Недавние запросы
Параметры поиска
Управляется:
@drq@mastodon.mlDear game graphics developers. Stop making frame upscaling and frame generation technology into a necessity.
Seriously, stop.
I mean, it's definitely a nice-to-have, and within its own limitations, it can be a good addition to the graphics pipeline, but it definitely isn't a replacement for graphics optimization.
Frame generation can help you slightly increase perceived smoothness - which is awesome because it's not a tragedy anymore if a game drops a frame or two, but in order for it to work properly, there has to already be a lot of frames.
Frame upscaling can help you slightly increase total resolution of a frame - which is awesome, because I can take a peek at new demanding tech like raytracing without buying a NASA computer, but in order to work properly, there has to already be a lot of the frame itself.
See how those algorithms are self-limiting? It's by design. Stop hiding bad graphics behind them.
We've already been there with bloom, depth of field and motion blur.
Speaking of upscaling. I'm hardly a graphics expert, but what if we apply the upscaler not to the frame, but to a texture - namely, a lightmap?
Raytrace at low resolution relatively cheaply, give the resulting lightmap to the upscaler, and tell it to work its magic, et voila, we get a decently raytraced lightmap for peanuts compared to full-scale raytracing.
You can even fill in some gaps with frame generation when the lightmap changes, lowering the frequency at which you have to fully raytrace.
@drq you're probably preaching to the choir here.
Competent graphics engineers are expensive apparently, players don't seem to care much, so management doesn't see this as a bad idea: picture pretty, sales go brrrrr.
I think the most workable angle today is to improve game competition by strengthening player communities when it comes to offering alternatives. Consumer associations of some kind, I'm not sure how that would work these days. That would work against other player-hostile practices too, like vendor-locked game servers. Only that there doesn't seem to be an actor with a financial incentive to do that, so probably not gonna happen anytime soon, unless someone from the 0.01% suddenly decides to pick up such a hobby.
@th3rdsergeevich well, kinda. Steam enforces *some* of the functions of a consumer association. But, at its core, it isn't one. And the difference is showing e. g. with the recent CS2 casino scandal which highlighted a source of Valve's income that is blatant psychological+financial exploitation – something I'd expect an actual consumer association to be able to boycott out of existence.
Still, it's a good head start, it offers a good bunch of living breathing examples of useful and reasonable consumer demands. Now if the same demands could be spearheaded by consumer associations with similar collective market power as Steam, we'd be good.
@lispi314 yup. Hence the idea to boost the players' signal. Advertise the idea that learning about the common pitfalls of videogames before picking them up might just be a good idea (better bang for the buck essentially). Not just in videogames preferably, but there too.
Because as it stands today, a common entry point is commercial advertising, and guess who defines that. 
I can only speculate on Gaben's actual reasons for banning ads on Steam, but I'd expect them to be along similar lines.
@drq@mastodon.ml my evil idea for game graphics: run the game loop render everything at ~1000FPS at low res, and then do the whole optical flow and temporal super sampling stuff but in reverse. You'd get proper (not fake!) motion blur and stuff with this approach, but I don't think current operating systems can truly keep up with such update rate.
See also, unreal engine 1
@lebronjames75@shitposter.world @drq@mastodon.ml oh
maybe viable without changing os that much then
running windows 10
ut2004, on smaller maps ~~800FPS
on big open-world maps ~~250FPS
unreal 1 (modern ver227 OpenGL patch, older ones without OpenGL should run faster but uglier) runs at ~~700FPS in most spots, main menu for unreal 1 runs at 2000FPS, simple geometry areas runs at around 1200FPS. Complex areas at 500-800 FPS. modern 227 patch of unreal 1 prevents the game from running slightly speedy at 1000 FPS.
@ignaloidas @drq xash3d can achieve up to 6000fps even on vega11 igpu...
@drq upscaler+framegen as replacement for graphics pipeline 
@drq dear game developers. Stop making 3D technology into necessity.
... We had pretty much the same problem back in the days, but turns out you can sell anything if you shill hard enough
@drq only takes care of diffuse. but RT enthusiast love reflections
@lritter Hmmmmm... Does it, though? Isn't what we're doing here basically render reflection at low res and upscaling it, so speculars get the treatment as well? Because in raytracing, everything is a lightmap, basically.
IMO, needs research.
