before-- looks interesting.

I think the general point most of the pro-fx people are tring to make is that video cards are becoming more and more able to fill in detail as objects get closer. This has NOTHING to do with the initial polygon count of the object. You could model a simple cube, with 12 triagles. As the cube gets closer to the screen, and ONLY as the cube gets closer to the screen, polygons are added, until the cube is right up against the front viewport plane, and THEN cube will have a million polygons or whatever. Not EVERY object in the scene will have a million polys, just the surfaces that are really closer to the screen. Most models in the scene, as Zaldron pointed out, can get by easily with much fewer polys. This is because they are farther away, and additional polys would be wasteful (there would be more than one poly per pixel, which is pointless on the current range of hardware). Again, NOT EVERY MODEL HAS 1.4 MILLION POLYS ALL THE TIME. It's just, as Zaldron alluded in the off-line situation, that smoothing and detail generation is needed when the objects get closer, so they don't look blocky. Polys are generated, ON THE FLY, as needed.

About the fact that the number textures and effects used should also be considered: naturally, however, as the vertex and pixel pipelines operate concurrently, we aren'y losing anything here-- the time to draw a triangle will be limited by the time to generate the polygons OR (NOT and) limited by the time it takes to do the pixel shading. Since the both work at the same time, the times for vertex proccessing and pixel shader proccessing are sort of parameters in a max(t_!, t_2) function, not an add(t_1,t_2) function.

In general, I don't think the GFFx vs. 9700 is what is important, but the direction the general hardwares are taking: distance-based adaptive tesselation of geometry.

Ok this is something new. Even though the images are blurry as hell unreal 2 is gona fucking rock!!!!!!!!1
http://www.stud.uni-goettingen.de/~s304280/

Those pics are fucking sweet man.
Unreal 2 looks pretty awesome.

Sorry, but the blurriness disguises the limits of the engine, making it look smoother and more impressive than it is in a crystal-clear UXGA display where you can see every microdetail.

Plus, the game isn't going to rock, it's going to be just another boring FPS game with predictibly bumped graphics. I am not one bit impressed. In fact, I'm disgusted.

[Mod]________S MOTHERFUCKER, DO YOU SPEAK THEM!?[/Mod]

AndrewB said:

I am not one bit impressed. In fact, I'm disgusted.

This is what bugs me the most about graphics. I enjoy them quite a bit (I tend to be a very visual person), but it's really getting to the point where games really are becoming like a walk-thru movies. And it's going to get worse, and companies scramble to produce art content for these new geometry and pixel hungry graphics engines. I heard once that Microprose developed entire games using just stick figures before any art work was done.

I think this would be a wonderful way to go about developing first person shooters. Just get the game designer and one or two programmers, grab the Quake1 engine off of id's site, and code the gameplay and interface using that technology. Since the engine won't support even 32bit color or highres models, there's no temptation to start off with "wouldn't this be a cool graphics effect." Just work the gameplay until it's so fun you could almost ship it with Quake1 graphics, and it would still fly off the shelves. After that work is done, hand it off to the artists and gfx engine designers, so they can go nuts for as long as the development funds hold out.

Fun factor doesn't make games fly off the shelves. Stupidity does, and mainstream gamers are the honeypot.

So who mentioned polycounts in the million ranges? Heh. Those shots show only mariginally more polygons than DOOM 3.

Looks like this new game suffers from the same lack of good visual design that the other Unreal games have, by the way. Crappy textures, crappy character design. Ugh.

Fredrik said:

So who mentioned polycounts in the million ranges? Heh. Those shots show only mariginally more polygons than DOOM 3.

Looks like this new game suffers from the same lack of good visual design that the other Unreal games have, by the way. Crappy textures, crappy character design. Ugh.

FFS READ READ READ. This is unreal 2, this isnt that NEW UNNAMED UNREAL GAME. Geeeez.

BTW 1.4 million polys * 20/30? Who said all of them where gonna be rendered as 1.4 million polys, and who says all the models are close? Maybe they can render 20-30 models at the same time in far distance.

