Not sure a video from 18 months and 30 driver sets ago is quite valid?
As said, hard to find many comparisons.
However, if you have a look at recent games, it seems to show a similar finding, which surprises me as I thought RDNA 2 was better than turing for RT but seems not, even a 3070 doesn't seem much better than the 2080ti in DL 2 here (which is unusual)
Are we still going on about RT in the first gen AMD ray tracing cards? oh lord. If anyone bought any of these for RT, in my opinion, they're an idiot. They are AT LEAST a generation behind Nvidia for that, and only worth buying for conventional graphics.
For raster they are awesome which is where they shine. Can we move on now?
Personally I always find that a poor excuse/reason, ok, it's amds first gen but are people going to pass up on that just because amd aren't there yet... Same way AMD customers are still waiting on a proper competitor to DLSS.... How long do people have to wait to get a similar experience to the competitor? This was one thing that did infuriate me when I had my amd gpus, recent ones being 290, vega 56, it always felt like you had to wait 1+ year to get something nvidia had/have or did better, 1 year is a long time in the pc gaming market, I'm at a point now where I want it now and not at some point "potentially" down the line....
But yes, I agree with the rest of your post although it kind of misses the point of the OP/thread.
I'm just curious why FSR 1 is providing not as much of a performance uplift compared to dlss in RT scenarios and it makes RT perf. of RDNA 2 even worse than we originally thought given amd can't use DLSS.
Probably methodologies , as FSR is performed after the rendering pipeline whilst DLSS isnt. I have emailed Ian Cutress to try and find out if AMD are going to use DirectML in the future.....
I guess when FSR 2 lands, it might be a bit better (using similar somewhat similar approach i.e. based on TAAU or TSR) but does seem that perhaps the tensor cores are helping here?
If you care about RT, and some people do, fair enough, its not a bad point to make, RT on RDNA2 is not much if any better than Turning cards.
But some people do it because its the only argument they have, they don't have rasterization performance, they certainly don't have performance per watt, so RT becomes vastly over stated in argument.
Nvidia know any metric that anyone can use to say "this is why Nvidia better" will keep their market share and sales up even if the GPU's cost a lot more than the competition.
So, with that in mind just sit back and watch the insane lengths Jenson will go to to say ahead with the next generation, its going to be a calamity of laughs, its going to be helarious, so just get the popcorn and enjoy it.
Interesting point that, no doubt for rasterization and stock, RDNA 2 is king, even when both are undervolted, I think RDNA 2 comes out slightly ahead, however, in RT scenarios, ampere demolishes RDNA 2 for performance per watt i.e. in matts video of his 6900xt compared to my 3080 in doom eternal (which is very light on RT), our fps is similar or mine is higher depending on the scene yet he's pulling 100+w more than my 3080.
2070S, overclocked to reference 2080S performance levels, RT is utter junk in almost everything, i only use it in FC6 because it only works at a reasonable performance in FC6, because apparently FC6 doesn't use proper RT
FC 6 RT is "ok", on the same level as guardians of the galaxy, resident evil village, doom eternal RT implementations. It's nothing to write home about when compared to the likes of cp2077, control, the ascent, metro ee, dl 2, chernobylite etc. RT.