Will tiling in 360 allow reaching 60 fps performance ?

[Griffith]

using predicated tiling with MSAA = 5% perf hit
Wo PT with MSAA = 40-45% perf hit

using PT = + 30-35% framerate


so if PGR3 has framerate between 35-45 fps, using PT this become near 45-59 fps


BUT developer can choose to use the extra framerate in other ways, example:

PGR3 wo pt @ 35-45 fps

has two options:

a) PGR3 + pt @ [45-59] fps
b) PGR3 + pt + new effect bandwidh eater @ 35-45 fps

so if dev choose 30 fps, the use of predicated tiling can permit a better IQ, more effects, etc
or the devs can choose to go with same IQ but more fps

 

[Hardknock]

Anyway, I could never give negative points to others for speaking out their (even erroneus) thoughts, so I am in disbelieve that someone gave me bad reputation on something so trivial as this.

Don't sweat it man. Reputation is a popularity contest. I don't see anything that you've said offensive either.


Edit: I also think people rashly give negative feedback before attempting to see what someone was trying to say. A lot of times it's just a misunderstanding.

 

[AlNom]

Don't sweat it man. Reputation is a popularity contest. I don't see anything that you've said offensive either.


Edit: I also think people rashly give negative feedback before attempting to see what someone was trying to say. A lot of times it's just a misunderstanding.

Sometimes for revenge i.e. giving bad rep to some post (the post isn't even remotely bad) just to get back.

 

[overclocked]

Havent even checked witch button you press? Is it the one report bad post?

 

[Titanio]

I don't see how tiling would improve the framerate. It'll allow for better use of the eDram's functionality, but it wouldn't relieve a bound anywhere. It's not like AA without tiling is eating into main memory bandwidth or anything - it just limits the level of AA or resolution.

In other words, lack of tiling isn't so much sacrificing speed as quality/"features", specifically IQ. If anything, tiling could introduce more expense to potentially slow things down, but I guess the idea is that the benefit (as above) is worth it.

 

[Shifty Geezer]

Don't sweat it man. Reputation is a popularity contest. I don't see anything that you've said offensive either.

Different people give rep for different reasons. Some people (like myself) only give bad rep when someone's attitude seems out of line. Other give negative rep when someone's providing wrong information as use the repometer to show not only what the person is like in discussion but also how much useful information they can provide to debates. There's no guidelines on when or how to rep, and there'd be pretty pointless even if there were. Most of the 'how to post FAQ' points are ignored a lot of the time, even on simple things like limiting embedded picture sizes.

 

[dukmahsik]

it's made for tiling, im sure we'll see way better perf once devs put into good use

 

[ihamoitc2005]

Pgr3

using predicated tiling with MSAA = 5% perf hit
Wo PT with MSAA = 40-45% perf hit

using PT = + 30-35% framerate


so if PGR3 has framerate between 35-45 fps, using PT this become near 45-59 fps


BUT developer can choose to use the extra framerate in other ways, example:

PGR3 wo pt @ 35-45 fps

has two options:

a) PGR3 + pt @ [45-59] fps
b) PGR3 + pt + new effect bandwidh eater @ 35-45 fps

so if dev choose 30 fps, the use of predicated tiling can permit a better IQ, more effects, etc
or the devs can choose to go with same IQ but more fps

If PGR3 frame-rate is 35-45 fps at ~576P resolution, 2X AA, then for higher 720P resolution, no AA, frame-rate will be much less because 56% more pixels has much more shader load.

Then for 720P and AA, for AA is additional minimum 5% "hit" because 720P with AA must have tiling for 10MB edram.

Because vertex load is less (less fps) then total shader "hit" and frame-rate hit is maybe little bit less than 56% for average. This is benefit of unified shader design.

So we can only guess but if PGR3 is 35-45fps at ~576P, 2x AA, then maybe 720P with 2x AA has 40% "hit" for 25-32fps. This is just silly guesses but point is very simple. To have real 720P and 2xAA PGR3 frame-rate will be much less.

But maybe PGR3 graphics is not so "efficient" for Xenos GPU so maybe with PGR4 we can have 720P with same effects and frame-rate because of more efficient graphics.

 

[Shifty Geezer]

PGR3 doesn't use tiling. It uses a smaller frame buffer that fits into the eDRAM.

 

[Griffith]

If PGR3 frame-rate is 35-45 fps at ~576P resolution, 2X AA

this is false speculation caming from a reviewer
he get a frame buffer screenshot from a devkit, but bizzarre claim that pgr3 is 720p and everyone can get a framebuffer screen at every resolution, even higher then 1080p
so this does't prove nothing

 

[Guden Oden]

So we can only guess but if PGR3 is 35-45fps at ~576P, 2x AA, then maybe 720P with 2x AA has 40% "hit" for 25-32fps. This is just silly guesses but point is very simple. To have real 720P and 2xAA PGR3 frame-rate will be much less.

This is just completely wrong. PGR's framerate is sluggish because the game is rushed and poorly optimized to the 360's overall design. Not because the graphics is so advanced it has to run at less than 600 lines rez in order to keep framerates up.

Like you said, it's just silly guesses. Using PGR as any kind of indication how fast a 360 game would run in 720P/AA, that's just dumb and bound to fail.

 

[BigGamer X]

this is false speculation caming from a reviewer
he get a frame buffer screenshot from a devkit, but bizzarre claim that pgr3 is 720p and everyone can get a framebuffer screen at every resolution, even higher then 1080p
so this does't prove nothing

The jaggies on my TV seem to indicate it is less than 720P. (it's by far the jaggies XBOX 360 game I've seen, and I've seen pretty much all of them.)

