Lightsmark - new realtime global illumination demo

Everyone who says things like that invariably eats their words when it's put on the table.

Oh, so everyone always have the same preferences as you do? Geometry may mean a lot to you, but it doesn't to me. Some people whined about Doom3 being low-poly, but I thought it looked great. It didn't bother me at all. Heck, I still enjoy playing Unreal classic, and what bothers me the most is not walking on 100m large triangles outdoors but the low texture detail, even though Unreal's geometry with modern lighting may look slightly weird.

In any case, wording it the way you did massively trivializes the sacrifices and just as massively blows the gain out of proportion.
Between the two options, the latter is several trillion times more marketable than the former.

ROFL
 
Oh, so everyone always have the same preferences as you do?
Who said anything about MY preferences? I'm talking about the market at large. My concern is throwing away as little as possible, else it's useless. You can't tell me your assertion is representative of the mass market. Why do you think people whined about Doom3? Because games with significantly higher polycounts had already existed. Why do you think Doom3 didn't do so well on the Xbox port? If the Xbox port came out first, followed later by the PC, it might well have been a completely different story. Why do you think everybody and his brother was applying normal mapping everywhere thereafter? Or that people worked so hard to develop normal mapping on the PS2? Or why ragdoll was bread and butter after HL2?

One good turn deserves another. ;)
 
I'm talking about the market at large. My concern is throwing away as little as possible, else it's useless.

Every game can't be as good as every previous game in every way. Having one or two very strong points creates a lot more buzz than just keeping up with the overall trend with everything.

You can't tell me your assertion is representative of the mass market.

Well, I'd say it's more representative than yours. The market is a lot more diverse than that. Heck, graphics isn't the strongest selling point anyway. CounterStrike was pretty ugly compared to the competition even when it was new, yet it became the most popular game of its time.
 
Besides, you can use low-poly input and get good quality with DPS techniques, and combine that with silhuette tessellation in GS and you can have an order of magnitude less geometry look every bit as good or better.
Seeing how nobody has done that yet and how much of a nightmare it would be to fit into existing game engines and art pipelines, that's not much of a counter point.

Well, I'd say it's more representative than yours. The market is a lot more diverse than that. Heck, graphics isn't the strongest selling point anyway. CounterStrike was pretty ugly compared to the competition even when it was new, yet it became the most popular game of its time.
That's exactly why this doesn't work. People who value graphics enough to reward a game's designers with a sales boost are already a small group.

People who would prefer this GI technique while not minding one or two orders of magnitude less geometry are even smaller, especially a technique that is so specific in its major effect - moving shadowed lights affecting secondary illumination. The only major game I can think of that could significantly benefit from this is Doom3, and even there how often did we use the flashlight?
 
Every game can't be as good as every previous game in every way. Having one or two very strong points creates a lot more buzz than just keeping up with the overall trend with everything.
The part you're missing is that one or two strong points is better, but having even 1 severely glaring weak point is unacceptable. With the levels of graphical throughput we get these days, people will forgive 10-20% weakness in some visual area, because you'll still be in the same league as your competitors (obviously applying some sense of perspective since it's not as though a DMC is out to compete with The Sims on AI nor is The Sims out to compete with DMC on visuals) but not 90%... Especially if you, as a studio, have already done work that far outdoes that weakness that on titles prior.

Well, I'd say it's more representative than yours. The market is a lot more diverse than that.
Genre selections and targets are diverse. The public demands within a given genre aren't very diverse at all. The phrase "Keeping up with the Joneses" seems to leave out that there isn't just one neighborhood. And the problem that everybody seems to be ignoring in spite of how many times I've mentioned it (and it seems quite deliberately so) is that graphics isn't the only loss.

Heck, graphics isn't the strongest selling point anyway.
People don't have the ability to evaluate anything more meaningful than visuals prior to purchase/rental (understandably so). You only have 2 seconds to grab their interest prior to that time of purchase. If you can't do that, it's already over before it started. And that means at least 1 of a few things. Graphics, a ridiculous marketing campaign that keeps the name on the tip of their tongues (e.g. "blasting covenant dogs is thirsty work" type of crap), brand loyalty, graphics, good timing, and indescribable luck.

Everything that's actually important about a game is stuff that keeps sales going beyond that initial 2 week window where your sales figures are pretty much determined, but it doesn't create initial sales velocity.
 
