At the SIGGRAPH conference, Nvidia presented several new products at once. The main one is the eighth proprietary platform named Turing, which is distinguished by a deep study of ray tracing. It is complemented by a new GPU architecture, which is implemented in three Quadro RTX 8000, 6000 and 5000 chips. They are intended for the film industry, video creation and simulation.
With the release of Turing, they made the biggest leap in a generation, according to the company's CEO Huang Renxun. A distinctive feature of the platform is that tensor kernels are widely used here, effectively implementing the work of artificial intelligence. The computational power of the RT cores is combined with the tensor core processing algorithms to achieve 6x the speed of ray tracing over its predecessor, the Pascal platform.
The Quadro RTX 5000 is an entry-level chip with 16GB of memory, 3, 072 CUDA cores and 384 Tensor cores. The RTX 6000 is slightly better, with 24GB of memory, 4608 CUDA cores, and 576 Tensor cores. But the top-end RTX 8000 chip with the same number of CUDA and tensor cores as the 6000 model received double the amount of memory. As for the key ray tracing ability, the RTX 5000 chip is apparently capable of handling 6 Giga beams per second, and 6000 and 8000 models up to 10 Giga beams per second.
Such powerful video cards cannot be cheap and are not meant to be bought spontaneously. The Quadro RTX 5000 costs $ 2300, the RTX 6000 model costs $ 6, 3 thousand, and the top version is all $ 10, 000. The area of application is described as "wherever you need to work with professional graphics." New powerful modules will go on sale at the end of this year.