Personal computing discussed
Moderators: renee, morphine, SecretSquirrel
Kretschmer wrote:I'd wait for the final hardware and software and sincerely doubt that Nvidia would bet the farm on a feature that hobbles their flagship GPU to 30FPS at 1080p.
If anyone wonders why these companies are so frugal with unfettered access to their products before release, this is why.
Krogoth wrote:It is the Geforce 3 all over again expect the entrance price was a mere $349-399 instead being north of $599.
Krogoth wrote:It is the Geforce 3 all over again expect the entrance price was a mere $349-399 instead being north of $599.
christos_thski wrote:Krogoth wrote:It is the Geforce 3 all over again expect the entrance price was a mere $349-399 instead being north of $599.
May be the GeforceFX...
DancinJack wrote:You guys have seen ONE "benchmark" of pre-production hardware on a pre-production software patch and you're calling it a failure already. Good stuff.
DancinJack wrote:You guys have seen ONE "benchmark" of pre-production hardware on a pre-production software patch and you're calling it a failure already. Good stuff.
DancinJack wrote:You guys have seen ONE "benchmark" of pre-production hardware on a pre-production software patch and you're calling it a failure already. Good stuff.
Krogoth wrote:It is the Geforce 3 all over again expect the entrance price was a mere $349-399 instead being north of $599.
synthtel2 wrote:There's presumably a lot of room left for them to work out performance issues, but I wouldn't be surprised if 1080p30 at max settings is an accurate characterization of it either.
Everything shadow-related I see there can be done by shadow maps, it's just kind of expensive (probably much cheaper than 1080p30). The trouble is that it's very complex to implement compared to RT shadows, enough so that almost nobody does given how few systems will be able to run it anyway.
Shadow mapping can avoid considering every object every frame by doubling up on maps, keeping one for static objects and adding the moving objects to it every frame. They're probably doing something similar here, either keeping separate BVHs for static and moving objects or just keeping a persistent-by-default representation of the world and updating whatever's changed each frame. Incremental updates to BVHs are a bit of a PITA, but Nvidia doesn't shy away from that sort of thing.
dragontamer5788 wrote:A shadow map definitely emulates the effect. But I'm not convinced it is the same.
[...]
derFunkenstein wrote:Maybe I'm in the minority, but to me the various demos of lighting effects are so convincing I'd probably play the game at 720p and sit back from my desk if I could do it.
synthtel2 wrote:dragontamer5788 wrote:A shadow map definitely emulates the effect. But I'm not convinced it is the same.
[...]
Shadow maps take some extensions to do that, but they can. PCSS (from way back in 2005) is a good example. I like the idea of SAVSMs to go with it (but haven't tried any such thing). There may still be a lot more complexity past that to get performance up, since the kind of quality on display in that demo needs pretty near full geometry detail and near 1:1 shadow map resolution and would really rather not be rendering unnecessary polys/texels. Either way gets very slow very quick when multiple lights get involved. RT shadows probably have a sizable advantage with multiple lights, but that demo doesn't look to be doing much with that.
GTA V comes to mind as a game that includes this (two implementations of it, actually).
synthtel2 wrote:derFunkenstein wrote:Maybe I'm in the minority, but to me the various demos of lighting effects are so convincing I'd probably play the game at 720p and sit back from my desk if I could do it.
That environment mapping stuff is where raytracing is exactly the holy grail it's cracked up to be, and it looks like Atomic Heart is relying on it. We can emulate all of that without raytracing, but it's a pretty shoddy emulation and it takes both a lot of programmer time and a lot of compute time to get even that. At least with the non-RT shadows you can get a seriously high-quality result for your effort.
derFunkenstein wrote:The thing I wonder most about this is whether resolution affects performance of the RTX effects. It's got to trace the rays of light regardless of resolution to fill in the picture. If you don't trace enough rays, presumably you get a grainy picture. Trace too many rays and performance tanks without improving the image. Maybe there's a "sweet spot" that performs and gives great image quality at 720p vs 1080p? Or 1080 vs 1440, or whatever? I have zero idea if that's even realistic, so I'm open to being corrected.
dragontamer5788 wrote:NVidia did 1-sample per pixel + a hell-of-a-lot of denoising in their demos (some kind of spacial-temporal effect, smoothing over space AND time to keep animated images looking nice). The "Star Wars" Demo was 1-sample per pixel + denoising IIRC.
"Typical" CG animations are 100+ samples per pixel, rumored to go up to 5000+ in the case of like... Davy Jones (Pirates of the Caribbean). I have no idea what settings were used for Grand Moff Tarkin (Star Wars: Rogue One), but I'd presume in the 5000+ samples per pixel zone or maybe even a few magnitudes more.
The general technique NVidia is hoping for is that their spacial-temporal denoiser magic sauce can erase the graininess. And as far as I can tell, it does a really good job at that. So props to NVidia. It will only work for blurry images however, not perfect mirrors or whatnot. But still, its progress. You simply need more than just 1-sample-per-pixel to achieve the photorealistic effects from Hollywood.
dragontamer5788 wrote:Reflections / Refractions are one of those other things that require raytracing. Although I'm not entirely sure if it really "adds" to the scene aside from realism. The thing is, without studying caustics / reflections / refractions, its very difficult to know what they are supposed to look like. So most laypeople probably won't tell the difference. Its not like people typically look at the shadows cast by semi-transparent objects with strange index-of-refractions all day. Water in a glass? Maybe, but how many people can tell where the light is supposed to go even in this simple case?
Captain Ned wrote:As one of the altekackers here, I still remember when GFX reviews set 30 FPS as the "threshold of realness".
dragontamer5788 wrote:NVidia did 1-sample per pixel + a hell-of-a-lot of denoising in their demos (some kind of spacial-temporal effect, smoothing over space AND time to keep animated images looking nice). The "Star Wars" Demo was 1-sample per pixel + denoising IIRC.
DancinJack wrote:You guys have seen ONE "benchmark" of pre-production hardware on a pre-production software patch and you're calling it a failure already. Good stuff.
Captain Ned wrote:As one of the altekackers here, I still remember when GFX reviews set 30 FPS as the "threshold of realness".