For the scenes that they’re showing, 76ms is an eternity. Granted, it will get (a lot) faster but this being better than traditional rendering is a way off yet.
Yeah, and the big caveat with this approach is that it scales quadratically with scene complexity, as opposed to the usual methods which are logarithmic. Their examples only have 4096 triangles at most for that reason. It's a cool potential direction for future research but there's a long way to go before it can wrangle real production scenes with hundreds of millions of triangles.
I'd sooner expect them to use this to 'feed' a larger neural path tracing engine where you can get away with 1 sample every x frames. Those already do a pretty great job of generating great looking images from what seems like noise.
I don't think this conventional similarity matrix in the paper is all that important to them
Because none of what they call “tariffs imposed on the US” are tariffs. It’s the trade deficit divided by the country’s exports to the US. They’re literally making shit up and expecting stupid people to take it at face value.
They’re imposing tariffs on uninhabited islands. They don’t care enough to even try and make it plausible. That’s how stupid they think their supporters are and how much contempt they have for them.
I used to believe this vehemently. It has become clear that that’s a notion from a bygone age.
The internet has created a global town square where the loudest voices are the ones that catch people’s attention, regardless of the veracity of their claims. There is no truth any more, only the cult of personality.
Tomorrow the US installs a racist, rapist, treasonous kleptocrat as president because the majority of people are unable to think objectively and swallow his promises at face value, despite every indication that life will be immeasurably worse if you’re not a billionaire oligarch.
What’s hilarious is there’s nothing physically based about the Disney model. It’s empirical and It’s not even energy conserving.
As sibling pointed out, physically based rendering precedes “PBR” by a looong time and goes much, much deeper than “I put a metalness map in my shader”
The upper bound of the approximation error of random Monte Carlo methods decreases with O(1/sqrt(N)), whereas the error for quasirandom Monte Carlo methods (omitting some details) decreases with O(1/N) (where N is the number of samples) [0]. I don't think path tracing is an exception here.
I would even argue the opposite: For the same variance I would expect the error to be more perceptible for (blue noise) quasirandom MC, because it can lead to regular patterns in the noise.
CUDA IS the merit. They’ve been developing the software stack to make GPU programming accessible for 2 decades now. They’re a software company as much or more than they are a hardware company. Ignoring this fundamentally misunderstands why they’re in the position they’re in now.
It's not the merit - it's the moat (i.e. lock-in) as the linked article states. Merit in this context would mean something you can compare across different GPUs. For CUDA - you can't. I.e. it's a tool to force you to use Nvidia.
