Digital Foundry Article Technical Discussion Archive [2014]

http://www.eurogamer.net/articles/digitalfoundry-2014-thief-next-gen-face-off

 

Thank you sebbbi for the info. So, these temporal upscaling algos are based on Video compression principles, in extremely simple terms. This is pure news to me, I had no idea we were already using previous frame datas in games to reduce rendering loads. I thought it would be too compute expensive as video encoding does take a lot of time and horsepower. Is such stuff happening in Frostbite too? saving rendering costs by analysing previous frames? This is whole new topic to explore !

In this case it did work. I play on the same monitor as always and it was nowhere blurry like a normal upscaled game. I just thought it looked different due to their optimisations. Didn't notice it at all. Did notice the artifacts, but had no idea it was due to halved horizontal res. Wish they give a complete talk during GDC, but considering this is highly volatile topic, I don't think we will see it.

 

I think there is some misunderstanding. DF says this effect (960X1080 > 1920X1080) "is not cheap from computational perspective". This means that another half of the frame (960 x 1080) is not directly copied and pasted from previous frame. If KZ really only updates 960 X 1080 and copies another 960 X 1920 from last frame then it is 1920i X 1080.

The only possible solution is that the other half of the frame is "interpolated" from current and last frame so this effect is not cheap. It's not directly upscale. It is more like "frame interpolation" discussed in console technology forum. The difference is that what we discussed is to update a whole frame 1920 X 1080 every 1/30 second and interpolate another complete frame to create 60 frame per second, while KZ updates 960 X 1080 (1/2 frame )and interpolates another half.

 

Exactly, I think even DF mentioned that it has a temporal aspect, ie. using information from other frames too.

 

Semantics, but it just sounds like a smarter deinterlace, judging by sebbbi's input. The artefacts are what they are. *shrug* I'm open to a name that doesn't take a world of explanation though.

Temporal field interpolation? Field Interpolated Temporal Reprojection?

Shifty's Bitch's naughty lace interpolation?

 

Obviously this should be the new official name for this technique from now on, on this forum.

 

What I don't understand is how suddenly people can claim this resolution looks blurry but it is somehow significantly better than 720p at the same time? A game like Ryse at 900p would not be noticed if pixel counting on zoomed in screen captures did not exist. That is why this wasn't noticed. Not because interlaced half resolution looks objectively better. It's because it was a technique that people weren't expecting and there wasn't someone to zoom in and count the pixels.

I think I'm mostly frustrated by the fact that there is a resolution hyperbole mostly by uninformed masses and it makes games look worse than they otherwise could. I swear everyone should just run Quake 3 at 4k and be done with it. Image quality in Ryse being better than Forza despite lower resolution should say something.

 

So destroy every image equally rather than make the best image based on temporal information? Frame rate aside (we don't know if it would be better), the image quality for 720P would be worse since it would be in every frame and not depend on motion.

Ethics aside (this is a tech forum), they found a method to improve the frame rate for MP. There are many ways to do it, this was one innovative way. We might see more of it, sure seems beats the over used 'drop the res' club devs have been swinging far too much in the past.

 

Forza is a 60fps game and Ryse basically was built to be a graphical showcase. Not a surprise that it does look good. Any way the importance of resolution varies on the type of setup one is gaming on. Larger the perceived screen size becomes the resolution also becomes more important. I personally prefer clarity in the image quite a bit and resolution helps there, especially on a large display. I hate all sort of blurs, whether it's due to low resolution, some post AA method, artistic choice or flaws in the display tech.

 

It's certainly something I'd like to see compared to a naive 1440/1536/1600 x 1080 (normal upscale, no temporal AA) since the increased blur was still noticeable to many a folk.

(Of course, they could have just stuck with 1080p30 as originally intended, and their backtrack with the patch for SP does kind of say something about their decision making post-E3.)

 

The world would be a better place without pixel counters

 

Indeed.
Grab you guns people, let's hunt some pixel counters

 

No, it'd be better if everyone were more interested in the technical aspects than console warring over scraps.

 

Maybe they were worried that 30fps was not possible to maintain with an unpredictable number of players shooting and exploding all in one area. Maybe the trade off was hard to detect interlacing trick with 35-50fps versus occasional sub-30fps in 1080P. DF would have a field day showing the worse case scenario in MP and the Internet would say nothing but 25fps KZ.

 

If they complained that the game looked blurred they actually saw that something was wrong, just because they didn't pinpoint the exact resolution or reason doesn't make them wrong in thinking something was wrong, right?

Yeah, shout a bit for ME as well.. fucking stupid world!

 

The Order 1886 also has a blurry look. Could it be doing something like this as well?

 

There's been no similar artefacting to date (including temporal ghosting). It's probably just their post-processing chain.

 

It's not wrong, it's just hypocrisy. If people gonna hammer on the 1080p debate, they need to have ONE STANDARD for ALL, not just when it's convenient.
It's even more hilarious that after like 3 month, people just think it's "blurry" but since they were told it's 1080p, they think it's something else (FXAA).

Imagine what kind of shitstorm would it be if CryTek had claimed Ryse to be 1080p, but one must ask, can people actually tell?

 

Excellent point. I completely agree with that. The DF article is fine –truth be told he missed the POM thingy being present on PS4, but they also said that PS4’s version never dips below 25 when it can be seen clearly in the video it dips to 21 right away, so give and take.

Still, if they weren’t judgmental they wouldn’t have a problem. But now they got GAF’s Seventh Cavalry against them, the knight errantry of justice by definition.

But seriously, the Digital Foundry Verdict does more harm than good, for all the parts involved.

It can also seriously affect sales.

You read my mind. I was about to post almost exactly the same, word by word. Ryse looks better than Forza -DrEvil is right though-, and better than Fifa 14, and better than NBA 2k14, and Crimson Dragon, and Halo: Spartan Assault, and NBA Live 14, and Need for Speed Rivals… etc etc.

In addition, I am not a big fan of UE3, never have been. It has never been console friendly and the games looked very similar for the most part, whatever the reason was. Give me CryEngine for next gen consoles any time of the week.

Still, the 1080p lovers seemed to have it under control, but… woah, they were waaaaaaay off....

 

It'd be better if people didn't care about a number but what things looked like on screen. Those of us who discuss the techniques do so purely from an analytical POV to understand how the developers approach the issue of working on finite hardware. Every game is a compromise; the interest comes from seeing which compromises are used and how, especially when new ones are developed (like Lair's alternative AA resolve).

I'm with Al. Looks like lens simulation as there are no pixel-perfect artefacts from what I've seen.