Agni's Philosophy runs at 60FPS on a GTX 680, uses 1.8GB VRAM. Can next-gen run it?

Ether_Snake

安安安安安安安安安安安安安安安
I'm not aware of any developer who builds levels 10x bigger than what current consoles can actually handle. So that's definitely going to take more time and goes back to the problem I'm speaking of.

And having better toolsets doesn't matter if games aren't being made faster. Well, you could hypothetically create current gen games faster, but is that the standard you want on hardware more capable than that?

Nope. Next-gen we'll be able to produce highly realistic characters with a few clicks. No need to hand-model all character models anymore. Heads are scanned then split into individual pieces, and can then be mixed-and-matched together to produce new heads. It will also be easier to re-use assets in different productions.

It will be very easy to create a huge landscape a la Skyrim, and automatically have grass, rocks, trees, plants, etc., all placed automatically. Even cities can be created like this. Weather, day/night cycles, temperature effects, can all be done in a way that doesn't imply doing micro work, it can all be implemented at a high level rather easily.

It will be a lot more easier to iterate on next-gen, to change the layouts during the production, or to make bigger worlds.
 

King_Moc

Banned
Nope. Next-gen we'll be able to produce highly realistic characters with a few clicks. No need to hand-model all character models anymore. Heads are scanned then split into individual pieces, and can then be mixed-and-matched together to produce new heads. It will also be easier to re-use assets in different productions.

It will be very easy to create a huge landscape a la Skyrim, and automatically have grass, rocks, trees, plants, etc., all placed automatically. Even cities can be created like this. Weather, day/night cycles, temperature effects, can all be done in a way that doesn't imply doing micro work, it can all be implemented at a high level rather easily.

It will be a lot more easier to iterate on next-gen, to change the layouts during the production, or to make bigger worlds.

Sounds like everything would look very generic with this approach. I suppose you can amend everything after though.
 

i-Lo

Member
characters will look even better, but i think they look pretty good right now.

environments will see the biggest jump in fidelity IMO.

Dear Mr. Credible Hulk, or anyone informed about graphics and polygons, I wonder if you would be so kind as to explain to a lay person like me a few key things:

  • Why is the throughput poly figures for un-culled and culled different?
  • Don't games today have some form of culling in the rendering pipeline to inhibit drawing of unseen polys on screen during final rendering operation?
  • Is culling used differently in engines that utilize "deferred rendering"?
 
Maybe i got this wrong my impression was just use tessellation because we have it?
It has it uses but please not like Crysis 2 cement blockades. That was Tessellation because we can do tessellation and write that of the checklist.

you have no idea of what you are talking about, and that Crysis 2 cement blocks article was complete bull shit. Tessellation can be used in this geometry or any geometry, see Crytek's adaptive tessellation. it elliminates pop ups and not as performance intensive as LoD on "ultra"
 

Mesharey

Member
02.jpg

This poster. :O
 

Shikoro

Member
Tessellation isn't a option or solution for all kind of geometry.
Of course not, but it is absolutely necessary to use it for LoD.
http://www.youtube.com/watch?v=sQQpCd_vvGU

Yes, the engine only draws what the camera sees.
http://en.wikipedia.org/wiki/Back-face_culling
Occlusion culling is the more important thing.

This poster. :O

HNNNG. I want that poster, and also her
be put in a game. :O

Why does she need a microphone? She's not real.
lol, didn't even notice. XD

Can next gen consoles run it? Should do as they mostly go for 30fps.
It all depends on what type of memory Sony and MS go with.
 

mrklaw

MrArseFace
Surely if you want to replace LoD and storing multiple levels of detail for models with tessellation, you'll still need to store multiple levels of normal map. You won't want to store the full detail one for objects far away. Even if you aren't drawing it, it'll still be taking up ram.
 

Shikoro

Member
Surely if you want to replace LoD and storing multiple levels of detail for models with tessellation, you'll still need to store multiple levels of normal map. You won't want to store the full detail one for objects far away. Even if you aren't drawing it, it'll still be taking up ram.
Sure, but compared to objects popping out of nowhere and swapping models every 20 game meters, this wouldn't be that noticeable, if at all. :D
 

i-Lo

Member
Occlusion culling is the more important thing.

