I think this is a little wide of the mark.
While Stadia is a different form factor, business model and value proposition, I think it's squarely aimed at next generation specifications that we're expecting for PS5 and the neXtBox.
If Stadia performance over/under shoots the next gen home console specs by a significant margin the ports will be more troublesome and come off less well to consumers damaging its position in the market. Google know they'll be depending timely ports to Stadia because their internal teams will take time to ramp up and may not even deliver anything that sets the world on fire.
I'm certain all the platform holders broadly know what their competitors are planning because the machines are becoming more similar with every generation and share common parts. I think it would be naive to assume that Google just plucked the Stadia specs out of thin air without an eye on what Sony and MS were/are up to. In much the same way that Sony mentioned game streaming platforms as a risk to the PlayStation business in their last results. Now, why would they do that I wonder?
I do believe that game streaming is (sadly) the future. And sadly it will be the future for all the wrong reasons. It will be the future because the people who make the call are also the ones who are not really gamer. It is the suits who see how successful Netflix and Spotify and all the other streaming things are. They do not understand that gaming is different. It will be a sad future for me because I am hyper sensitive towards additional latency.
I agree. Stadia is definitely aimed at taking share away from Microsoft and Sony (and I guess it is safe to expect next gen consoles to be roughly in the ballpark of 10.7 TFLOPs) and they all want to take a big piece of pie from the "2 billion gamers worldwide" market. (Which is a bit of a stupid goal in my opinion because even though 2 billion people have access to some sort of internet or device that could stream it to, they don't have the internet to handle that, they don't have the money, and most assured no company has the infastructure to target every single one of those people).
My comment was more aimed at people basing expectations for next gen consoles on the google streaming thing like in a napkin math scenario á la google has 10.7 Tflops in 2019 -> that means Sony will have 14.2 TFLOPs in 2020 and Phil Spencer said Microsoft is going to have the most powerful console -> so Xbox Big Scarlett is 16.8 TFLOPs.
There might be a case to be made about saying Sony and Microsoft want to go above 10.7 Tflops for their next gen consoles for marketing reasons so they can stick one up to google again.
I think that I would consider that expectation to be reasonable. 60CUs @ higher than 1400MHZ and you have around 11 TFLOPs which is probably a very realistic expectation for a late 2020 console around $399-499.
i just want to say that i dont disagree with anything you have said here, and i do gree that we will have to wait to see the power draw of those cards, but we do have some real life examples of what happens when you take these AMD cards and put them on a smaller die. We saw this with both thebase consoles and the mid gen upgrades X1X and PS4 Pro. they went from 28 nm to 16 nm and were able to the CU count AND increase the clocks. In MS's case, with an extra $100 budget over the Pro, they were able to push the clocks by by over 250 Ghz with some fancy cooling.
Going from 16nm to 7nm should allow them to double the CU counts unless of course they hit the GCN 64 CU limit. but even in that case, the jump down to 7nm should let them push the clocks even higher.
i think 14 tflops is very possible if you just go by the precedence we have here and if Sony is willing to hit the $499 mark. I agree that 10 tflops for $399 is a much more realistic but MS should be able to hit 14+ tflops with their $499 anaconda model.
Oh yes my clockspeed estimation is very high. I think 1400MHZ-1500MHZ for the Navi GPU inside the PS5/XboxScarlett is going to be realistic, but I think that the amount of compute units they can fit into the die space they feel comfortable is lower than people think.
My estimation (as of right now) is still 56 CUs @ 1400MHZ for exactly 10 TFLOPs. (Those are my lowest estimations for a 399$ PS5 console in Q3-Q4 2020).
And I am not sure how expensive a 7nm die with 72CUs would be (if we go by the double amount of PS4pro). I can understand the logic behind it going from PS4 Pro has 36CUs at 16nm so 72CUs at 7nm for PS5 is possible (I haven't done the math and I don't know if it exactly lines up like that especially considering early 7nm yields and binning) plus the additional core clock increase we inevitably get from going to a smaller manufacturing process.
So yeah 72CUs @ 1500MHZ would give us 13.8TFLOPs, but if that is possible for a console in terms of cost per unit / power draw / die size I can't say. I feel like it would be too expensive, too power hungry and too big for a console.
With all that being said I would love to be proven wrong and being surprised by next gen, but in all fairness there are also other things that are going to cost money which I think are more important than raw TFLOPs. Like you mentioned better cooling would be very nice to have considering it is going to allow for higher clockspeeds but also (if done properly) it makes your console quieter, which I do have a problem with in my PS4 being too loud imho.
And then we obviously need more and faster RAM, a better build quality Controller, and an SSD would be a substantial upgrade for the next generation of consoles. In my opinion SSD prices in 2020 should come down so far that going with an SSD is the only logical option for both Sony and Microsoft.