Game Development Stack Exchange is a question and answer site for professional and independent game developers. It only takes a minute to sign up.
Sign up to join this community
We usually use OpenGL or DirectX to work with the GPU. If I want to use the CPU to render 2D/3D graphics, how does it work and what API can I use for it on Windows?
I know there is no benefit in using the CPU for graphics when GPUs exist, but I'd still like to understand how it works.
GDI (Graphics Device Interface) is the software renderer under Windows. Basically any language/runtime platform under Windows that is not GPU-accelerated is going to be using GDI under the hood at some level. While Java AWT might use GDI directly via the C code that the Java runtime is written in, something like Flash running in Chrome will OTOH be using GDI if the desktop isn't GPU accelerated, vs DirectX if it is.
I know that there is no benefit of using the CPU for this purpose, but only to understand the operation.
Not at all. It can be a lot easier to simply plot lines, pixels and images, than to know the ins and outs of GPU-based graphics programming. If you intend to write native code to access GDI directly, maybe check out the gdiplus lib, as outlined in this tutorial. Good luck!
EDIT: For accuracy, it may also be worth noting that there are CPU implementations of OpenGL; Mesa being the common example. So, strictly speaking, OpenGL does not exclude software rendering. But that's worth mentioning only if you really want to be pedantic!
Actually, starting with Windows 7, Direct3D 11 is your answer. Of course the API defaults to using a GPU if you have one, but you can create a Direct3D device targeting the Windows Advanced Rasterization Platform (WARP), which is meant to be a high performance software rasterizer supporting the Direct3D API. You should not expect performance anywhere as close as that of a GPU, but it should be pretty good as far as CPU rasterization goes since it exists for Windows to support systems without hardware acceleration. Direct3D also provides a software reference rasterizer, but it's only designed for correctness and is terribly slow.
For more details on Direct3D 11 rasterizers, see the D3D_DRIVER_TYPE enumeration.
Edit: But can it run Crysis? Yes!
I wouldn't we so fast in jumping to "there's no benefit" conclusion. Whilst GPUs are absurdly fast at processing triangles, not all of the images are made using triangle rasterization!
For example, voxel-octree based renderers (used, for example, in medicine, for visualizing CT scans) often utilize multicore CPUs and write raw pixels basing on completely different algorithms. That can also be true for some raytracing algorithms.
In general, the simplest way to render using CPU is to output the data to, for example, a BMP file. You can also use OpenGL to create a window, and simply write raw pixel data to a texture and render over the whole screen.
