Graphics options/capabilities

[Rob]

I definitely agree that realistic image quality is generally much more important than fancy effects. Kudos to that.

I also agree that there is more scope for parallelism than for pure speed increases, due to physical limitations. Indeed, we should be trying to push parallelism forward, and only software developers can do that. As Tony said, the challenge is to create software that scales well. Specifically, two sorts of scaling are required:
(1) ability to efficiently utilise whatever hardware is available, whether in serial or highly parallel environments. This simply involves implementing efficient practically unbounded multi-threading where possible;
(2) ability to smoothly ramp up the quality of the image or features, in accordance with the hardware (or rather, the desired frame rate). Features such as AA and tessellation are perfect for this, e.g. supersampling could in principle use 1, 10, 100 or 1000 samples, depending on available resources (although in practice there will obviously be an optimal number of samples).

 

[Rob]

Madoc, out of interest, I've got a question about your implementation of stochastic supersamping. Given a fixed number of samples, doesn't that mean that you won't have a unique deterministic colour for each pixel? In which case, doesn't that mean that a static image can have a different pixel colouring in different frames (because it's re-sampled each frame)? Am I correct, or am I being naive? Does this cause a kind of shimmering effect? Do you solve that by sampling until convergence (which would be expensive or cheap depending on sub-pixel heterogeneity)? Or is the assumption that time-averaging will automatically sort that out by tricking our eyes? Is that why you've rendered each frame 8 times and taken the average? Interesting problem, and I'd like to know what you're doing - I think it's great that you're breaking the mould and innovatively implementing your own solutions! Cheers.

 

[Rob]

We're PC gamers and we hate what consoles have done to games. We want more from games than something you play on a sofa with friends while eating pizza and having a conversation, we want something that requires more than 1% of your attention. I'd rather watch a film than play games like that.

PCs and consoles are completely different beasts. Like panthers and sloths, they both happen to exist in the same world. Some people like panthers. Some people like sloths. But you can't take a sloth very seriously. Here, in the PC community, I think we should keep 100% focus on our panthers, and let the sloths do what they want.

 

[The Hedge Knight]

To bump this thread, can we expect ambient occlusion for Nvidia users at all?

 

[Ziegenpeter]

This Infos and Statements by Madoc (especially the Quality of InGame-"Render") are really kickass!!!!!!!

Make sure to get an Update out of this, the people out there should be aware of this amazing stuff, Madoc!!!!

 

[hredthel]

We're PC gamers and we hate what consoles have done to games. We want more from games than something you play on a sofa with friends while eating pizza and having a conversation, we want something that requires more than 1% of your attention. I'd rather watch a film than play games like that.

Nothing overly fancy going on there, it's just efficient stochastic supersampling. This yelds much
superior results to hardware AA methods (which yeld surprisingly poor results) and it allows us to do
a number of effects beyond anti aliasing. We can get away with it because the engine is very fast to
begin with. I consider image quality to be often more important than fancy effects. Once you try it it's hard to go back.

I don't see ray tracing having a significant role in real time graphics any time soon. Raster
rendering is just so much more efficient and still it's too damn slow. When all the Larrabee hype was
going I kept saying it wasn't going to work, everyone disagreed with me and... it epic failed.

Truth is hardware isn't getting much faster at all and manufacturers are running into physical
limitations they don't know how to overcome. Recent advances are more about parallel processing and
architecture than faster chips. Even GPUs haven't been improving much in terms of pixel fill rate
which is where my engine starts to bottleneck.

The modern trend seems to be mid range graphics hardware that is vastly inferior to the high end and
games substituting realistic graphics for stylised simplified graphics. Yuck.

We're PC gamers and we hate what consoles have done to games. We want more from games than something you play on a sofa with friends while eating pizza and having a conversation, we want something that requires more than 1% of your attention. I'd rather watch a film than play games like that.

Yeah, o.k., I'm doubling my pledge.

Madoc, I like your take on the relation of our "civilian" hardware and current games. It really is about still running games on a 4 series and a 2.9 intel dual core. I've watched the tech demos for the new CryEngine and Unreal Engine: and what I see is the first breach (I've come across) in that everlasting wall of "faking" good graphics. Honestly I'm ok with my game using POM vs physical displacement for detail (fuck, why not use both?), but I'm never down with adding voracious specular and bloom effects to COVER up how bad the engine is rendering the game. In all, I think a little substitute in graphics is nice since, seriously: It opens up headroom for both your cpu and gpu to work on processing some new and expansive feature in your game that would normally be a system hog.

