If this is your first visit, be sure to check out the FAQ by clicking the link above. You may have to register before you can post: click the register link above to proceed. To start viewing messages, select the forum that you want to visit from the selection below.
![]() |
|
|
#1 |
|
Junior Member
Join Date: Feb 2007
Posts: 96
|
I mean, it should be possible provided scene complexity is modest enough, right? I doubt it'll happen with UE4 given the engine architecture, but it'd be nice to play around with their lighting on a GTX 260 knowing I couldn't exceed a certain poly count and texture resolution. I can accept that tradeoff with the understanding that even a 10 year old game could look dramatically better with global illumination.
I suppose Crytek have made inroads toward this goal, but unless Crysis 3 is a miracle on consoles, they're going for the big game now the same as epic. And soon enough cards will be cheaper and GI will be standard. But until then I want to see real-time lighting at the expense of everything else and I'd like to see that happen in the mid-range, if anyone's willing. |
|
|
|
|
|
#2 | |
|
Junior Member
Join Date: Dec 2006
Posts: 28
|
Quote:
Doom 3 certainly could have benefited from a decent GI implementation, FEAR too. All those early shadow volume games where anything that wasn't in light was pitch black. |
|
|
|
|
|
|
#3 |
|
a.k.a. Ingenu
Join Date: Feb 2002
Location: Apsley, U.K.
Posts: 2,738
|
There are some but always (unless mistaken) partly precomputed.
__________________
So many things to do, and yet so little time to spend... |
|
|
|
|
|
#4 |
|
Member
Join Date: Aug 2005
Posts: 278
|
DX11 hardware is cheap as chips these days, are you still using a GTX260 on your main machine?! I really doubt there will be much work devoted to DX9/10 going forward.
Last edited by CNCAddict; 16-Jun-2012 at 00:19. |
|
|
|
|
|
#5 |
|
Entirely Suboptimal
Join Date: Mar 2003
Location: WI, USA
Posts: 6,847
|
I wonder how much life DX11 has left, for that matter. The current hardware is 11.1. DX12 is probably coming sooner or later (next consoles?)
|
|
|
|
|
|
#6 |
|
Senior Member
Join Date: Feb 2002
Posts: 2,020
|
The AMD Ping Pong demo implemented real time GI on DX10 hardware and the same basic technique of using cubemaps applies to DX9 hardware as well.
http://userpages.umbc.edu/~jshopf1/Projects_AMD.html I don't know of any games that used this technique. |
|
|
|
|
|
#7 |
|
Member
Join Date: Feb 2007
Location: Irvine, CA
Posts: 426
|
Crytek's LPV technique could be done on DX9 or DX10-class hardware. Michael Bunnel's point-based GI was originally implemented on DX9-class hardware using pixel shaders. The Geomerics middleware does a lot of its work on the CPU and can work with DX9 hardware, and you could probably whip up a similar solution that does it all on the GPU.
|
|
|
|
|
|
#8 |
|
Junior Member
Join Date: Feb 2007
Posts: 96
|
do any geomerics solutions operate completely in real-time or do they all require at least some geometry preprocess? i mean, i guess with sufficient alteration even that preprocess could be done in real-time on lower-end hardware, but with less precision
i'm talking 2/3/4 teraflop cards. and yeah, still rocking the gtx 260 unfortunately. don't have the space or the cooling necessary for a modern card. need a total system overhaul at this point |
|
|
|
|
|
#9 | |
|
Darlek ******
Join Date: Jun 2004
Posts: 9,498
|
dx11 not 9, but found this quote from a review of dirt showdown
Quote:
__________________
Guardian of the Most holy Two Terabytes of Gaming Goodness™ |
|
|
|
|
|
|
#10 |
|
a.k.a. Ingenu
Join Date: Feb 2002
Location: Apsley, U.K.
Posts: 2,738
|
For the record, I was commenting on D3D9/D3D10 class hardware.
D3D11 class hardware is much more capable. When it comes to GI though, nothing beats raytracing ;p (bi-directional path tracing generates great images)
__________________
So many things to do, and yet so little time to spend... |
|
|
|
|
|
#11 |
|
Junior Member
Join Date: Feb 2007
Posts: 96
|
I'd love to see someone take something like Ico, or a similarly beautiful but performance-lite game these days, and run it through some path tracing. It kinda blows my mind that's it never happened. The closest things we get are like Quake 3 raytraced, but only as proof of concepts.
|
|
|
|
|
|
#12 |
|
Member
Join Date: Feb 2007
Location: Irvine, CA
Posts: 426
|
I'm pretty sure that they always pre-compute the form factors between patches. Computing form factors can be cheap if you use approximations, but computing visibility is expensive. Bunnell uses a clever method for approximating visibility, but it's not perfect. Without visibility it also takes you longer to evaluate the lighting since you end up with many more non-zero form factors between patches.
|
|
|
|
|
|
#13 |
|
Junior Member
Join Date: Feb 2007
Posts: 96
|
I've been playing the steam build of Half-Life (not source) and I've been thinking: how many light bounces did goldsrc bake into maps by default? Two? I feel like we should be able to do Quake-era/Quake quality light bounces in real-time on modest hardware. It shouldn't matter how low-res those shadows would be provided you use lights effectively in any given scene and you maybe separately render self-shadows for characters and particular objects.