I just saw the video of Occlusion Culling from Unity. In it they mention that "Umbra" is the industry-wide provider of this technology. So, is this the new standard when it comes object culling (because the procedure seems logical and remarkable) and is it used by all major middlewares (UE3.9/4, CryEngine 3, Phyre Engine, id Tech 5, Gamebryo etc)?

Surely if you want to replace LoD and storing multiple levels of detail for models with tessellation, you'll still need to store multiple levels of normal map. You won't want to store the full detail one for objects far away. Even if you aren't drawing it, it'll still be taking up ram.

You could stream it off of the HDD. Let's assume for a moment that an object imbued with tessellation requires 5 normal maps going for distances ranging from 1KM to Point blank. Given distance between dynamic objects doesn't change too dramatically from very far to very close, and given it'll most likely be linear, couldn't you resort to storing a subset of those 5 normal maps (say 3 for objects that move unpredictably) which should provide enough detail at a given distance and enough time to stream other normals to RAM for the potential transition? It'll be akin to those clutch-less transmissions that pre-select two additional gears (one above and one below) instead of all the available ones.
 
Xenos supposedly is rated for 500 million polygons, whilst that is obviously much greater than is realistically possible, there's no reason to think that a GPU that is more than 13x faster won't be able to at the very least reach 100 million above the Xenos' theoretical max.


Someone mentioned that certain scenes would be pulling about 10 million polys in real time + texture. IIRC RSX in PS3 is capable of around 250 million poly/sec (I don't know if they're textured or not) which amounts to around 8.33 million polygons per 30th of a second in a 30 fps game. So, a jump of 1.7 million polys per 30th of a sec equates a total of 300 million polys/sec. Doesn't sound like a lot and was wondering if it's indicative of what sort of GPU power are we talking about for next gen...

That's nothing more than simple wireframed untextured polygons with nothing else going on. PS3/360 don't even come close to reaching 100m polygons per second in any actual video game scenario. Hell I doubt any 360/PS3 game has crossed the 50m threshold.
 
So regarding character polycount for example, kratos was 5k in god of war, then in gow3 he was about 22k. What does anyone think will be the normal polycount of characters next gen? I say about 60-100k.
 

Lathentar

Looking for Pants
Nope. Next-gen we'll be able to produce highly realistic characters with a few clicks. No need to hand-model all character models anymore. Heads are scanned then split into individual pieces, and can then be mixed-and-matched together to produce new heads. It will also be easier to re-use assets in different productions.

It will be very easy to create a huge landscape a la Skyrim, and automatically have grass, rocks, trees, plants, etc., all placed automatically. Even cities can be created like this. Weather, day/night cycles, temperature effects, can all be done in a way that doesn't imply doing micro work, it can all be implemented at a high level rather easily.

It will be a lot more easier to iterate on next-gen, to change the layouts during the production, or to make bigger worlds.

So much fantasy in this post.

Highly realistic characters with a few clicks! However, every important character needs to be unique. If your game is full of only humans at least you could share the same rig and animations! Still need to make a unique set of animations for everything your character can do. Probably can't get those from the animation bank.

Does that character need to talk? Well either have a series of canned animations for chatting or you'll need to custom animate each scene or most likely with such high fidelity you'll need performance capture with high quality voice actors. (None of this can be automated).

Non-human characters will most likely be entirely custom and need to be done from scratch. Sure the character model/textures might be created at a higher quality than what you see in game but the character rig probably isn't. With a high quality model you need to have a higher quality rig so all the animations and weighting work well. This takes the animators a lot more time then they used to. Combined with the game most likely needing more animations than they used to. Once you get it into the game and find out something needs to be tweaked, the character pipeline is much longer and causes more of a headache for each change.

Procedural generation of vegetation and rocks, etc. They did that in Oblivion and then went back to more manual placement in Skyrim. Tod Howard (Executive Producer) has said that this was because they got much better results out of manual placement than procedural generation. Even with something like SpeedTree you still need to have someone learn the middleware, integrate it into your platform, verifying and checking the results.

What I think the person you quoted was referencing was even though high quality assets are created for characters using ZBrush you don't see this type of work being done for the game world itself. It would take far too long.

Tools are getting better, but productivity gains are not outpacing the capabilities of hardware and expectations of users. Especially for games breaking new ground.