 

[mozmo]

I guess some people are confused as to why tiling would improve performance. Well part of the predicated tiling involves doing a z only pass with all your geometry. This basically populates the z buffer so that when you actually do the real rendering pass, a lot of stuff that is not visable won't actually be rendered and waste pixel shader resources, thx to a populated z buffer which will all polygons/pixels that are not visible. That's why tiling can actually give you a performance boost. Although i'd say a few games already do a z only pass for effeciency so they're liking to not see any improvement due to tiling.

 

[3dcgi]

The z pass is just a performance optimization. It can be done with or without predicated tiling so it doesn't make tiling faster than not tiling.

 

[ERP]

The z pass is just a performance optimization. It can be done with or without predicated tiling so it doesn't make tiling faster than not tiling.

The ZPass is used to fill in the predicates aswell as the ZPass itself.

You would almost always want to do the ZPass when using tiling, without it you will be doing significant redundant transform work, this is not necessarilly true of not tiling, where the ZPass might be a cost.

Not tiling will always be faster clearly....
But there are a number of ways to approach tiling, you could do the simple brute force (no predicate) render the screen twice thing.
Or you could spend the time to do it right, which involves versions of shaders specifically for the ZPass and a datastructure that lets you easilly swap them, and potentially writing transformed data back for subsequent passes if the transform cost is significant enough.
None of it is complicated, it just takes time to try it and time it and see what the best approach is.

 

[ihamoitc2005]

Optimize

This is just completely wrong. PGR's framerate is sluggish because the game is rushed and poorly optimized to the 360's overall design. Not because the graphics is so advanced it has to run at less than 600 lines rez in order to keep framerates up.

Like you said, it's just silly guesses. Using PGR as any kind of indication how fast a 360 game would run in 720P/AA, that's just dumb and bound to fail.

Maybe you are right my friend, I cannot know so this is why I also said this.

But maybe PGR3 graphics is not so "efficient" for Xenos GPU so maybe with PGR4 we can have 720P with same effects and frame-rate because of more efficient graphics.

So if PGR3 shader use is not so efficient then maybe 720P is no problem with same effects for PGR4. Also then if tiling method is used then AA for 720P also possible.

I think many games are not showing true Xbox360 power and I am do not know why this is. Maybe flexible unified shader is unusual for developers so they must learn new "mentality" for this. I dont know.

 

[3dcgi]

The ZPass is used to fill in the predicates aswell as the ZPass itself.

You would almost always want to do the ZPass when using tiling, without it you will be doing significant redundant transform work, this is not necessarilly true of not tiling, where the ZPass might be a cost.

Not tiling will always be faster clearly....
But there are a number of ways to approach tiling, you could do the simple brute force (no predicate) render the screen twice thing.
Or you could spend the time to do it right, which involves versions of shaders specifically for the ZPass and a datastructure that lets you easilly swap them, and potentially writing transformed data back for subsequent passes if the transform cost is significant enough.
None of it is complicated, it just takes time to try it and time it and see what the best approach is.

I agree with what you wrote. The bolded part is what I was trying to say to mozmo as he seems to think tiling is a performance improvement.

 

[ERP]

I agree with what you wrote. The bolded part is what I was trying to say to mozmo as he seems to think tiling is a performance improvement.

I guess the addition of the ZPrepass could be a win, but that would be independant of tiling.

Where I disagree is since Gotham shipped with a none tiled frame buffer it was as fast as it could be.

Most of your graphics performance comes down to arranging you're data optimally for the GPU, no document can usefully describe how to do this unfortunately so you're down to trying things and timing the results, and time to do this is extremly tight on a launch title.

The only conclusion you can draw from a game without a very detailed performance analysis is that the game runs as fast as it does given the way it's written. For example it could be that Gotham was CPU limited and the CPU overhead of their tiling implementation tiling made that worse or it could be that they were short of RAM or it could be that they were simply Fill limited, there is no way to know with the information publically available.

 

[MfA]

You would almost always want to do the ZPass when using tiling, without it you will be doing significant redundant transform work

Significant, but still far less than doing it all twice. It's only going to add around half of the redundant work you do outside of the frustum even without tiling, assuming 2 tiles, which itself is only a fraction of the work. It shifts the balance, but Id say that for a Z-pass to become interesting just because of tiling you would have to be pretty close to the break even point even without it.

 

[3dcgi]

I guess the addition of the ZPrepass could be a win, but that would be independant of tiling.

Where I disagree is since Gotham shipped with a none tiled frame buffer it was as fast as it could be.

Most of your graphics performance comes down to arranging you're data optimally for the GPU, no document can usefully describe how to do this unfortunately so you're down to trying things and timing the results, and time to do this is extremly tight on a launch title.

The only conclusion you can draw from a game without a very detailed performance analysis is that the game runs as fast as it does given the way it's written. For example it could be that Gotham was CPU limited and the CPU overhead of their tiling implementation tiling made that worse or it could be that they were short of RAM or it could be that they were simply Fill limited, there is no way to know with the information publically available.

I'm a little confused as to what you're disagreeing with. I wasn't trying to comment on PGR and say that a z pre pass would help it. My only point was that tiling is not a performance improvement (which you stated yourself). Ideally tiling wouldn't ever be necessary, but edram cost money so...