Details of nvidia's turing architecture appear

Turing is the new Nvidia graphics architecture that brings GeForce RTX 2000 series graphics cards to life, for now few details of this architecture are known, although Videocardz has released some pretty interesting data.

Turing offers a host of floating point and memory compression enhancements

The Turing architecture adds a new unit of execution (INT32). This unit will allow Turing GPUs to run floating point and non-floating point processes in parallel. Nvidia claims that in theory this should provide an additional 36% performance on all floating point operations. This parallel execution will be made possible by the new unified architecture for L1 shared memory and texture caching. Nvidia claims that the INT32 / FP32 core design and other changes to the new streaming multiprocessor provide a 50% improvement in performance delivered by CUDA core.

We recommend reading our post on What is rasterization and what is its difference with Ray Tracing

The Turing architecture also offers new lossless memory compression techniques. Nvidia claims that its further improvements to Pascal's algorithms have provided a 50% increase in effective bandwidth in Turing compared to Pascal. Turing also features a new DisplayPort 1.4a compliant video engine, making it possible to use 8K resolution at 60Hz. Turing graphics cards can control two 8K displays at 60Hz via DP or USB-C. The new engine incorporates an improved NVENC encoder that can encode H.265 stream at 8K / 30 FPS, and a new NVDEC decoder with HEV YUV444 10 / Support 12b HDR, H.264 8K, and VP9 10/12 HDR.

Lastly, the TU102 silicon features two second-generation NVLINK x8 lanes, while TU104 is equipped with a single x8 link. TU106 silicon is not NVLINK compatible, so multiple card configurations will not be possible.

Videocardz font