Seeing how nobody has done that yet and how much of a nightmare it would be to fit into existing game engines and art pipelines, that's not much of a counter point.

Well, it's on my list of things to experiment with, so I can't give you a final verdict, but there shouldn't be much of an issue. DPS is an already proven technology that's been used in several games. Smoothing silhuettes with GS should work fine. Performance should be OK given that most triangles are not tessellated, and it should not have the art pipeline issues of say N-patches because you have access to adjacency information.

That's exactly why this doesn't work. People who value graphics enough to reward a game's designers with a sales boost are already a small group.

No, that's why it works. No one's arguing that implementing this in a game will make it a top seller, or even sell better than without it. However, the OP argued it could be used in a game. I agree with that. Using this technique, with its drawbacks, won't be a catastrophy for the game.

and even there how often did we use the flashlight?

Constantly! :p

The part you're missing is that one or two strong points is better, but having even 1 severely glaring weak point is unacceptable.

Well, I disagree, and I don't think the market at large agrees. I don't have any statistics to back that up, but I haven't seen any numbers from you either supporting your notion, so this is a pretty pointless argument.

People don't have the ability to evaluate anything more meaningful than visuals prior to purchase/rental (understandably so).

So you never had a friend go "you should try game X, it's pretty good"? I would say the friend to friend factor is far more important that any other factor in purchase decisions. Graphics may give you a good headstart in reviews, but what gives you a long-term success is the fun factor. Lot of people just happened to think CS was a fun game, despite its ugliness, and hence it became popular. Same with WoW. A few guys at work liked it and recommended everyone else, and soon everyone was playing it, except me (and to this day I'm refusing to try it out!!! :p)
 
So you never had a friend go "you should try game X, it's pretty good"? I would say the friend to friend factor is far more important that any other factor in purchase decisions. Graphics may give you a good headstart in reviews, but what gives you a long-term success is the fun factor.
You talk as if word-of-mouth just happens as if out of nowhere, which is just plain absurd. Simple logic should tell you that word-of-mouth only works when there is a first mouth... You have to have someone's interest first before you can get word to spread. If you don't have that, there never was a shot.

And again, you don't have this enormous time window to do that. Why do you think Psychonauts was a failure? It wasn't anything about the game itself. It's that those few moments they had to catch the interest of at least a handful of gamers and create impetus for the early sales velocity ended up being seconds where they lost the interest of almost everybody and would-be buyers had no idea what the game was about, assuming instead that it was some kiddie title that wasn't worth wasting their time reading about. The critical acclaim was just too little, too late.

What you refer to as a "head start" is really the one and only opening you have. If you bothered to read the last parts of that post, I also said that long-term success is not written by the things that give you the initial sales. But without the initial early success, there is no opportunity for long-term success in the first place. This has been shown time and time again, both with major AAA hits and with hidden gems that failed in spite of themselves. WoW, for instance, had massive anticipation, and a following driven by the attraction of the "Warcraft" name, so it had that window well before it was out. And Starcraft 2 will be no different even if it is somehow completely outclassed.

If you aren't in the public eye early, you never will be. There's no such thing as a belated success, and this isn't just seen in games but entertainment media in general. Sleeper hits and unexpected successes do happen once in a dozen or so blue moons, but they're designated as sleepers and unexpected for a reason. They are the exceptions, never the rule. Relying on the hope that you'll be an exception to the rule is always a fool's game.

Sure graphics isn't the only way to get that initial sales velocity, but it is the most obvious, especially since the first thing people typically know about your game is the graphics alone. It appeals at the most basic of levels, and in a manner of speaking, it's a form of idiot-proofing. Conversely, failure to excite early on is self-dooming. For something like this to even have a shot at working, you'd first have to be working on something that has little need of flexibility in the variety of scenes you can show, and secondly, never show a single screenshot until you've crammed the words "global illumination" down people's throats so far that they can't help but develop diarrhea of the mouth over GI. Quite simply, it creates the impression that you're doing something nifty that is beyond simple words to describe (hence why people will wait for the thousand words). And then you show screens which are so specific in their viewpoint that they show only what does work particularly well and doesn't show any sign of its failings, which itself is not easy to do, because the opposite is so easy.
 
