If I recall correctly; they already do this by really limiting mixed precision performance on the GeForce line. I believe Titans typically where excluded, but otherwise for full FP16 support you needed a Quadro or Tesla card.

Is that limitation in hardware or software? I always thought it was a physical difference in chip design.

Yes, that is worrying. But, they already have license in place to ensure no one uses their gaming GPUs in the cloud, where the big money is already. (And it worked, I cannot find any low-cost cloud providers with gaming GPUs for GPGPU computing).

I hope they don't, given that there is limited upside and a lot of push back from dev community.

So let’s wait until BTC hits $100k and a generic videocard for any purpose costs $5000. That would be much better than paying extra $200 over MSRP for ML. /s

People are not mining Bitcoin on videocards.

But BTC price influences all crypto. Read it as “ETH hits $5k” if that matters.

That would suck in the short term, but it would also be a HUGE incentive to massively ramp up chip fabrication capacity, leading to long term cost decrease due to economies of scale.

What ML card are you buying that is only $200 over its gaming equivalent?

I don’t, that was a blind guess that missed. Insert a real extra and compare to the prospect of crypto prices rising and nvidia doing nothing. What would ML people gain from that while gamers lose all?

Yeah, the ML example doesn't really work because we're already paying through the nose for the privilege of running GPU's in production processes. This development is really nothing new, Nvidia and ATI (and Intel and AMD) have always charged extra for their "professional" lineups usually enforced through software alone or even just the EULA.