PDA

View Full Version : Number of triangles for DVD quality games?


eGoatee
10-28-02, 10:40 PM
I think current high end video cards push around 3 to 4 hundred million triangles per second. Wonder if any advanced Nvidia or other cards for Cad and gaming developing like 3DStudio put out half a millon triangles/sec. Anyway how many billion trianges per second is needed for 3d games to play with Dvd quality? 10 billion or something like that? I can't seem to find it on the net yet. Is pushing more triangles the main obstacle to getting movie dvd quality gameplay?

And by dvd quality I mean fully interactive with characters and objects that are as clear as dvd movies, like The Matrix. So it would be like playing and being able to move all around in Mortal Kombat or the Die Hard series. Actually gameplay graphics as clear as a 60's regular movie like War of the Worlds would be a temendous leap foward, but everything's going to original dvd production.

eGoatee

thcdru2k
10-28-02, 10:52 PM
wow..i don't think you can really pinpoint how many triangles you need..by dvd quality, its essentially a step below real life quality. this is probably the holy grail as far as graphics are concerned. with current techniques and technology we are years and years away. something revolutionary will be created in the next decade or the next. its certianly is more than triangles. take a look at the world for a second..colors..lighting..shadows. we are definately decades away.

thcdru2k
10-28-02, 10:56 PM
by revolutionary, i'm thinking something similar to the technology in matrix. something along the lines of brain stimulation, using a direct creation to the brain to recreate realistic lifelike gaming. sounds far fetch but i wouldn't be suprised if it happens in next hundred years. they already have the ability to implant a camera into a person's brain and sending signals along the brain's pathways, and allow the blind to see in black and white. though its nowhere usable as of now, because its not portable. this was 2-3 years ago i believe.

Chalnoth
10-29-02, 02:17 AM
Number of triangles for DVD-quality, eh?

Well, first of all, modern high-end graphics cards for the consumer market are (very roughly) capable of about 50 million triangles per second. It's not yet close to approaching a few hundred million per second, not for real-world applications. For movie-quality effects, it would be even less, as better visual quality will mean more processing per triangle, not just more triangles.

Now, onto the thought experiment. This will be very rough, but I'm only attempting to get within about a multiple of ten of a realistic value.

Let's say that the resolution is 720x480 at 30 frames per second, non-interlaced. Movie-quality graphics generally require polygons that are much smaller than screen pixels.

So, let's say that 16x supersampling is used (multisampling doesn't make sense as triangles are smaller than screen pixels). Let's also say that there is roughly one triangle per pixel subsample.

That makes for 16 triangles per pixel, or 720 x 480 x 16 triangles per frame, for a total of 5.5 million triangles per frame.

At 30 fps, that's about 166 million triangles per second.

But we're not done yet. At a mere 30 fps, motion blur is an absolute necessity. Let's again select 16 samples for the motion blur, leading to a grand total of 2.65 billion triangles per second.

This is very roughly 50 times the triangle count today's high-end video cards are capable of. And don't forget that movie-quality graphics would likely require more processing per vertex than today's graphics.

In the end, we're a number of years away yet.

OliverRedfox
10-29-02, 05:20 AM
You also need to take into account memory and cpu power for physics. For example: Right now they use tricks to make hair. But for true quality, you would need to be able to render every hair individual, calculate wind effect on each strand, plus gravity, plus of motion of the person they were attached too. This would have to be done for every hair on a persons head, even if the hair wasn't visable, since the location and movement of one hair could effect other hairs.

Then for a person faces to look real, you would need to take into account bone structure, muscles, and skin. When muscles move in your face to talk, they flex and relax. As a result your skin is moved in many different ways. It would take calculating all of this information to create a proper facial movement.

The computational power would be massive to just render one person corretly. You would need tons of memory to keep everyone of these details stored. So to creating MATRIX quaility games is a long way off.

However, with techniques to fake it(like pixel shaders), we'll keep seeing higher and higher quality games. 'DVD like' games will eventually appear. But there will still be rendering tricks used to make it look more real.

Chalnoth
10-29-02, 09:11 AM
And there's also the fact that many other visual effects, such as reflection/refraction caustics (when a reflecting/refracting object, such as the surface of a pool, lights the surroundings) are N^2 algorithms. That basically means that when you double the resolution, the computational power required goes up by a factor of four.

For realtime rendering on silicon-based proessors, it is impossible to work with N^2 algorithms, which means you have to do tricks, and will always need to do tricks.

Quantum computers will be able to reduce many N^2 algorithms to N algorithms, while not decreasing the accuracy of the calculations.

DaveW
10-29-02, 09:33 AM
I think it depends on the scene and what texturing techniques are used (e.g. displacement mapping).

The inside of a solid plain box can be rendered in DVD quality with 12 polygons (2 polys per side).

Phyre
10-29-02, 06:17 PM
I hope this isn't changing the subject, but I'm wondering how much quantum based CPU's, GPU's, and memory might help in this situation?

Phyre

Chalnoth
10-29-02, 07:14 PM
Originally posted by Phyre
I hope this isn't changing the subject, but I'm wondering how much quantum based CPU's, GPU's, and memory might help in this situation?

Phyre

Well, they might. I'm not really certain.

The primary benefit of quantum computers is the ability to turn many N^2 algorithms into N algorithms, which could make them usable for realtime.

The only thing I don't know for certain is whether or not graphics algorithms that are N^2 algorithms (such as radiosity-type lighting) can be satisfactorily translated for computation by a quantum computer. But, I do really doubt that it is impossible to do so. I think somebody will find a way (It may be really easy...don't actually know).

raystone57
10-29-02, 09:40 PM
These articles seem to say the triangles per second for latest vid cards are a lot more than 50 million. This one for the 4600 apparently says 136 million triangles/sec unless they mean a different function.

http://www.3dsolutions.co.uk/3doffer_geforce4ti4600.htm

and this one says the 9700 is "300M" triangles/sec.

http://www.anandtech.com/video/showdoc.html?i=1656&p=6

Could they be confusing triangles with vertices per second or something else? Dunno for sure. But the other poster said 50 million, so if he's right are these articles referring to some other process?

raystone

Chalnoth
10-29-02, 10:40 PM
No, they're not confusing anything. Those are the theoretical maximums. They're just not publishing real-world numbers.

But by the way, vertices per second is actually the only true measure of hardware T&L performance. Triangles per second and polygons per second numbers depend mostly upon software optimization. A best-case scenario places the triangles per second number at twice that of the hardware's vertices per second limit.

raystone57
10-30-02, 03:42 AM
Clalnoth, that clears up what I've read on the net about it. I didn't realize the real world numbers before.

raystone