This PDF (https://www.nvidia.com/content/dam/en-zz/Solutions/design-visualization/technologies/turing-architecture/NVIDIA-Turing-Architecture-Whitepaper.pdf) provides a lengthy and detailed description of how new GPU architecture (called Turing) makes ray tracing faster. Interestingly, it seems as if these GPUs have an order of magnitude more transistors than a CPU (which may have previously been used for ray tracing?). I wish there were more detail on how the Turing architecture actually makes ray tracing so much faster.
THWG
I think for ray tracing you can refer to this article: https://blogs.msdn.microsoft.com/directx/2018/03/19/announcing-microsoft-directx-raytracing/
For the number of transistors, you can refer to this wiki: https://en.wikipedia.org/wiki/Transistor_count. GPU does not necessarily have more transistors than CPU as you can see. GPU is an ASIC that throws away lots of redundant functionality (e.g. complicated branch predictors, control handling structures) and becomes specialized on highly parallelizable programs such as graphics program. Ray tracing, similar to rasterization, is a highly parallelizable kernel so more powerful GPUs like Turing are enabling it to happen in real-time.
InfinityAxiom
Not only does RTX need dedicated RT cores, it also needs to be able to denoise the ray-traced image in real time. It uses an algorithm called ross-bilateral filtered denoising to accomplish this. It is a proprietary algorithm developed by NVDIA, and you can read more about it here in this awesome GDC talk introducing RTX: https://www.gdcvault.com/play/1024813/
This PDF (https://www.nvidia.com/content/dam/en-zz/Solutions/design-visualization/technologies/turing-architecture/NVIDIA-Turing-Architecture-Whitepaper.pdf) provides a lengthy and detailed description of how new GPU architecture (called Turing) makes ray tracing faster. Interestingly, it seems as if these GPUs have an order of magnitude more transistors than a CPU (which may have previously been used for ray tracing?). I wish there were more detail on how the Turing architecture actually makes ray tracing so much faster.
I think for ray tracing you can refer to this article: https://blogs.msdn.microsoft.com/directx/2018/03/19/announcing-microsoft-directx-raytracing/ For the number of transistors, you can refer to this wiki: https://en.wikipedia.org/wiki/Transistor_count. GPU does not necessarily have more transistors than CPU as you can see. GPU is an ASIC that throws away lots of redundant functionality (e.g. complicated branch predictors, control handling structures) and becomes specialized on highly parallelizable programs such as graphics program. Ray tracing, similar to rasterization, is a highly parallelizable kernel so more powerful GPUs like Turing are enabling it to happen in real-time.
Not only does RTX need dedicated RT cores, it also needs to be able to denoise the ray-traced image in real time. It uses an algorithm called ross-bilateral filtered denoising to accomplish this. It is a proprietary algorithm developed by NVDIA, and you can read more about it here in this awesome GDC talk introducing RTX: https://www.gdcvault.com/play/1024813/