Yes it can, lole giving every player $100 or even better $1000 or even better 10000 or even better 100000 or even better 10000000 or even better 100000000 or EVNE bettrm10000000000 or even better b100000000000000

This approach has caused a lot of bad press too. Notoriously "you guys don't have phones?" was said by a project lead at Blizzard and caused massive backlash. You can't just put anyone on a stage and have them captivate people and really sell your product. I still love the transparency and passion that is evident from presentations like this.

I imagine most of the people at the bleeding edge of computer graphics feel the same way as he does in regard to the sharing of information. They all come from heavy research backgrounds with PhDs in something related to computer graphics, they are used to the necessary sharing of information in the context of scientific research. These guarded proprietary solutions to general graphics problems have to be frustrating for them.

​@@GooseMugs Difference here is one is talking about graphical techniques for a free to play game that people actually like verses one trying to 'sell' a extremely shitty mobile cash grab.

@@GooseMugs that "do you guys not have phones" guy doesn't meet the passion citeria.

I don't get it. it's not true, we got a hundreds of Game dev talks everywhere sharing about game design / game architecture / rendering / network / physics / AI and such as GDC or SIGGRAPH in a more generic field. There is a lot of talk from all sort of companies from AAA to indie companies. But it takes time, cause you need to focus on the project first. And you have to prepare a big presentation after the game is released else you broke the Non-Disclosure Agreement.

Right ?! Damn, actually got goosebumps listening

Ok smart aleck

@@darkbibni he’s providing a lovely overview of performant solutions to common problems in realtime graphics, you don’t often get these techniques packaged together in a sampler that provides you with all the terminology you need to *actually* dig deeper if you want to.

Thanks, good sir.

thanks

Agreed. I think most people just kind of take this kind of presentation as something to skip over but you can tell he absolutely loves and knows what he is doing and it's wonderful to see.

It's extremely impressive. Lightning is one of the biggest costs in a video game outside of poly count, so being able to manage that cost is extremely important in any game these days. The technique expressed here (Hierarchical Radiance Cascades) is a fantastic solution.

Thanks!

no, let's GI

If you are the presenter, thank you for this mazing talk. I still have troubles understanding how increasing the amount of light sources (thousands of fireballs) doesn't end up being O(n*n) or maybe O(n logn) but you end up ... O(1) ?? I will need to rewatch it and try to grok it. What number os that constant-limit based on? pixels on the screen? cubes of space?

@@olivierdulac basically it calculates lighting from every pixel of the screen on every other pixel of the screen. Fireballs obviously still have rasterization cost, but the cost of calculating indirect lighting of the scene is fixed, because the total number of pixels does not change.

Hold on to ur paper

I have since the talk published a paper on it. Lots of people in the field have read it since then, and a bunch of them have already independently replicated my results by implementing the algorithm themselves.

I think the big O() is just wrong. He had just computed it "backwards", because the "n", the probe grid structure and render resolution, was already chosen to fixed value... It is like if you take binary search on 32 bit sytem. You have only 32 bit numbers for array indexing so O(log n) can be at worst case O(log 2^32) which is O(32) which is O(1) so we can conclude that binary search is constant time algorithm in practice.... I would say that the GI algorithm still grows O(n^2) with render resolution which grows at exactly same O(m*n) aka O(n^2) rate. Which is nice because 2*O(n^2) is still O(n^2) so the rendering complexity increase is multiplication by a constant which is all that really matters.. (and yes I was blown away too these presentations are cool)

@@ladislavseps4801 big O notation is an asymptotic behaviour in relation to some parameter at infinity. the paper discusses the way raymarching cost scales as a function of the total number of cascades, and not the screen resolution. in other words, O(1) absolutely does not mean "1", it means "asymptotic cost does not depend on this parameter". and just to clarify, the number of rays does absolutely depend on the screen resolution in the screenspace version, and grows as O(w^2) as a function of viewport width. note that this O(w^2) includes "each pixel lighting each pixel" behaviour that normally takes O(w^4) rays to calculate naively (O(w*w) pixels affecting O(w*w) other pixels). I just assume O(w^2) as a baseline that I compare against, assuming this cost is unavoidable in any image-space algorithm. in other words you can think of it as per-pixel cost.

I think lumen is way different because it needs to use light from outside the screen while poe don't, but maybe they have figured out a similar tech

@@xxseesxx9234 Alexanders mentioned that his algorithm can support world space AO and he showed an example of it.

At 37:24 he also mentions this works in world space for lights outside of the screen@@xxseesxx9234

@@xxseesxx9234 i believe he said in the talk, poe uses screenspace but he made a demo that is world space so it should be roughly the same there unless im missing something

@@emmas1366since then I wrote a paper with detailed analysis, and also a slightly different approach that captures world space radiance (including offscreen light sources and occluders) in screenspace probes.