You talk as if word-of-mouth just happens as if out of nowhere, which is just plain absurd. Simple logic should tell you that word-of-mouth only works when there is a first mouth... You have to have someone's interest first before you can get word to spread. If you don't have that, there never was a shot.

In this time and age with internet and forums there's no lack of first mouths.

Sure graphics isn't the only way to get that initial sales velocity, but it is the most obvious, especially since the first thing people typically know about your game is the graphics alone.

I don't know about you, but I find it rather rare that I hear about the graphics first about a game. Screenshots tends to come long after the initial hype.

To be honest, I don't know a single game that succeeded without having the fun factor. There have been plenty of good looking hyped up games that failed, because the fun wasn't there. And there are plenty of games with non-impressive or even ugly graphics that succeeded anyway. As much as I care about graphics myself, with it being my profession and everything too, even for me I'd rather play a fun game than a pretty one. That's why I prefer UT classic to UT2003/2004.
 
Well, it's on my list of things to experiment with, so I can't give you a final verdict, but there shouldn't be much of an issue. DPS is an already proven technology that's been used in several games. Smoothing silhuettes with GS should work fine. Performance should be OK given that most triangles are not tessellated, and it should not have the art pipeline issues of say N-patches because you have access to adjacency information.
Smoothing silhouettes with the GS messes up your art pipeline because you can't apply it to a lot of models designed without it. The artist need to know about this smoothing beforehand. Anyway, smoothing is not the big issue here. It's the silhouette detail lost with low poly models that really impacts image quality.

I was experimenting with some techniques to use the alpha channel and curved rays in a form of parallax occlusion mapping, but there are cases that give you horrendous artifacts (holes in the model!), and now I'm pretty sure this route is a dead end unless you have convex objects.

The best solution to scale this technique to more detail is not low geometry hackery but simply use granularity less than per vertex calculation. Use fixed weights and vetex texturing to sample from a few nearby radiosity points. It requires work (possibly automated) to place these points around your models, and each point needs some method of capturing the lighting properties of all the vertices around it, but it can be done.

The problem, as always with GI approximations, is scaling this technique to real scenes.
 
I don't know about you, but I find it rather rare that I hear about the graphics first about a game. Screenshots tends to come long after the initial hype.
How can initial hype be based on fun? The game isn't finished yet and nobody's played it!

Games with big hype are almost entirely based on graphics. Gran Turismo 5, Gears of War, Killzone 2, Crysis, UT3, etc. Otherwise, it's due to history of a game franchise, and even though that could well be based on fun or storyline, the vast majority of game developers don't have this luxury to generate hype. Screenshots are everything for hype.
 
In this time and age with internet and forums there's no lack of first mouths.
Not if you don't have people's interest first. By the same logic that makes it easy to spread word quickly, it's every bit as easy to stop the spread of word -- all it takes is one mistake. But if you prefer, then I'll rephrase to "first mouths that speak positively and loudly" You also seem to forget that the average buyer isn't out scanning the net for info constantly or chattering on forums... or even buying from gaming-specific retailers as often as from general retailers like Walmart.

I don't know about you, but I find it rather rare that I hear about the graphics first about a game. Screenshots tends to come long after the initial hype.
You haven't paid much attention then. That's certainly true of major huge-budget AAA projects with production values so great that early hype is a must, but those really aren't the majority of titles that come out. These are just the majority of the titles that most of us hear and talk about as if . There are several times more titles that come out where the first thing you'll ever see is screenshots and/or trailers. The thing is that it's these major titles that get all the attention, which only goes to show that the more common case lacks "first mouths."

To be honest, I don't know a single game that succeeded without having the fun factor.
I reiterate, you haven't paid much attention. In fact, for this one, I'd all but say you haven't paid any attention at all. Because the majority of your sales happen in the first two weeks after release, there are things which are commercially more powerful than fun like "hype" and "name" and "getting in bed with the press" (the last one isn't a regular occurrence or anything, but it isn't exactly unusual, either). And like Mintmaster posted above, fun doesn't often impact these.

There have been plenty of good looking hyped up games that failed, because the fun wasn't there. And there are plenty of games with non-impressive or even ugly graphics that succeeded anyway.
The latter is an exception, not the rule (I'm assuming that you're still keeping things in persepective wrt what sort of titles a given title would be competing against). I can think of just as many if not more games which weren't particularly great on gameplay or in some cases, just plain bad, but looked good enough or had something worth hyping enough to hold people's attention so as to get that early sales velocity. Did such releases have great long-term staying power? Not often, but they didn't need it to be financially successful.