Good luck if you try a unique art style with non-standard lighting! Not going to find much for you in that art data bank!
 

Dr. Light

Member
That's nothing more than simple wireframed untextured polygons with nothing else going on. PS3/360 don't even come close to reaching 100m polygons per second in any actual video game scenario. Hell I doubt any 360/PS3 game has crossed the 50m threshold.

Joke post? We already had PS2 games running at 10-15 million polys per second (Jak series). Dead Rising was apparently 3 million polys per frame x 30fps which equals 90 million polys, which sounds about right. I wish they released the geometry figures more often, we rarely get them these days. I know the Ratchet PS3 games must be pushing shit tons. Call of Duty is probably up there as well.
 

JordanN

Banned
So regarding character polycount for example, kratos was 5k in god of war, then in gow3 he was about 22k. What does anyone think will be the normal polycount of characters next gen? I say about 60-100k.
With shaders coming to play, simply adding more polygons wont be cut and dry.

For example, here's a "next gen" character even though she has only 20k polygons.
You wouldn't know why it's next gen till I told you she has displacement maps.

Also, correct me if I'm wrong but engines currently have a limit to the amount of how many polygons a character can have. So you can't insert a 60k poly character if the limit is 40k.
 

drexplora

Member
So regarding character polycount for example, kratos was 5k in god of war, then in gow3 he was about 22k. What does anyone think will be the normal polycount of characters next gen? I say about 60-100k.

wasnt kratos 80k polys in gow3?
also wasnt there only 1 lod level for his model?
 

Pachinko

Member
The listed polygon pushing power has always been a bit of a BS statistic. The rule used to be to say how many triangles the GPU could toss out per second. But games run at mostly at 30 FPS or even 60 FPS . And on top of that, the triangle number used is implying no textures , no filters, no shaders, nothing. Just flat, possibly single color polygons. Textures and lighting take a sizable dent out of those figures but I'm not tech savvy enough to know just HOW big a dent that is. Filtering those textures and adding shaders is a monumental hit as well although the shaders can hide the low poly models quite well if used correctly.

One thing I do know from running the command console in crysis- that game is pushing between 700,000-1.2 million triangles in any given frame. I don't know what 5870X2 means in terms of raw polygon pushing (the bullshit number) but I do know that performance peaks for me around 650,000 polys a frame, I still manage 40-60 fps. Every 25000 or so extra triangles beyond that start costing me frames and when it hits that high point of 1.2 million or so the frame rate drops to a stuttery 30 fps.

----

Another poster speculated on max poly output for character models for game engines on the next set of consoles and honestly , I've long figured developer can go 2 ways with this. Since the VAST majority of the best selling content on 360/ps3 ran at 30fps while barely being 720p (quite a bit was just shy of that resolution) I think we'll see a ton of games using roughly the same engines as today but they'll run at 1080P and at 60FPS with vsync and higher end filtering as well as much better antialiasing. In those cases obviously the models will be identical to what they are now. That number seems to be a large range but I'd say 10-20,000 polys is what constitutes normal these days.

Likewise though, we will of course see plenty of developers pushing the hardware in new ways with direct X 11 features cranked to the nuts, aiming for 30 FPS , possibly knock down the resolution to 720p. Would not be unheard of to see character models in such a scenario be 50-100,000 a piece. Given the time needed to make that level of art though I suspect less polys + improved shaders will be closer to what we'll get. So think, better clothing textures, better hair and cloth physics , better lighting and dust effects, sub surface scattering on skin , etc etc . All that extra detail won't need more polys, I'd say the raw polys would be better used making larger environments and rendering more things on screen at once.
 

nasos_333

Member
http://game.watch.impress.co.jp/docs/news/20121129_575412.html

New article has revealed the specs of the PC that the Agni's Philosophy tech demo was run on: GTX 680, i7-3770K 3.5GHz and 32 GB of RAM. Apparently the CPU was hardly stressed and it could have run on something much weaker. It also only used 1.8GB of texture data so next gen consoles should have very little problem fitting it onto memory. In terms of anti-aliasing I think it may have used a combination of 8xMSAA and FXAA which is massively impressive if true.

