Digital Foundry Article Technical Discussion Archive [2015]

Status
Not open for further replies.
The problem is that there isn't just direct light coming from one direction in reality, so if you just simply multiply the lighting pass with an occlusion pass in comp, you're already wrong.

As VFX Veteran has already mentioned, offline CG has moved to brute force path tracing and that takes care of more than just the phenomenon that you've mentioned, it calculates the amount of light that hits a surface point... kinda properly. The renderer will take into account all kinds of bounce light and such, so the results are more accurate compared to just using occlusion.
 
It's still occlusion - it's just occluding light from all around instead of a singular point. The realtime hacks are fudging the effect of objects blocking rays of light and reducing illumination strength. Hence, being pedantic, there is such a thing as occlusion. Ambient occlusion doesn't really exist because there's no such thing really as 'ambient light'. Ambient light is either from a light source or reflected light. But all shadows, hard and soft, are occlusion. ;)
 
Imagine a simple scene - a character standing in front of a wall.

Ambient occlusion will darken the wall behind the character, no matter if it's a 2D hack or a "proper" 3D raytraced solution, because the character blocks the environment's visibility.

Now imagine it's Star Wars, the character is a Stormtrooper in white armor and there's a light right above his head on the wall.
The armor should bounce back a LOT of that white light on the wall so it should actually be brighter than where there's no Stormtrooper in front of it.
AO can not take that into account as it does not calculate bounces, not in 3D and definitely not in screen space.

Now imagine it's an Imperial Guard in red armor... ;)

The general idea behind the tech is wrong from the start, it doesn't account for bounced light and as far as I know it's not really possible to do so. But if you do proper raytraced lighting with bounces, it'll look correct. It's a hack, an approximation, that worked well enough at the time when brute force path tracing was way to expensive to be practical but it's still wrong. Offline CG has eventually replaced it, and hopefully realtime will follow sooner or later, too.
 
Article mainly seems to analyze frame rate, but I was hoping they focused more on identifying graphical features.
Based on this beta, the visual progress from Halo 4 is largely brought by way of a new lighting model and screen-space effects, such as motion blur, ambient occlusion, bloom and lens flare.

Hmm still curious as to what they mean by a new lighting model, would have been nice to provide some context of Halo 4 vs Halo 5, I'm not entirely sure to know what to look for.

And here's an amateur question, if Halo 5 final is developed for DX12, then the current beta is only leveraging the subset of features that are common between the two APIs until consumer units are patched or is this a wrong way of looking at it? Like why bother optimizing for D3D11 when you know everyone will be on 12 right? i.e. This beta is basically a complete throw away as a rendering showcase
 
Imagine a simple scene - a character standing in front of a wall.

I agree, but you have both occluded light and reflected light in effect. Adding occlusion is a step up from none and a step down from reflected light.

I don't disagree at all with your sentiments; I'm just being pedantic about the terms and what occlusion means. It's a physical phenomenon, the counterpart of emission and reflection, and possibly transmission - I'm sure you know more about these than I do! Quite often the high-level science isn't particularly accurate.
 
Well, you hope 343i is aware of the framerate hiccoughs at least, and perhaps the animation issue in that pre-battle sprint thing.

I think it's not an issue but intended by the devs in order to process the data being loaded. Remember in MCC where the game could have ~30fps drops at the beginning of some levels because of data streaming/processing, they must have wanted to avoid that in Halo 5, at least in the beta.

Because I found that the camera updates at exactly 30fps during those "cutscenes" (well at least in the gamersyde vid) contrary to what Mr Morgan wrote:
but the character animations and camera update at around 30-40fps.
 
Last edited:
Yeah I was talking about ambient occlusion as a tech, an intermediary step that's technically wrong, I thought that was clear. Reflected or bounced light is in my opinion a part of the lighting as a whole.

The point is that games are still stuck in that intermediary step and while it's a good solution for now - and something I don't expect to be gone in this generation of consoles - it's still not correct and should be left behind eventually. Any additional developments here, as good as they might be, are a technical dead end.
Then again I also think the same about normal mapping, for example, or shadow buffers.

