Archive notice: This article was originally published on December 27, 2018. Links and embedded videos are preserved as part of the historical record.
Built in are: 576 multi-precision Turing Tensor Cores, which are supposed to deliver 130 teraflops of “deep learning performance” – to what extent that is relevant in our workflwos remains to be seen. If everything works as intended, that should make real-time 8K workflows possible. ANd 72 Turing RT cores, which technically enable up to 11 GigaRays of ray tracing per second, will probably speed up quite a bit in the next round of renderer updates.
Key data
What is relevant is 24GB of GDDR6 RAM with 672GB/s bandwidth – roughly twice that of the previous generation. and while we are doubling things: With 100GB/s “Nvidia NVLink” you can connect two T-Rexes. The image of two thunder lizards in a Spartacus-style chariot race will then appear on the usual platforms fairly soon. There is also a “VirtualLink”, a connection for the nexxt generation of VR headsets. We will not say anything about that.
Target group?
The card is therefore intended primarily for AI applications and “deep learning” – doubling the throughput should allow neural networks to be trained considerably faster, and the shorter turnaround means much faster development.
Price and availability
The Titan RTX is supposedly coming at the end of December – so, realistically, in mid-January when everyone is back from holiday – and is expected to cost around 2,499 US dollars. Do you have anything left from your Christmas bonus?

