alanm:

I'm seeing some disturbing news re Nvidias new NDAs. https://videocardz.com/76645/nvidias-new-non-disclosure-agreement-leaked German site Heise.de has refused to sign it as its a long term (5 years) contract which could limit even speculation about technical aspects of Nvidia products unless its "beneficial" to Nvidia. https://www.heise.de/newsticker/meldung/In-eigener-Sache-Nvidia-NDA-als-Maulkorb-fuer-Journalisten-4091751.html I wonder why now? Could the timing also be related to how Turing may turn out? That it may not live up to expectations?

I wouldn't say that has to give us any specific clues to the chip itself, but more to how Nvidia tries to control the environment around them. Same as with GPP, now they're just doing it with tech journalists and not board partners. But this time around there's actually a paper to read so everybody knows what's up, not just hearsay. I wish back with GPP such a document would have been leaked too. Good find though, thx for sharing.

Michal Turlik 21:

I understand the enthusiast point of view but I am also sure that at the Nvidia headquarters they try to focus on the 250 watt tdp limit for their gaming cards so how much power the new theoretical 12nm architecture may need to be capable of 120 hz at 4k? At this moment it seems that only the Titan V is capable to do what you would like to see in the mainstream segment.

I think you missed the point of "New Architecture", and ignored it for 12nm. 12nm may not do much, likely same achievable clock at 100mV less, so around ~10-15% smaller power consumption, or bit better OC. But point of Turing is in changes inside building blocks of GPU itself. Some things may be done within fewer GPU cycles. It may have those new Tensor cores. Maybe better intra frame voltage gating for parts which are done their work and are waiting for next batch of data. (which may bring even bigger energy efficiency over 12nm) There are really many things next architectural step can bring. What if nVidia improved HW level compression for textures and next generation had effectively 20% more space for textures on card with same amount of physical memory?