I know that some types of renderers, like those used in animated films, can take several minutes to render a single frame. Obviously games like this one render much more quickly. What's different between them? I presume those used for film are rendering higher quality images in some way, but what are they doing that games aren't? Is it just a question of resolution?
anonymous_panda
I guess that resolution will be an important factor. The movie screen is much larger than the desktop screen. Even the new GTX 2080 can only render 4k gaming at 60fps.
harveybia
@silentQ Films are made with high vertex objects with ray tracing, more real physics simulation and other weird CG technologies because they do not need to be real time. For games since we need high throughput, engines are designed with lots of tricks like fake lighting, fake physics (especially fluid stuff), etc. and objects in games certainly dont have as high-poly surfaces as films do. There must be even more differences and resolution definitely is not the only factor.
merc
Piggybacking off this conversation: what are some tricks people use to render in real-time for video games? Probably one of them is the whole raytracing+rasterization thing, but what kinds of shortcuts are taken with physics, etc.?
zbp
To the best of my knowledge, high-performance computer graphics applications make liberal use of near-realistic approximations of physical phenomena to trick the viewer without paying the full price for realistic simulation. For example, many physical phenomena like refraction, N-body gravitation, mirrors, etc. are all computationally expensive to emulate precisely. For the phenomena I listed I'm aware of shortcuts like surface modeling, aggregating across neighbors, and secondary + downsampled camera views. In general, most of these kinds of approximations violate laws like conservation of energy or momentum, etc. so I doubt they'd be well-suited for accurate physical simulation.
keenan
It's not about resolution; pixels are cheap.
It's more about illumination: figuring out how light bounces around the scene is expensive.
We'll talk about global illumination in depth later in class, including why it's slow (but also beautiful!) :-)
keenan
@merc There are way too many "cool real time tricks" to list; to name just one, screen space ambient occlusion is a pretty common trick these days for increasing the shading quality in a non-physical but still plausible way.
I know that some types of renderers, like those used in animated films, can take several minutes to render a single frame. Obviously games like this one render much more quickly. What's different between them? I presume those used for film are rendering higher quality images in some way, but what are they doing that games aren't? Is it just a question of resolution?
I guess that resolution will be an important factor. The movie screen is much larger than the desktop screen. Even the new GTX 2080 can only render 4k gaming at 60fps.
@silentQ Films are made with high vertex objects with ray tracing, more real physics simulation and other weird CG technologies because they do not need to be real time. For games since we need high throughput, engines are designed with lots of tricks like fake lighting, fake physics (especially fluid stuff), etc. and objects in games certainly dont have as high-poly surfaces as films do. There must be even more differences and resolution definitely is not the only factor.
Piggybacking off this conversation: what are some tricks people use to render in real-time for video games? Probably one of them is the whole raytracing+rasterization thing, but what kinds of shortcuts are taken with physics, etc.?
To the best of my knowledge, high-performance computer graphics applications make liberal use of near-realistic approximations of physical phenomena to trick the viewer without paying the full price for realistic simulation. For example, many physical phenomena like refraction, N-body gravitation, mirrors, etc. are all computationally expensive to emulate precisely. For the phenomena I listed I'm aware of shortcuts like surface modeling, aggregating across neighbors, and secondary + downsampled camera views. In general, most of these kinds of approximations violate laws like conservation of energy or momentum, etc. so I doubt they'd be well-suited for accurate physical simulation.
It's not about resolution; pixels are cheap.
It's more about illumination: figuring out how light bounces around the scene is expensive.
We'll talk about global illumination in depth later in class, including why it's slow (but also beautiful!) :-)
@merc There are way too many "cool real time tricks" to list; to name just one, screen space ambient occlusion is a pretty common trick these days for increasing the shading quality in a non-physical but still plausible way.