 

[BrecMadak]

Great reply, yet great visuals and innovative potentials you and co have right there Madoc ! It's still unbelievable that we would have an isometric RPG like this.. These kind of news solely make hollows come back into life. Haha ha...

 

[leafspring]

Truth is hardware isn't getting much faster at all and manufacturers are running into physical limitations they don't know how to overcome. Recent advances are more about parallel processing and architecture than faster chips.

Well, graphics hardware does get faster as rendering is a highly parallel process and, as you said, current advances are mostly towards parallel computing.
Technologically, we're not that far away from realtime raytracing in gaming by now, but we certainly won't see it happening for another 10 years on the consumer side. Simply because there is no proper down-scaling and it thus requires a certain level of hardware from every consumer which isn't really the best premise if you want to sell a game to millions of people, I guess.

Interesting choice on the AA btw. You really seem to value quality more than performance which is a good thing seeing as barely anybody does these days.
Going to be interesting to see how this performs on lower-end systems (in general, not just AA), though.

 

[Rob]

You really seem to value quality more than performance which is a good thing seeing as barely anybody does these days.

Yes, this is great. It really annoys me on other games forums when you get people saying things like "Oooo, those graphics features look too demanding... I hope it works on the laptop I got off ebay for £50, without it looking too pixellated". Geez. To those people, I say: "what are you doing with a PC? Just get a console if you're interested in inferior products". Even if you're on a budget, you can build a perfectly modest gaming PC without breaking the bank, if you know what you're doing.

Going to be interesting to see how this performs on lower-end systems

Of course, we'll have to wait and see... but tentatively I suspect that Madoc's stochastic AA will scale down as well as up. Depending on the implementation, stochastic AA can be far superior to all other AA techniques (including FXAA, MSAA and SSAA), from all points of view (although FXAA *may* be more appropriate for low-end systems). If you're smart about it, you can design it so that it's applicable to low-end and high-end machines (and future monster PCs), producing results that are high-fidelity, efficient, scalable, self-regulating, with floating-point adjustable detail levels, that result in no permanent nor temporary FPS dips below 60FPS. Definitely sounds like the future of AA to me!

As I said, we'll have to wait and see how it's presented in Sui Generis... although I trust that Madoc will have produced something good.

 

[leafspring]

Of course, we'll have to wait and see... but tentatively I suspect that Madoc's stochastic AA will scale down as well as up.

The problem with super sampling (stochastic or not) is that it is generally the most expensive way to handle AA as it basically comes down to throwing raw computation power at the problem instead of a clever algorithm.
So in the end, SSAA might just not be possible (at a reasonable quality level) on hardware that could very well handle MLAA/SMAA and thus make Sui Generis look considerably worse than other games on lower-end specs.

 

[Rob]

I agree that it's always a good idea to be able to have multiple other methods of AA too, e.g. FXAA & MSAA, in order to cover all bases, depending on what hardware is available.

However, I think that there is a stigma associated with stochastic SSAA due to the similarity in name and approach to traditional SSAA, which is far less effective and much more computationally expensive. However, it all depends on how intelligent the algorithm and implementation is. For example, it could be possible to design a stochastic SSAA algorithm that could operate on low-end systems whilst maintaining 60FPS (providing 60FPS is achievable without AA). The point is it scales. True, I'm not implying that FXAA wouldn't be more effective on a low-end system... that is yet to be seen/proven either way.

Think future tech... not traditional approaches.

 

[Tony]

Not to mention they've already stated that the game engine is well optimized and runs smoothly on a four-year-old PC! And that recording the videos for their updates was more demanding on the system than actually running the game itself. Most technology is limited by how well it is implemented, not on the technology itself. And finding new and innovative approaches seems to be the entire design concept behind Sui Generis

EDIT: oops, after rereading my post I realize I basically just restated what Rob said in his previous post

 

[Madoc]

As I said earlier we still support hardware AA which realistically is impossible to beat in terms of performance, but also pretty damn ugly. Our supersampling is demanding and basically it's something to throw exrta performance at when you can spare it, easily done if you have a decent PC and a great thing to throw it at in our opinion given the results. On lower end systems hardware AA might be a better choice.

 

[Rob]

You really seem to value quality more than performance which is a good thing seeing as barely anybody does these days.

Yes, this is great. It really annoys me on other games forums when you get people saying things like "Oooo, those graphics features look too demanding... I hope it works on the laptop I got off ebay for £50, without it looking too pixellated".

