NVIDIA has been marketing their new RTX API and at SIGGRAPH 2018 they have showed several new slides.
The new RT Core is introduced. Since it is “RT”, as the name implies, it is directed to Ray Tracing, which is used to speed up the processing of light in a three-dimensional environment. The speed of processing light rays is 25 times that of the current Pascal architecture.
This could lead to some powerful ray-traced shaders which would make games look much better with some tweaking of the models etc.
This is on top of the known tensor cores and CUDA. NVIDIA seems to be doing hybrid.
At first these GPUs are being sold for workstation use. The flagship RTX 8000 is likely to be rather expensive.
It’s interesting that despite these individual speed-ups, NVIDIA’s overall performance promises aren’t quite as extreme. NVIDIA should have a SDK available soon for developers to use.
One thing is obvious that NVIDIA is now designing logic to do several things depending on the job assigned. This suggests NVIDIA may want sell such hardware to consumers down the road.
The RTX 8000 can handle an 8K professional display which suggests this card may find use in the film industry. NVIDIA is well known for rendering in the film industry where they compete with AMD.
Ray Tracing has a long history on the PC. In the early days of the PC, Intel sold a floating point chip which typically was very expensive. Rival floating point hardware was available with our AMD 386-40 with a Cyrix FPU which was slightly faster than the 80486 which ran at 25 Mhz.
POVRAY is an open source ray tracer that runs on the PC. It started off using a DOS extender before it was eventually converted to Win32, POVRAY using the CPU and if available it will use the FPU.
Images used to take many hours to render at 640×480 with POVRAY but faster modern machines are able to now render the original award winning artwork in a few minutes at higher resolutions.