Xbox One (Durango) Technical hardware investigation

Status
Not open for further replies.
The textures will be read from DDR3 though. A few inconsistent textures probably speaks more of time constraints or something else rather than memory bandwidth. I can't see how memory bandwidth would be the concern. The game is 60FPS though.

The design on top of the car looks softer because it's at a sharper angle. It seems fine at other times. Same texture, different angle and lighting.

I agree with you the bulk of textures would be read from DDR3, while textures reused every frame like those of your car might be stored in esram given the framebuffer would likely not be using the full 32MB. Perhaps a full 2kx2k set of textures could not fit in the remaining esram for this car model. Yeah you have to give the game a break on memorybandwidth since its running at 60fps.

No LOD just refers to polygon models, does not mean no mipmapping ;) Yes in the prerace ingame cinematic you see various angles of the car, but come on thats one of the oldest developer tricks in the 3D rasterizing book. They could easily use a different set of textures given the gigs of memory int he system. Its similar to blendshapes the are rendered in the ingame Ryse cinematics creating incredibly realistic skin, and toned down during the actual gameplay. All one LOD for Marius also.
 
Last edited by a moderator:
It does look like the filtering, not lower quality textures than are present for the rest of the car in-game, only affecting textures viewed from steep angles. That'd be one hell of an optimisation though.
 
Last edited by a moderator:
It does look like the filtering, not lower quality textures than are present for the rest of the car in-game, only affecting textures viewed from steep angles. That'd be one hell of an optimisation though.

Lets hope you are right.
 
It does look like the filtering, not lower quality textures than are present for the rest of the car in-game, only affecting textures viewed from steep angles. That'd be one hell of an optimisation though.
Thats how mipmap calculation usually works, anisotropic filtering is the solution for that exact problem.
 
I agree with you the bulk of textures would be read from DDR3, while textures reused every frame like those of your car might be stored in esram given the framebuffer would likely not be using the full 32MB. Perhaps a full 2kx2k set of textures could not fit in the remaining esram for this car model. Yeah you have to give the game a break on memorybandwidth since its running at 60fps.

No LOD just refers to polygon models, does not mean no mipmapping ;) Yes in the prerace ingame cinematic you see various angles of the car, but come on thats one of the oldest developer tricks in the 3D rasterizing book. They could easily use a different set of textures given the gigs of memory int he system. Its similar to blendshapes the are rendered in the ingame Ryse cinematics creating incredibly realistic skin, and toned down during the actual gameplay. All one LOD for Marius also.

Most recent footage shows ingame cars have simplified, or non existing interior elements, so pre-race you will see a 30 polygon bolt in closeup, but then, when you see your full car, it doesn't even display the correct dashboard. So I am pretty sure that, like all forza games before it, the ingame car, the one that you control, is of significant lower detail, thus proving the game has different sets of geometry and detail in memory.
 
Some of the car textures may be stored in the esram since the framebuffer wouldn't take up the whole 32MB cache and they are reoccurring every frame. Unlike Ryse which is 1600x900 its possible that texture limitations had to be made because the framebuffer is slightly larger in Forza's case running at 1080p and there is only 16MB or less remaining to store the cars data.

It would be wise to store the cars data in the esram in order to elevate memory bandwidth requirements on the main memory.
That's just crazy talk. The ESRAM isn't a static buffer to store a few MBs of much-used assets. It's the full working space of the GPU. It'll hold multiple render targets being worked on and composited. Of course, devs could choose to stick a few static car models in there if they so desired, but none with any sense would completely underuse the hardware that way!
 
Its similar to blendshapes the are rendered in the ingame Ryse cinematics creating incredibly realistic skin, and toned down during the actual gameplay.

What exactly do blendshapes have to do with realistic skin rendering?
Or you've meant facial animation?
 
you can store whatever you want in ESRAM, including compressed data. Graphics already used compressed data natively in the form of DXTC compressed textures, for example. However, for speed purposes you wouldn't want to store much compressed data in there beyond what can be natively worked with via the hardware (compressed textures).
 
Thats how mipmap calculation usually works, anisotropic filtering is the solution for that exact problem.

Not the same as just having lower resolution textures for car tops in-game. Just a little more AF could be memory bandwidth limited?

It is a 60fps game.
 
you can store whatever you want in ESRAM, including compressed data. Graphics already used compressed data natively in the form of DXTC compressed textures, for example. However, for speed purposes you wouldn't want to store much compressed data in there beyond what can be natively worked with via the hardware (compressed textures).

Must elaborate my question
I'm reading that 32MB are not enought to store the full framebuffer with all that is necessary to make it work (and that I don't understand) and achieve a 1080p resolution.
Can the framebuffer's tiles, maybe on a 30fps game, be compressed to store all in the esram?
Id'll like a suggestions to study myself how modern framebuffers works.
 
Can the framebuffer's tiles, maybe on a 30fps game, be compressed to store all in the esram?
No. I don't think GPUs can write compressed texture data; only read it. Even if you could, you'd lose quality if they were compressed. FBs have to be in the best, native quality for all subsequent blend and composite operations. We already have some forms of compression via the pixel format, such as MS's 10 bit per colour float format. These produce smaller buffers with lower quality. ESRAM is enough for a 1080p framebuffer. What it's not enough for is some deferred rendering buffers, and deferred rending is all the rage. That's mean some jobs will have to be handled piecemeal, or devs will have to come up with new approaches.
 
I think this works here

http://www.businessinsider.com/xbox-one-command-and-gesture-cheat-sheet-2013-11

xbox%20commands.png


xbox%20gestures.png
 
Making a selection by tapping is a massive improvement over holding still for a moment. I hope they got that to work well. :smile:

The zoom gesture seems sensible too.
 
The long list of specific voice commands is interesting. People are either going to memorize it all or just skip it completely and use the controller. Will be interesting what the breakdown is in a years time. I for one can never remember the basic voice controls for my phone and thus don't use them.
 
The most useful commands are the most obvious. For turning it on, launching content/apps, stop/pause/play, searching for content... Those should stick because they'll beat the controller much of the time. "Show my stuff" too.

"Xbox pause/play" certainly beat powering up controller up to press the A button on the 360.
 
The long list of specific voice commands is interesting. People are either going to memorize it all or just skip it completely and use the controller. Will be interesting what the breakdown is in a years time. I for one can never remember the basic voice controls for my phone and thus don't use them.

Theres gonna be a lot of people yelling things out at their screens this holiday...

I bet its gonna be a meme :LOL:
 
As with muscle memory it will just click once you do it enough times. In other words the most used voice commands/functions will automatically click over time. There's really no complicated learning process. When you pick up a phone you say "hello"...it's no different.

Anyway the hand gestures are a nice compliment to the voice control...it's like navigating an Ipad without actually physically touching it. I'm thinking this might be integrated into a future Ipad given that Apple acquired Prime Sense recently.
 
Last edited by a moderator:
As with muscle memory it will just click once you do it enough times. In other words the most used voice commands/functions will automatically click over time. There's really no complicated learning process. When you pick up a phone you say "hello"...it's no different.

Anyway the hand gestures are a nice compliment to the voice control...it's like navigating an Ipad without actually physically touching it. I'm thinking this might be integrated into a future Ipad given that Apple acquired Prime Sense recently.


More like the future Apple TV.
 
Status
Not open for further replies.
Back
Top