What was most impressive was that it runs at 60FPS as shown by this slide:
1040kmz.jpg



What implications does this have for next gen? Considering that the rumour for PS4 is that it has a 7870, is there any reason to think that it won't be able to run this same fidelity of graphics but at 30FPS?


Here's the original video:http://www.youtube.com/watch?v=UVX0OUO9ptU

Here are some new screens:

DXT compressed textures are 1:4, so it would just need 1.8/4 = 450MB VRAM to run it with compressed textures
 

TheD

The Detective
One thing I do know from running the command console in crysis- that game is pushing between 700,000-1.2 million triangles in any given frame. I don't know what 5870X2 means in terms of raw polygon pushing (the bullshit number) but I do know that performance peaks for me around 650,000 polys a frame, I still manage 40-60 fps. Every 25000 or so extra triangles beyond that start costing me frames and when it hits that high point of 1.2 million or so the frame rate drops to a stuttery 30 fps.

Crysis has tons more polygons than 700000 to 1200000 per a frame.

At least 2x that.
 
Image quality should definitely be a priority, I've played so many PC games with advanced graphical features that look like shit in motion thanks to jaggies, flickering shadows, pop-in etc.
Meanwhile a game with much less advanced graphics looks better simply cause it has really good IQ, it looks clean, sharp but most of all it has a good AA solution that works.

Not to mention the part where more angular, less busy, higher fidelity graphics and higher framerates lend themselves to smoother and better gameplay experiences. This is part of the reason why CoD "feels" so much better on the consoles than most of its competition. I still think the Source engine is all around the best engine for gameplay purposes for these reasons. Great performance, great IQ, not too much BS postprocessing or special effects, dust and blur and other stuff crapping up the frame.
 

charsace

Member
Yes it DOES!

(Yes it is warhead, but it is not much different than Crysis 1).

1 million is not that much, even for a console game.
My mistake, I was confusing it with something else.

LP pushes more than Crysis though and at its peak it pushes 3million polys. Dead Rising pushes ever more.

You have to remember that Max numbers are listed as polys per second. Most devs are talking about per frame.

And you have to keep in mind that a lot of systems can push more polys, but usually have hard limits. The xbox360 for example can handle 6billion verts, but the triangle limit is half a billion when it could be 2 billion or more.
 

i-Lo

Member
First level, can be over 2.6 million

ipTX8N4DajGqZ.jpg

Wow, this is pretty revelation worthy stuff. Thank you. Is this the where the number peaked in the entire game?

Also, did anyone run a similar benchmark in Crysis 2 in a scene he/she deemed to be geometry heavy?
 
you have no idea of what you are talking about, and that Crysis 2 cement blocks article was complete bull shit. Tessellation can be used in this geometry or any geometry, see Crytek's adaptive tessellation. it elliminates pop ups and not as performance intensive as LoD on "ultra"

That was a example off how not to do tessellation. Tessellation on say landscapes is a better options.
Like mountains or that nvidia alien demo. Dont think lod will be such a big problem next gen i want tessellation to be used for more exotic things.
 

orioto

Good Art™
I don't think i would go for crazy high poly model for characters. I would use the best level of geometry we have this gen, with better normal maps and textures, hopfully big improvment in dynamic animations and their transitions, same level of geometry for background (when its oftenly low poly right now) + lots of post processing tricks and dynamic stufs, particles, clothe etc... And a good iq. I'm sur it would look amazing and movie like.
 

DieH@rd

Banned
Crysis 2 tessellation support was heavily influenced by Nvidia and their bullshit need to provide better FPS than their competition.
 

i-Lo

Member
search my post history to find out how this is wrong. I have to post this all the time

With all due respect, you can show it to us.

Also, are you saying it's wrong because the evidence documented is falsified or that later patches rectified the issues first introduced with tessellation functionality?
 
if next gen couldn't run it, then there would be no point in developing it...why would Square Enix gather all the resources to make an engine that only runs on PC? Yes, it will run at the highest spec on PC but next gen systems should be able to handle it.
 

Stallion Free

Cock Encumbered
With all due respect, you can show it to us.

Also, are you saying it's wrong because the evidence documented is falsified or that later patches rectified the issues first introduced with tessellation functionality?

The claims about the water tesselation being present in every level being the performance drag was proven to be a bullshit claim.
 
Top Bottom