But even if - raytraced diffuse is always nicer than traditional ad-hoc tricks.
@lritter I mean, I'm okay with reflections being a *little bit* lower fidelity than the rest of the frame. Especially on a curved surface. I doubt that anyone would really ever notice that. Bonus points, if the surface is a little matted - then you can blur and lose detail to Hell and back.
Straight mirrors are actually pretty rare in nature anyway. That's why it was considered unlucky to break one back in the day - it was very not easy to make, so the superstition stuck.
@drq @lritter sad thing is: modern cards can't properly trace rays at native resolution so most games already do this. We already have blurry mess instead of reflections.
Also, modern RT is noisy as hell and to mitigate it we either need to up a sample per pixel count (but in order to do it in 60 fps we need monster of a video card) or blur and denoise result heavily to get decent image.
Real-time RT still, mostly, a hot garbage that just eats up all of you resources to produce something that is barely distinguishable from baked in lighting in most games. (Only if you pay close attention to nuances). But it allows developers on some engines to do virtually nothing and get okayish performance and visuals.
Honestly, I don't know what RT did in new Indiana Jones game — it looked exactly as if I didn't turn RT on, but my PC did much harder job to get this result. Why? 
@dotterian @drq i cannot agree with this. i enjoyed the path tracing in cyberpunk 2077 immensely. and it is true, the technique is still too costly to provide a clean image without some trickery.
what you call NASA level performance, they already called NASA level performance 20 years ago. computers get faster through innovative breakthroughs, and things thought unreachable before suddenly become commonplace.
for path tracing, this is that avenue. start out shit, become better in 2 decades.
@dotterian @drq though i recall when the 320x200 VGA era ended and the first 640x480 games came out i was like "why? this adds nothing" -- so if we're perfectly honest, that's where we should have stopped.
@lritter @drq I didn't say it won't be better eventually. And yeah, PT is way better than RT, but you still need good card (80-90 models of Nvidia and virtually none of AMD will do) to enjoy it properly.
What I was saying is: it's bad now and really not worth it. And, sadly, it's been 5 years since RTX 2000 series and we're still in a place where proper RT without some crappy tricks are virtually inaccessible for most players. I'm still trying not to loose hope, but future of RT for me looks dull for next 10 years.
@dotterian @drq yes, not digging the proprietary development status and the exclusivity of it (and don't get me started on CUDA as The Other Big Problem Child), but please, let's differentiate and don't shit on the technique for the failures of the market.
@lritter @drq problem is: RT as a "software technique to get better image" is inseparable from hardware which is, still, in it's infancy. And if we're talking about optimizing RT here, we have to take hardware and current objective reality into account.
Sadly, objective reality of current hardware is less then good for awesome RT without generating frames, upscaling and adding whole bunch of temporal artifacts to image.
And "let's keep image clean and optimize properly" was a point of first post in this thread, iirc.
@dotterian @drq and if one day, images are clean and everything is optimized properly, it will only be thanks to you. :)
(but situation presently is not as bleak. there are many ways to get GI apart from HW accel RT, one of which hasn't even been exploited to the fullest, and which i'm using, because it gives me infinite bounces.)
@lritter Care to elaborate?
@drq @dotterian the high quality one is Voxel Cone Tracing (Crassin et al). around the time when the paper came out it was a little too much but by now several games have implemented variants of it, The Tomorrow Children is one i can think of right away.
the one I'm fancying is called Light Propagation Volume Cascades, which started as a Crytek paper, and was used in Crysis only to do bounce light. I implemented it as a prototype and quickly found that there were errors in the paper. (cont)
@drq @dotterian then i discovered someone masters thesis using the same technique, who had found the same errors and figured out the correct coefficients to fix light loss, and found that the technique now worked without requiring any lightmap injections. due to its low coefficient SH it's quite blurry and lowres, but it fits the style of our game very well.
media molecule's Dreams uses it as its GI solution.
I believe lightmaps are a feature of pre-baked lighting, something that that raytracing doesn't use at all. Raytracing's way is converting all world's geometry into octree or some other such structure and bouncing an astronomic amount of rays on it. It's basically a "Monte Carlo" method, that is it works by doing lots of random sampling. And the total number of rays used can be changed for different tradeoffs between quality and performance. But I bet in existing games they already tuned it so it's using minimum amount of rays necessary to avoid artifacts.