Some of those features are pretty cool, but, you don't really need proprietary hardware and drivers to do them. Phones can do a lot of those things. The part that really confuses me is why they're showing off the green screen. Background removal has been a thing for years, even in open-source software. If the green screen can work on a moving background, then I'll be very impressed.

schmidtbag:

Some of those features are pretty cool, but, you don't really need proprietary hardware and drivers to do them. Phones can do a lot of those things. The part that really confuses me is why they're showing off the green screen. Background removal has been a thing for years, even in open-source software. If the green screen can work on a moving background, then I'll be very impressed.

I don't know how good the Nvidia implementation is but have you ever used any of the automatic background removers? I've tried several and they are all really, really bad. Pretty much everyone on twitch continues to use a greenscreen and chromakeying.

Denial:

I don't know how good the Nvidia implementation is but have you ever used any of the automatic background removers? I've tried several and they are all really, really bad. Pretty much everyone on twitch continues to use a greenscreen and chromakeying.

I suppose live/realtime background removers might be pretty bad; I've never used it myself and I never watched anything on Twitch (nor do I intend to). But I've seen plenty of pre-recorded content that does background removal and works perfectly fine; no AI required. That being said, assuming Nvidia's "green screen" does yield better results than other realtime options with minimal effort, then sure, I suppose that's a nice feature.

Sounds like better use for the Tensor cores than a blur filter. I actually thought it's rather interesting gaming cards had the AI cores, it was just the price premium that pushed me off.

i wonder what are the specifics of an hardware tensor operation? what does it need to be faster at that a general purpose core does not do?

Very enticing features indeed. Pretty much like the latest smartphones today with their advanced video capture capabilities (for example Sony Xperia 1), these potential new features can make rather high quality content creation possible for lesser price. It is absolutely wonderful when technology gets cheaper like this, so more individuals have access to create and play with such things, possibly filming entertaining and inspiring short movies or whatever content that pleases us. From GPU market perspective, these features make RTX 2060 and the Super variant certainly more appealing for streamers. Little performance loss compared to RX 5700 is mitigated by possibilities, which means for AMD that they really need to drop their prices. I'm one who basically never watch streams, so I'm all for the Radeon solution, but even I am compelled to have those Tensor cores available if price difference is small between RX 5700 XT and RTX 2070 Super. I want inexpensive gaming hardware first, but all these neat things surely mean many will "go green" from this point onward if AMD doesn't act.

OK. So is nVidia turning Turing into MEME generator or something? Was not that AI stuff meant to improve general gaming? And what about performance if one uses it for streaming of game that uses same HW blocks?

Fox2232:

Was not that AI stuff meant to improve general gaming?

Nope , gaming is just a really small part. [youtube=GiZ7kyrwZGQ]

WhiteLightning:

Nope , gaming is just a really small part. [youtube=GiZ7kyrwZGQ]

I meant it for consumer RTX cards. 🙂

Fox2232:

I meant it for consumer RTX cards. 🙂

Well its multi-functional, so it helps in more then just gaming. more like system wide.

WhiteLightning:

Well its multi-functional, so it helps in more then just gaming. more like system wide.

But then there is that other part of my post. How big performance impact it will have on game that uses same GPU part for its calculations.

Fox2232:

But then there is that other part of my post. How big performance impact it will have on game that uses same GPU part for its calculations.

what makes you say its the same part ?

WhiteLightning:

what makes you say its the same part ?

What makes you think that it can't be same part? It can, and therefore sooner or later there will be collision of workloads. And something will take performance hit.

Fox2232:

What makes you think that it can't be same part? It can, and therefore sooner or later there will be collision of workloads. And something will take performance hit.

because it is done by the tensor cores.

WhiteLightning:

because it is done by the tensor cores.

I think his point is that if a game is using dlss or RT, which already uses the tensor cores and now your streaming that game with AI green screen, will it effect performance of the game? The answer is presumably yes.

Denial:

I think his point is that if a game is using dlss or RT, which already uses the tensor cores and now your streaming that game with AI green screen, will it effect performance of the game? The answer is presumably yes.

But doesnt RT use RT cores and not Turing cores ? dont know about dlss, that is true. i forgot about that tbh. And the number of Turing cores will also matter of course for performance fo the deep learning. here is the whitepaper btw, if you have nothing to do and havent read it yet. it is a very interesting read, exciting times ahead. (since this will be included in future cards as well) https://www.nvidia.com/content/dam/en-zz/Solutions/design-visualization/technologies/turing-architecture/NVIDIA-Turing-Architecture-Whitepaper.pdf