But of course I'm also quite aware of the performance considerations of brute force path tracing - the computational needs are extremely high, and also quite variable, so it'll be quite a while before we'll start to see that kind of tech in games. On the other hand, game rendering programmers have been very clever in coming up with approximations for all the 'proper' tech that had more reliable performance characteristics, so I fully expect some new, currently unimaginable tech to arise in the next decade that'd be a more accurate or 'correct' solution.
 
i.e. This beta is basically a complete throw away as a rendering showcase

Yeah, that makes the most sense. If anything, the beta seems to be about gathering experience and user feedback about the gameplay and the renderer is probably just a placeholder.

Then again, it could still be relatively close to their final version - especially if you consider that 343 will probably have enough time to release a Halo 6 on the X1 where they can leverage a lot more knowledge about the architecture. Focus on gameplay in H5 with a scaled up H4 renderer, and then build a new one on top of that...?
 
Then again, it could still be relatively close to their final version - especially if you consider that 343 will probably have enough time to release a Halo 6 on the X1 where they can leverage a lot more knowledge about the architecture. Focus on gameplay in H5 with a scaled up H4 renderer, and then build a new one on top of that...?

Agreed H5 could definitely swing either way. 343i bit a off a lot this year with having to make Halo 5 and to some degree assist with Halo MCC and to a degree Halo Channel. Their largest priority may not even be graphics as well; the resilience, consistency and quality of their multiplayer platform is more pivotal to the success of their game (in the long term) and indirectly Xbox Live and as a service.

ps. Thank you mods for the stealth edit, I scrolled around for a bit too quickly and didn't realize it was already posted.
 
The foliage on Regret and the ships on Empire and Truth seem to be moving at a low frame-rate, no? Looks weird. Hopefully they get that patched for the final game.
 
Interesting is the fact that the good utilization of PS3's cell yields results that rival those of PS4's and XO's CPUs.

Cell (7 SPEs + one PPE) at 3.2 GHz actually has twice the theoretical peak FLOPS of a 8 core 1.6 GHz Jaguar. I am suprised that it doesn't beat the Jaguar in this particular scenario since cloth simulation (heavy linear vector crunching) is a pretty much a perfect use case for SPUs. This result also points out the fact that Cell loses very badly in everything more complex (branches, memory pointer chasing). 600 cycle memory latency and no branch prediction makes complex things hard. Jaguar on the other hand has nice vector units with super low latency, allowing you to mix vector math with other code easily. Jaguar can be easily beat Cell by 5x or more in more complex code.

The slide actually says they used 5 SPUs vs 6 Jaguar cores in the PS3 vs PS4 time.
They probably need to reserve remaining Cell resources for other tasks (e.g., OS chores, audio, help out RSX, blah). It's unrealistic to benchmark all 7 SPUs + 1 PPU for cloth sim.

Without knowing what exactly they did, it's hard to say how well the SPU cloth sim are coded.

Compared to a console CPU in similar era, the 360 CPU (2 [?] cores @ 3.2GHz) performed at about 1/3 of the 5 SPUs or 6 Jaguar cores.

But yes, the SPUs suck at "general purpose" programming.
 
Last edited:
I am surprised about the H5 DF review...it seems that they really aren't impressed and are especially disappointed with the 720p image compared to the remaster.

They especially mentioned that for large scale and long distances, the current resolution is not very easy on the eye. This is what I also mentioned in the BF4 thread that for large scale battles, resolution matters.

I hope they go the KZ:SF route and have a 30Hz SP campaign with upped resolution and AA.
 
Granted in h5 beta there is 0 AA solution visible compared to bf4. And despite how hard it is to see players have adapted quite well.

Arena shooters are very much based on map control so you should already be looking where your opponents could be relative to you. That and you can't go prone behind foliage.

Bf4 is completely wide open with squads needing to do firing arcs to cover larger areas and people will camouflage behind cover. And lastly time to kill is so much higher in halo, most long range hits provide you ample time to seek cover after being hit.

