Hard Light Productions Forums
Off-Topic Discussion => General Discussion => Topic started by: Hellstryker on April 15, 2008, 01:19:10 pm
-
I always wondered how many polygons the Galatica or Pegasus had, or more importantly... How the hell do you render such a thing? You would need like 20 9800 GX2s O_o
-
Need a computer fast enough the crack the human genome :p
-
Network Rendering, you use about 30-40 Computers and each computer only renders one frame, and saves it as a numbered bitmap, then it doesn't matter when the frame is finished as they can all be joined up later :)
Some colleges etc actually rent out their terminals for exactly this sort of job overnight and during holidays :)
-
Holy mother of god :jaw: FORTY computers!?
-
Large number of polys aren't really a problem it's when an artist starts adding lighting and such that a pc grinds to a crawl.
-
Large number of polys aren't really a problem it's when an artist starts adding lighting and such that a pc grinds to a crawl.
So that in the end, the large amount of polygons is actually a problem?
Disclaimer: I don't know anything about rendering in the computers. But what I do know is that I have written a couple of ray-tracers, for different purposes than rendering. Those programs take hell a lot of processing power (and yes I optimized the code if you wondered), funny things are that if you either have many parts or a complex surface (like a polygon model), ray-tracer has to do a lot of intersection tests to find out to which surface ray hit. And then, if the actual surface is defined by a polygon or a spline, the fun will actually start when trying to find the accurate hitpoint. Unfortunately, real time rendering with Splines (in this case actually NURBS) will need a couple of years of processor development until feasible.
Mika
-
It the combination of the 2, a 20,000 polygon model with high levels of shaders will still render up far faster than a 2 million polygon model, because there's only 1% of the number of objects in the scene being calculated for.
-
Holy mother of god :jaw: FORTY computers!?
Shouldn't be that impressive...I've been in computer labs with rows upon rows of opteron and xeon processors all working together (probably 150-200 computers) in high speed researcher computing. You also need three truck sized air conditioners to keep the place cool. When you do CGI for TV and movies you need to have a whole bunch of computers thrown at the task for the final production render as they take ages to calculate and output. Often what will happen is the scene is divided up frame by frame with each CPU working on one and then moving to one down the line as it finishes up.
Real time rendering of that stuff will eventually happen but allot more power needs to be thrown at it first.
-
In Babylon 5 they originally had a cluster of (IIRC) 10 Amiga's.
-
And then, if the actual surface is defined by a polygon or a spline, the fun will actually start when trying to find the accurate hitpoint.
Damned. Instead of polygon it should have said polynomial.
It the combination of the 2, a 20,000 polygon model with high levels of shaders will still render up far faster than a 2 million polygon model, because there's only 1% of the number of objects in the scene being calculated for.
This is true, but given the context of either polynomial or spline surface, the numbers are more even. This is because there is a lot more to calculate when determining ray hitpoints when surfaces are parameterized by functions.
I would believe that rendering splines is where the computer graphics will be going next. This is because a model needs a lot less control points (i.e. memory usage) when it is represented by spline, rather than a polygon, and it is more easier to blend it, or to deform it smoothly, without adding points on the fly. The surfaces will appear continous even when zooming close. But I'm not sure about the texture maps if they could avoid pixelization there. Some strange effects would also be visible when applying textures since there actually are smooth bumps in the models. On the contrary, separating two pieces from a single control point mesh is harder when there is a requirement that the pieces should be cut along that arc and should preserve their shape. It could be difficult to make it look like something blew up with those models.
The limiting factor with splines is computing power. Current papers state something like a 100000 control points object being rendered 5 frames/second, but this needed a couple of standard PC computers, and I suppose, not that accurate ray-tracing.
Mika
-
I seem to recall there is limited spline support in DX 10 cards, but I'm still amazed it has taken them this long to create graphics cards with internal particle processing abilities to take the weight off the CPU itself.