1.4 millions per model can't be true. For one thing, could you try to imagine what memory that would require? It would require 400 megs of RAM just for vertex positions. Yes, that's without animation, textures and so on.

Fredrik said:

So who mentioned polycounts in the million ranges? Heh. Those shots show only mariginally more polygons than DOOM 3.

Looks like this new game suffers from the same lack of good visual design that the other Unreal games have, by the way. Crappy textures, crappy character design. Ugh.

Teletubbies design.

Fredrik said:

1.4 millions per model can't be true. For one thing, could you try to imagine what memory that would require? It would require 400 megs of RAM just for vertex positions. Yes, that's without animation, textures and so on.

Like i said, they where probably less poly when they where all on the screen due of distance.

BTW they might have had a over 1 gig of ram comp to show it on.

Like i said, they where probably less poly when they where all on the screen due of distance.

Uhh, are you suggesting that objects aren't loaded into memory when they're far away? So the game reads 50 megs from the hard drive every time you close up on an object. That sounds very intelligent indeed.

BTW they might have had a over 1 gig of ram comp to show it on.

Add in composition and texture data and you're FAR over a gig of memory.

Give up, it's not even remotely realistic.

Fredrik said:

Uhh, are you suggesting that objects aren't loaded into memory when they're far away? So the game reads 50 megs from the hard drive every time you close up on an object. That sounds very intelligent indeed.

Add in composition and texture data and you're FAR over a gig of memory.

Give up, it's not even remotely realistic.

Less polys = alot easier for the VIDEO CARD. And geez afcourse is realtime loading slow. BTW, if not 1 gig, they can easily have 2 gig of ram. And ever heard of swapping?

Look. The hi-res character models in SHREK were about 800,000 polygons. With newer games like Unreal 2, you're looking at the low thousands.

AndrewB said:

Look. The hi-res character models in SHREK were about 800,000 polygons. With newer games like Unreal 2, you're looking at the low thousands.

Shrek has alot more polys. Far in the millions. ( I remember they matched doom 3 high poly version model with shrek, which was about 1/10 of shrek). And this engine isnt anything like unreal 2. Pls stop about unreal 2, the game we're talking about uses a whole updated next-gen engine.

AIRMichael, you're a retard.


The Shrek character model had 1.5 million polygons, and here's a fact:

EACH FRAME IN SHREK REQUIRED ONE HOUR TO RENDER.

Do you think computers have gotten like 50000 times faster in one year?

Fredrik said:

AIRMichael, you're a retard.


The Shrek character model had 1.5 million polygons, and here's a fact:

EACH FRAME IN SHREK REQUIRED ONE HOUR TO RENDER.

Do you think computers have gotten like 50000 times faster in one year?

Each frame = a whole ENVIRONMENT.
BTW read again, did i say each model has 1.4 millions of polys in that new unreal game? Even Tim didnt say that, he just said 20 to 30 models being rendered. Geez man, if they where only 10.000 polys a model due distance, its only 200.000 to 300.000 polys for ALL the models.

First you need to learn how to read. And second, you dont have the nerve to call me a retard in real life.

Each frame = a whole ENVIRONMENT.

So, assume that the character model makes up 1/100 of the whole scene, and you're still dealing with a rendering time 2000 times worse than realtime.

BTW read again, did i say each model has 1.4 millions of polys in that new unreal game? Even Tim didnt say that, he just said 20 to 30 models being rendered.

You're actually right, you didn't say that there was going to be 1.4 millions of polys per character in the new Unreal game. However, you've been arguing in your last five or so posts that 1.4 million in-game polys per model is possible and that despite people actually having shown proof that it isn't possible while you haven't provided any facts to back up what you said.

Jesus motherfucking Christ put a goddamn 1.4 million sphere filling up the entire screen, untextured, running under a Radeon 9700 and see for yourself the kind of framerates it gets.

STOP FUCKING ARGUING

/me pours some c0ff33 for everyone so they can talk it out

And I do agree, AirMichal is a freakin clownboat, bar none :P