@Pedro: I can't put a number on the performance gains because I don't know the details of their next architecture. Nvidia does not always follow their roadmap. We were expecting Volta for their last consumer GPUs and then they came out of nowhere with Turing and their RTX line. So, we will have to wait and see.
Both Volta and Turing shares CUDA compute capability 7.x . https://docs.nvidia.com/cuda/turing-tuning-guide/index.html
The Turing Streaming Multiprocessor (SM) is based on the same major architecture (7.x) as Volta, and provides similar improvements over Pascal.
With RTX 2080 Ti, we have Volta+ CUDA 7.x compute capability without usable FP64 and HBM 2.
https://en.wikipedia.org/wiki/CUDA
Volta has CUDA Compute capability version 7.0 to 7.2
Turing has CUDA Compute capability version 7.5
Optimizations for Volta is applicable for Turing. There's only minor CUDA compute capability differences between 7.0/7.2/7.5 versions just as AMD's GFX 9.x Vega family.
In terms of features, Volta is similar to GTX Turing with extra Tensor cores . RTX Turing has RT cores and Tensor cores.
From programmer's POV, Turing is treated like a Volta.
Developing AI apps for NVIDIA's Volta CUDA CC v7.0/7.2 embedded platforms via RTX Turing equiped workstation PCs are valid. There's dev kit hardware inside a gaming PC.
Log in to comment