The former happens a fair bit, because it does take more than graphics to compete, but either way you have to make people think you are competitive enough to be worth picking up, or you're done.

Of course, the most common story (more common than pretty much all others combined) is the game which is more than adequate in all areas, but simply was not big enough -- it just fell in a shadow. Ultimately, the overall problem with marketability is one of being noticed in a positive light and staying there... That's why so many cookie-cutter titles are out there -- certain things are known to work. People aren't totally non-forgiving, but there are limits to what will pass, and dropping below that range is not safe nor will it ever be. In terms of absolutes, you can actually sacrifice quite a bit graphically or physics-wise or AI-wise or whatever for a little more power to be dedicated elsewhere, as long as you make up for it in some way. But making it sound as simple as "fun=everything" is incalculably naive. It's not that people don't want fun or want graphics only, but that they want everything.

It's easy to say something like we've dropped the polycount on our main character by ~15,000 tris from a year ago, but that means nothing without some point of reference. What people notice is the relative difference, and you can't hide everything behind a veil of "but look at the lighting!". And there's only so much you can get away with anywhere and still have a shot at success (let alone actually achieving it). You can't cross that line and hope lightning will strike.

Doing that is just never a good idea. Sure, you can refer to it as bravery or blazing a trail, but there's a fine line between bravery from foolhardiness.
 
Oh wow, you guys are completely off topic. Take it to another thread or to PM, please.
I don't know if it is. We're (indirectly) arguing the feasibility of using this technique from the OP in games.

It's a per-vertex technique using CPU readback, and as such the vertex count is severely limited. That has implications.
 
Last edited by a moderator:
Thanks for feedback.. I was busy fixing driver issues etc. Updated version has also shimmering fixed.

Regarding (difficult) integration into modern engines, it was done before and it is in progress again, it's not difficult.
 
Dee,

The demo is very impressive. On a 2x2.8Ghz AMD with a QuadroFx 4500, 34.4 fps at 1920x1200, 50.6 fps at 640x480.

I would be very happy to have realtime radiosity in a game, even if it didn't work in all possible scenes.

Can you describe the drawbacks and challenges of your system and what might be done to overcome them?

For example:
Transfering data for each vertex back and forth between the CPU and GPU is difficult to scale. Could it be done instead with a texture where each texel represents a sample point on the surface of the mesh? That would decouple the data set size from the mesh complexity. Instead it could scale with lighting complexity.
If you pipelined the render/read-back/process/write-back stages across multiple frames, the latency would obviously increase, but would it avoid stalling? Or are you already avoiding stalling while waiting on CPU<->GPU transfers?
 
Dee,

The demo is very impressive. On a 2x2.8Ghz AMD with a QuadroFx 4500, 34.4 fps at 1920x1200, 50.6 fps at 640x480.

I would be very happy to have realtime radiosity in a game, even if it didn't work in all possible scenes.

Can you describe the drawbacks and challenges of your system and what might be done to overcome them?

Thanks,
the biggest problem was to find workarounds for driver bugs that were not fixed yet or that were introduced recently.
Bad tessellation creates some artifacts. As a game developer, you can fix it in preprocess step.

For example:
Transfering data for each vertex back and forth between the CPU and GPU is difficult to scale. Could it be done instead with a texture where each texel represents a sample point on the surface of the mesh? That would decouple the data set size from the mesh complexity. Instead it could scale with lighting complexity.

Yes, this is good idea.
Radiosity calculation is more expensive than data transfers, but your solution improves both.
I work on different approach to scene complexity, but it's still too soon to reveal.

If you pipelined the render/read-back/process/write-back stages across multiple frames, the latency would obviously increase, but would it avoid stalling? Or are you already avoiding stalling while waiting on CPU<->GPU transfers?

I experimented with lag before and it would look ok in demo, but it's not good for all situations, so I prefer stalls.
I fill CPU stalls with dynamic character jobs, but demo with only one character is pretty simple, so multicore CPU has still lots of free time (good e.g. for AI).
When I filled GPU stalls by reordering work, fps usually dropped, so I stayed on straightforward path with one stall. I believe that unrelated GPU jobs like particle physics would fit in perfectly, without decreasing fps.
 
Back
Top