The processors of modern videocards (GPU's - Graphical Processing Units) are SO powerful, that they're much faster at rendering than CPUs. Here are the first results of a test project that introduces GPU rendering in LuxRender. If they pull this off, it promises an incredible speedup!
The frames in the following sample only took seconds to render:
From the demonstration video's description:
Testing the incredibly tiny OpenCL software demo SmallptGPU (and a quick look at the OpenCL accelerated SmallLuxGPU and Bullet Physics as well). The OpenCL fun appears to be just beginning!
Path tracing, yes the typically very slow path tracing, which used to take hours to render, is now getting a helping hand from OpenCL. The frames in these animations took only seconds to render almost entirely on my GPUs (CPU utilization literally was at 0%). Notice the soft shadows (from true spherical area lights), indirect bounced light, color bleeding, caustics, etc.
Using AMD Stream SDK v2.0 with OpenCL 1.0 support on two ATI Radeon HD 4890 GPUs (Catalyst 9.12 Hotfix drivers on Windows 7, no CCC, not Crossfired). Now I want 5970s :P. AMDs implementation of OpenCL and SmallptGPU also allow exploiting the full CPU as an OpenCL device as well, with SmallptGPU it helped with one GPU, however not with two. With SmallLuxGPU, you seem to be better off using native threads for the CPU and OpenCL for the GPUs.
In the spirit of this test I also used the experimental partially (btBatchConstraintSolverOCL) OpenCL accelerated version of the Bullet Physics engine to simulate the physics. It appears still too early to be using this version.
Note that none of the software programs tested in this video are full featured packages, they are for demonstration purposes at this stage.