The Titan RTX is aimed at AI researchers, data scientists, and content creators, with NVIDIA claiming "breakthrough performance" from FP32, FP16, INT8 and INT4 operations that allows for faster training and inference of neural networks. Researchers will also be able to leverage NVLink to combine two Titan RTXs when working with large data sets. Of particular interest for data scientists will be the fact that the Titan RTX supports RAPIDS, the open-source libraries that integrate with popular data science workflows to boost machine learning use cases.
I have a big bang house. I had all the materials for my new house brought in and dumped on a vacant lot. I took a stick of dynamite, lit the fuse on fire and tossed it into the pile of materials. BOOM went the explosion and when the dust settled I had a new house! Amazing isn't it?
So will the 7nm cards in 2019 supersede these GTX 20 series and Titan or do you think 7nm will just be a turing refresh
It should be similar to the transition from Maxwell to Pascal. re-using same micro-arch except for faster speed on a smaller node. However NV will never name the micro-arch with the same name, so expect Turing refresh to be named something else (Ampere?)
They added dedicated hardware for Tensor with the Titan V and increased memory. Of course pricing will be much higher than the previous gaming Titan.
Let's not forget the the Titan V and RTX Titansupport NVlink.
No one looks at the die size comparison when considering what products should cost yet its a main contributor. There will be nothing but complaining about product costs for cards with tensor cores until they are used by nearly all games.