The fact that the ViT and CNN DLSS models have the same performance impact is evidence that the tensor cores on at least the high end chips (but probably mid-end too) are under-utilized, given that the ViT model was supposed to be 4x as intensive at inference. 

Did you see the article with the prices in some european countries (btw, the spanish site is called Wipod)? Yeah, I'm not even surprised by those prices, tho I admit Asus maeket dominance is getting the better of them.

So, about 30% faster with ∼30% more cores and power consumption for a theoretical 25% price increase. You can call that stagnation.

That doesn't look good for the 5080.

Let's hope they really uses those extra two months and, for once in God knows how long, AMD launches a GPU with proper drivers, game support for its features and in enough quantity.

But that seems to be asking for a lot.

Probably the best review I have seen of the 5090. Shows the 3090 to 4090 had a 74% uplift but 4090 to 5090 had 27%. But the power draw difference is nuts. The 4090 usually used the same power draw as the 3090 while the 5090 consistently uses 100 watts more. I'd watch the entire video cause he brings up some good points and it's only 9 minutes long.

The requirements also show that a RT capable gpu is required. I think next gen, we will be firmly in the "Ray Tracing" era where you will need a Ray Tracing capable GPU if you even want to launch the game let alone play at max settings.

RT is the future, no doubt about it, especially for the benefits it brings for Global Illumination. It's just a matter or how long does the GPU install base need to be for studios to make it a requirement. And so far, only two MSoft owned studios have decided to take that risk.

16 GB vRAM card, damn.

Hopefully I can get at least 1440p or maybe by the time I catch up to the Doom series, I'll have a better computer lol

The full PS6 (not cross-gen) era, I'm guessing. So maybe 2030?

Good video, one thing that surprise me is in this video and multiple others like the level one tech review for example for a single player game like cyber-punk multi frame generation seem to work really well to the point that if you have a 4k 240hz monitor it might be the best experience using Muti-frame generation on.

I plan to use mine for more then gaming like running AI LLMs and eventually training which is where the 5090 does have a big advantage over a 4090.  Either way am coming from a 3090 so I still see a huge increase in gaming + the huge increase in AI stuff.

My only concern with power draw is Tripping the circuit breaker in my office as it be a pain trying to get a electrician to run another line to my office as it on opposite side of house then my breaker box. Already running a 1600 watt PSU from that standpoint I got that covered.  

I wouldn't mind the power if the performance was a lot better but it feels like the 6090 is gonna get a node jump like Lovelace did and that is gonna have a big performance increase.

A half-node jump to N3, that is. The leap from Samsung's N8 to TSMC's N5 was much bigger.

Assuming Nvidia releases another ~600 mm² chip first like the 4090, that's a 17% transistor increase from the 5090 with either 35% less power consumption or 15% higher clocks (going by TSMC's estimates).

So it would be rather comparable to this generation's performance increase, barring some crazy IPC boost.