The situation is not easy on the eye today but does not nearly affect the game as much as one would believe.
 
There's definitely a peculiar perspective to it. In the early days of Halo, we played split-screen with a 109-degree field of view at 640x240, sent over interlaced composite. That wasn't entirely perfect, but we mostly tolerated it. Now we've got six times that many pixels in our view, a significantly smaller FoV, and we're currently restricted to fairly small maps... and objects being resolved well enough is being regarded as a significant issue.
 
In Halo 5 players are not difficult to see because of lack of resolution, but because of the saturated colour palette used, since Halo 4 the developers have had opted for a Killzone esque approach to colouring where maps/levels have a dominant colour palette...for example in the beta it is easier for Blue team players to blend in Truth than it is for Red team player and the opposite for the other map Empire.

We've had Halo games in the 6th gen era too you know with 480P resolution. Granted Halo 5 looks pretty poor due to low resolution, the maps itself in Halo games aren't large enough for resolution to affect visibility in any significant way. Battlefield is different because the 6th gen era consoles had BF games limited in scope and likewise for the 7th gen era games, and you could see how hard it was in that game to figure out players especially because they'd all blend in, and this is an issue that exists even today if you play on PC because the clothes the soldiers wear look alike for friend and foe and they match the environment quite well.


EDIT: I find it surprising that the article made no mention of Halo 2 Anniversary multiplayer. Those maps were built on an upgraded Halo 4 engine as well and ran at 1080P/60FPS. It would have been an interesting comparison between that and Halo 5 beta. I am disappointed in how DF articles these days lack any depth and any technical commentary is limited to the captions under an odd few screenshots.
 
Last edited:
In Halo 5 players are not difficult to see because of lack of resolution, but because of the saturated colour palette used, since Halo 4 the developers have had opted for a Killzone esque approach to colouring where maps/levels have a dominant colour palette...for example in the beta it is easier for Blue team players to blend in Truth than it is for Red team player and the opposite for the other map Empire.

We've had Halo games in the 6th gen era too you know with 480P resolution. Granted Halo 5 looks pretty poor due to low resolution, the maps itself in Halo games aren't large enough for resolution to affect visibility in any significant way. Battlefield is different because the 6th gen era consoles had BF games limited in scope and likewise for the 7th gen era games, and you could see how hard it was in that game to figure out players especially because they'd all blend in, and this is an issue that exists even today if you play on PC because the clothes the soldiers wear look alike for friend and foe and they match the environment quite well.


EDIT: I find it surprising that the article made no mention of Halo 2 Anniversary multiplayer. Those maps were built on an upgraded Halo 4 engine as well and ran at 1080P/60FPS. It would have been an interesting comparison between that and Halo 5 beta. I am disappointed in how DF articles these days lack any depth and any technical commentary is limited to the captions under an odd few screenshots.

I agree they speak to much about gameplay and not enough about technical side of games.

And when they will use tools to give average framerate... They only give extrême sitâtions...
 
Granted in h5 beta there is 0 AA solution visible compared to bf4. And despite how hard it is to see players have adapted quite well.

Arena shooters are very much based on map control so you should already be looking where your opponents could be relative to you. That and you can't go prone behind foliage.

Bf4 is completely wide open with squads needing to do firing arcs to cover larger areas and people will camouflage behind cover. And lastly time to kill is so much higher in halo, most long range hits provide you ample time to seek cover after being hit.

The situation is not easy on the eye today but does not nearly affect the game as much as one would believe.

I am not talking about MP mode. I

am interested in SP only and hope we get super large scale battles. Here I personally would prefer 30Hz to enable large open scale battles and I am convinced that in this case high resolution is important.
 
I thought the Halo5 beta was kind of hideous when I tried it. Would have been more interested in an article explaining some of the rendering tech, because right now I think it's ugly. I'm guessing they're shooting for some interesting things on the tech side, but it didn't come across as I was playing.
 
Status
Not open for further replies.
Back
Top