Half-Life running on final goldsrc builds may not hold up to modern eyes, but it is still a triumph of lighting design. Half-Life 2 even more so. Both games make such effective use of contrast - positioning lights minimally to create great and subtle textural shifts, limiting area of effect, combining natural and artificial light - and all I ever want from a real-time GI solution is to be able to handle those transitions without my input, so I'm free to position lights as effectively and let the engine do the rest. This should've been possible years ago. I can deal with low-res shadows and less than accurate light calculations as long as there's still some simulation there - as long as it means I still get the right variety in lighting conditions, instead of the nasty wall of ambient occlusion you see so often in even Cryengine games. I have some naive ambitions of doing something like this with Unity, but I don't know that I'm the right programmer to do it. Maybe UDK2 will make that all irrelevant. edit: I mean, 16-bit shadows with color banding in their falloff would be preferable to the flat shaded mess of Cryengine interiors Last edited by inlimbo; 10-Aug-2012 at 00:15. |
|
|
|
|
|
#14 |
|
Senior Member
Join Date: Mar 2007
Posts: 1,147
|
What do you think of the indirect shadows in 'The Last of Us'? Is that like what you're describing in your post?
|
|
|
|
|
|
#15 | |
|
hardly a Senior Member
Join Date: Jul 2008
Location: still camping with a mauler
Posts: 3,637
|
Remember folks, always preview and review your posts. It doesn't take too much time really.
inlimbo, I don't think the tradoffs you describe would go over very well with most people. HL1 quality lighting would not be acceptable even if it were all done in realtime IMO.
__________________
Quote:
|
|
|
|
|
|
|
#16 |
|
Junior Member
Join Date: Feb 2007
Posts: 96
|
I think that reaction owes more to the cynical gaze of the modern gamer than the reality. The mind hears "low-res" anything and it reels from the very idea. And, yeah, it's hard to separate Half-Life's antiquated visual qualities - and even its lesser successes functionally and artistically - from its greater success at building and following through with distinct, palpable atmosphere, but that's what I'm asking here.
Allow me to use this as an example: ![]() there's banding there, but you just don't see that mix of light and shadow in many real-time lighting solutions, yet I think it's crucial for delivering atmosphere without smothering your scene in color filters and high contrast monochrome or thousands of deferred light sources. Of course the latter are sometimes more a style choice than a technical limitation, but compare that 14 year old Half-Life scene to this: ![]() there's just nothing there. Might as well not be lit in my mind. It doesn't matter to me that it's showcasing a decent polycount, clean texturing and pixel-perfect specular; it doesn't feel like a real location. Even if you knock out at a few lights in favor of drama, you're stuck with large swaths of single-toned bright and dark where there should be more subtlety. I suppose you can mitigate that in CE3 by hand-placing ambient lights (if that's possible). And maybe there's something to be said for Crytek's light propagation system when given more room to breathe instead of strangled thin by their tech push, but that's just the point I'm trying to make. You pair a relatively modest (in the quake sense), but flexible real-time path tracing lighting system with modern tech and the right artists (artists who aren't strictly pushing for bleeding edge results) and you'll find some great middle ground. You're not gonna end up with a 1998-looking game. Make no mistake, Crytek's engines are beyond impressive. I just wish they had the sense to juggle their ambitions more effectively. That's what I'm arguing for here. I think you can arrive at similar, and in some cases better, results under limited circumstances - if you've got the basic simulation and the right finesse. edit: I do like what Naughty Dog's doing with indirect shadows. I think it's proof that if you render falloff the right way, you can get away with low-res shadows that don't feel like a jagged mess slathered across the screen. Penumbra is essential for that. I wish it was somehow possible they were using no prebaked lighting for their environments too, but ah well. That's another matter. Last edited by inlimbo; 14-Aug-2012 at 02:21. |
|
|
|
|
|
#17 |
|
Junior Member
Join Date: Feb 2007
Posts: 96
|
I stumbled upon this on gaf:
This is what I want. And this looks better than the SDF-based ray traced occlusion you can see in Fairlight's presentation from the Assembly 2012 thread, even if it's less accurate. God, that clip makes me really excited. I can't believe I hadn't seen it sooner, though I suppose Michael Bunnel's technique that MJP linked is closely related. Last edited by inlimbo; 16-Aug-2012 at 22:57. |
|
|
|
![]() |
| Thread Tools | |
| Display Modes | |
|
|