Thought I'd share some comments I'd seen in another forum, which humorously exemplify this point:

• Wanted to chip in to say should the game come with optional high end graphics, please do your best to ensure the low end graphics do not look like crumpled, pixelated whatever. I cannot afford a high end gaming beast
• I can imagine that no matter what sort of PC you've got, it's probably pretty beasty compared with the PCs of 5 or 10 years ago. In 5 or 10 years time, today's PC beasts will likely seem underpowered.
• unfortunately I AM using a 12 year old pc XD. its amazing what these old workhorses can do with proper maintenance and care. i probably need to get off my wallet and max the ram out but to date ive done nothing to the specs and am handling things quite well XD.

WTF?!!! Where do these people come from?!!! And they want/expect modern PC gaming?!!! (...well, they might get away with online flash games...)

 

[Tony]

"...unfortunately I AM using a 12 year old pc XD. its amazing what these old workhorses can do with proper maintenance and care. i probably need to get off my wallet and max the ram out but to date ive done nothing to the specs and am handling things quite well XD."

WTF?!!! Where do these people come from?!!! And they want/expect modern PC gaming?!!! (...well, they might get away with online flash games...)

The sad thing is... most smart phones would be more powerful than a 12-year-old PC

 

[leafspring]

The sad thing is... most smart phones would be more powerful than a 12-year-old PC

But thanks to Android most of these phones would still be slower.

Putting a 1 Ghz dual core CPU and 512 MB RAM in a phone and it still lags at times with ICS (and I'm talking browsing and basic stuff here, not gaming or video playback). I mean, how?

 

[Brendan]

Not even sure where to start on this thread anymore, but y'all (yeah you, Rob) seem pretty knowledgeable on graphics cards etc, so can anyone explain to me why pixel fill rate isn't improving with recent cards? Is it actually a limitation of massive parallelism, or just an oversight due to the inefficiency of typical graphics engines?

 

[Rob]

Not even sure where to start on this thread anymore, but y'all (yeah you, Rob) seem pretty knowledgeable on graphics cards etc, so can anyone explain to me why pixel fill rate isn't improving with recent cards? Is it actually a limitation of massive parallelism, or just an oversight due to the inefficiency of typical graphics engines?

Hi Brendan,
What recent cards have you tried? What's your observation? With ATI or Nvidia or both?

Are you referring to the 8800 -> 660 that Kieran mentioned in another thread? The 660 should be much better than the 8800. The 660 has way many more transistors (around 4 times as many) and shader processors, etc. so it should be able to handle highly parallel jobs better. Plus the cores are faster, and it has more RAM. In fact, it's much better in every way possible. It's very unlikely that a 660 can't do much better. (or is it not the 8800 -> 660 that you're talking about?)

First, I'd start by trying to identify exactly what the bottleneck is. Look at GPU usage (e.g. with MSI afterburner) and see what the usage is on the 8800 compared with the 660. If usage is 100%, then GPU is likely the/a bottleneck. Otherwise, there's another bottleneck, which may or may not be due to the graphics card. First place to check is the CPU - is it maxing out? If not, is it maxing out usage on just one core (i.e. does it require great single-core performance)? Then check the RAM - how much is it using, and what speed RAM is it? If you're 100% sure that there isn't a hardware bottleneck, then the next question to ask will be whether the software is configured to indefinitely take full advantage of the whole GPU, i.e. does it actually take advantage of the more highly parallel environment of the more recent GPU? If unsure, what you could do is try plugging a few monitors into the graphics card - more pixels to render means more work for the graphics card - see if you can get that GPU working closer to 100%!!!

Lastly, stupid question - I assume you're using the same quality/performance options in the nvidia control panel? You haven't accidentally got high-quality settings enabled for the recent card?

Afraid I've got to go now, but will check the thread in the morning!
Good luck!!!

 

[Madoc]

I believe Brendan is referring to the "pixel fill rate" as quoted by hardware manufacturers. While other specs have been increasing significantly this one hasn't. So if you optimised the hell out of everything you're left with the limitation of how many pixels you can output, which is independent of how complex the calculations you perform to compute its values or how much data you fetched doing so.

 

[Rob]

I believe Brendan is referring to the "pixel fill rate" as quoted by hardware manufacturers. While other specs have been increasing significantly this one hasn't. So if you optimised the hell out of everything you're left with the limitation of how many pixels you can output, which is independent of how complex the calculations you perform to compute its values or how much data you fetched doing so.

Perhaps it depends what graphics cards we're talking about - fillrate has more than doubled between the 8800 and 660. Are we talking about those cards?