News & Rumors: Xbox One (codename Durango)

Yes, things like that are somewhat good at predicting the direction someone might be looking.

The problem here is you need to know exactly where someone is looking, not just the general direction.

Regards,
SB

 

I doubt that would matter too much. That hasn't been an issue for 3D gaming after all. It would look worse than crap though...it'd be straight up weird since you'd see low res, blurry fidelity all over the place except for some circular region darting all over the screen to follow the player's eye.

I doubt the Tobii system they are tracking the eyes with there is using a depth camera anyhow. Seems like a strange approach to take in trying to track what are largely 2D eye movements. I don't see a reason why it couldn't work via Kinect 2.0 by just running image analysis on a zoomed in RGB feed, except for the fact that in dark conditions this would obviously not work.

Again though, the eye tracking angle here isn't relevant to my point. I'd really rather discuss ppl's ideas for implementing the display planes in interesting ways or the potential for performance gains there.

Supposedly the display planes stuff is hardware in Durango to do exactly that.

I think you misunderstand why I posted the foveated stuff. I posted it because in both that case and the display planes stuff we've heard about in Durango MS is leveraging 3 different layers that dynamically adjust their visual fidelity across a host of parameters to net major performance gains. At least in concept.

I didn't mean to suggest that the display planes stuff works via embedding rectangles or any other regular shapes within one another. In the foveated rendering setup that is only being done to replicate how the human eye works (it's actually more elliptical than straight up circular but they ignored that for the study).

I see no reason to think that the display planes in Durango would be limited to displaying rectangles. Otherwise it wouldn't make sense to have foreground/background distinctions as no modern 3D games have sharp rectangular boundaries on the screen separating foreground/background. Same goes for the HUD or even OS overlays. Those won't always be rectangles either. So we don't need to limit ourselves as you have suggested here. When these 3 distinct sets of imagery get composited/blended together they can then be upscaled to fit the tv as necessary.

 

Bogus, at least for Microsoft. There was never a plan for a 2012 console launch.

 

Sounds like something that could be done pretty easily on the GPU pretty fast don't know why they would put money towards dedicated silicon for it.

 

Interesting. Thanks for the reply. If it was true, it would have to be Sony he was talking about.

I guess its all moot now anyway!

 

The issue with eye tracking, is you need to

A. Be able to see the pupils
B Have pupil movement make a significant change to the image.

Think about how many pixels you're pupils cover on a 1080P image when the camera is trying to take a wide angle shot to do motion tracking.

 

When I heard of the display planes, I thought of the following.

1. UI, OS, windows for skype, social medial. ... 1080p 30 fps updates but only renders changes.

2. Background - i.e. anything that isn't animated 720p 30 fps, high IQ

3. Animation frame - ultra high poly characters and other things with animation, e.g. master chief and all enemies, 1080 p or higher, 60 FPS or high, ultra high IQ and detail.

Overall output is 60 fps, 1080p. with overall graphic card burden less than 1080p, 30 fps, current levels of IQ.

Hopefully, we get the display planes tomorrow from VGLeaks.

 

I don't see why you couldn't do this now with current consoles then again I'm not a graphics dev

 

remedy actually did something like this for Alan Wake. The results were quite awesome. The idea is to make happen in hardware and have it optimized, by using say tiles.

 

The question how much faster it is in dedicated hardware, and I'm not convinced that it would be that much faster then just using the GPU straight up the GPUs in both machines are plenty capable of blending variable sized buffers very quickly.

 

I think the Tobii eye tracker is more sophisticated than that. It bounces near infrared light off your eye and analyzes the reflection. I read that on their web site around CES time.

 

I don't think the point is whether it's possible or not, but rather the minimum impact on development this brings.

From the patent, for each of those planes the developer could set up a fixed resolution, or a fixed refresh rate, and the system would automatically scale the other to achieve this goal. The advantage being that they would always have dynamic resolutions depending on load, rather than being restricted to a single resolution throughout, which, going by the number of current gen games that are sub hd but use dynamic scaling, is not a trivial task.

 

You could but that would require the graphics card to do additional work.

If we assume the 3 planes that some are speculating, then the graphics card would have to...

Upscale the background plane to 1080p and hold that in memory.
Upscale the middle plane to 1080p and hold that in memory.
Composite back, middle, and foreground planes into one image and then output that to the display.

It could be that as you upscale you could also composite the different planes, I'm not sure. Either way, it's going to be resource intensive.

Regards,
SB

 

Something that isn't trivial this gen could be easily achievable next gen I don't see his providing a lot of benefit over doing it straight on the rumored GPUs and it adds cost and complexity to the design aswell.

Aren't current gen games already doing this just with only two buffers ? (usually particles I believe)

 

Particles I'd imagine would be rather trivial to upscale. At some point someone with more knowledge of 3D rendering may pop in and give more detailed information on what would likely need to be done.

Regards,
SB

 

Ah okay thanks for that . I'm now wondering where this device read/writes from/to because it's going to need 30/20/10MB space roughly depending on how you partition it.

 

What I really don't get about the vgleak rumor is the 102GB ESRAM speed. As the ESRAM is on the APU it should also be connected to the internal GPU data bus. An old GPU from 2009 had a 1024Bit crossbar and 4 memory controllers. So somebody calculated 800Mhz Core * 1024 = 102GB. I couldn't find such detailed images for Southern Islands though.

But today a 7970HD with a memclock of 1500Mhz with 384Bit Bus(6 Memory controllers) would completely overload such a crossbar. Even a GPU with 256Bit GDDR5 should easily overload it. So I can only assume that the crossbar itself is wider than 1024Bit in the current generation and/or runs with a higher clock than the GPU core clock.

So why should they limit the ESRAM connection to the crossbar?

 

How do you know what MS wanted? The leaked doc is just an old doc, we know (thanks to shockingAlberto in NeoGAF) that MS listened to EA, Activision, DICE, etc they want power.

 

Do you have the link to that comment?

EDIT
Found it
http://www.neogaf.com/forum/showpost.php?p=36353477&postcount=9316

But this was from a Year ago and if the Durango specs are real than this comment was bullshit.

 

I think there's an element of truth to them.

The original spec MS stuck to pretty solidly... and Durango is likely within the range they quoted (as is the CPU I think), with one notable exception.

In Dec 2011 there was the "reliable" pastebin leak. In terms of memory this would seem to put the consoles somewhere around these rough numbers:
Orbis 2GB RAM, 0.5-1GB O/S, 1-1.5GB for gaming.
Durango 4GB RAM, 2.5-3GB O/S?, 1-1.5GB for gaming.
Wii-U 2GB RAM, 0.5-1GB O/S, 1-1.5GB for gaming.

At some point after the leak (and probably after Nintendo had finalized the Wii-U) both Sony and MS decided to increase their memory size... it seems entirely possible this was at the request of 3rd party developers.

(personally, the more interesting part is what pastebin didn't say - what type of RAM was in Durango prior to the re-spec?)