Xenos and its special features: Educated expectations needed READ: post #126!

[liolio]

I've just red again these threads :
http://www.beyond3d.com/forum/showthread.php?t=28500&page=2&highlight=kameo
http://www.beyond3d.com/forum/showthread.php?t=29521&highlight=kameo
http://www.beyond3d.com/forum/showthread.php?t=28830&highlight=kameo
http://www.beyond3d.com/forum/showthread.php?t=37564&highlight=crackdown
http://www.beyond3d.com/forum/showthread.php?t=33423&highlight=lost+planet

Everybody here is wandering about cell capacities, as a matter of a fact, I think we have far less shared knowledge about what the xenos is able to bring on the table and with witch kind of performance.

In the aforementioned thread I've find interesting links and quote, but sadly most of them were quiet old.
I guess everybody should have a quick look before posting in this thread.

The goal of this topic is raise educated expectations for the future.


Now we're starting to see some games doing "strange stuffs" with xenos or other games using edram logic in a weird way to achieve better performance with particles.

What can we expect in the futur?

Can we know if the last havoc release bring something on the table in the these regard?
think at this (quiet old) :
http://www.havok.com/content/view/187/77/
Will MS push Havoc to put together a middle ware that would take advantage of ms hardware in a more balanced way between cpu and gpu calculations.

Everybody speak about Some big news coming from sony at GDC, where is MS at this point do we have rumors coming from insiders?
Will Ms come at gdc with empty hands or with huge improvement in this regard, directx10 it's here now, we see gpgpu applications coming ( not that I think that xenos is comparable in power with say a g80). I hope Ms will have something interesting to bring on the table, say some grat tech demo could be a good marketing move.

I know we do very few 360 devs here to enlight us but I still hope some insight.
I'm still under the impression that xenos is used in a fairly conventional way.
Anyway automatic tiling only makes its way to devs during autumn and still doesn't allow memexport.

 

[Love_In_Rio]

Has any developer used already memexport ?

John Carmack started to use it but it seems he finished giving it up. Maybe Fable 2 will make use of any exhotic way of programming, Frank ?

 

[almighty]

I really cant see many developers using memexport, i remember correctly if a developer decides to tile render to achieve 4xMSAA it disables the option to use memesport? So a developer can either have one or the other and i see them favouring the 4xMSAA TBH.

 

[ShootMyMonkey]

I really cant see many developers using memexport, i remember correctly if a developer decides to tile render to achieve 4xMSAA it disables the option to use memesport? So a developer can either have one or the other and i see them favouring the 4xMSAA TBH.

Why would you think that? All else aside, I'd see devs favoring NOT using hardware MSAA because predicated tiling can be a source of headaches. And I don't mean specifically that tiling in itself is problematic, but that having to alter the way you issue geometry can be disagreeable to a lot of things about the way particular engines might be structured -- much to the point that it can be a deal-breaker. There's no such thing as "just put in support for it."

The way I see it, memexport is a convenience if you have a use for it. Whereas tiling is an inconvenience, even if it might sometimes be an insignificant one. Especially considering that there are many ways to avoid having to rely on hardware MSAA or at least tiling, which is probably the preferred thing to do if Xbox360 isn't the only platform you have to worry about. Building the engine around that is probably fine if you're a 360-only developer. The requirement is only to get an equivalent effect to that of 2xMSAA, so most people would normally prefer to cheat with something that works out on all platforms.

 

[Rangers]

Why would you think that? All else aside, I'd see devs favoring NOT using hardware MSAA because predicated tiling can be a source of headaches. And I don't mean specifically that tiling in itself is problematic, but that having to alter the way you issue geometry can be disagreeable to a lot of things about the way particular engines might be structured -- much to the point that it can be a deal-breaker. There's no such thing as "just put in support for it."

The way I see it, memexport is a convenience if you have a use for it. Whereas tiling is an inconvenience, even if it might sometimes be an insignificant one. Especially considering that there are many ways to avoid having to rely on hardware MSAA or at least tiling, which is probably the preferred thing to do if Xbox360 isn't the only platform you have to worry about. Building the engine around that is probably fine if you're a 360-only developer. The requirement is only to get an equivalent effect to that of 2xMSAA, so most people would normally prefer to cheat with something that works out on all platforms.

How do you "cheat" on Xenos AA? Are you meaning shader based AA or something?

Is it even possible for 360 to do normal AA, aka like any normal GPU? Bypass the hardware AA?

And I thought all 360 titles used tiling? Isn't it necessary at 720P?

It really seems to me that tiling is the real shame of Xenos. It seems to be performing at par with RSX anyway, so one wonder if it couldn't gain even more perfomance if it didn't have to rely on tiling. What a tidy little piece of kit Xenos is. 170mm^2 or something like that.

 

[Cheezdoodles]

Is it even possible for 360 to do normal AA, aka like any normal GPU? Bypass the hardware AA?

No, because the ROPs are on the eDRAM i believe..

And I thought all 360 titles used tiling? Isn't it necessary at 720P?

You do not need tiling for neither 720p nor 1080p. You need tiling if your going to have MSAA.

It really seems to me that tiling is the real shame of Xenos. It seems to be performing at par with RSX anyway, so one wonder if it couldn't gain even more perfomance if it didn't have to rely on tiling.

Actually, when we actually begin to see games that use tiling, it will be a performance gain. Forza 2 will be the first game that will use tiling for the X360. (Maybe im missing a game or something), anyways, the eDRAM if used correctly, is brilliant, it basically means "free" AA (or AA at a VERY low cost). Or you can use it to do other heavy bandwidth operations.

The Xenos, is imo, not only on par with the RSX, its much more sophisticated than the RSX.

I do belive, RSX will have a BIG problem doing 4x AA @60fps, because its to low on bandwidth. (Forza 2, the first game that uses tiling on the X360, will do just this)

 

[ShootMyMonkey]

How do you "cheat" on Xenos AA? Are you meaning shader based AA or something?

Not necessarily shader-based AA, but also just using other things like Depth-of-field and motion blur and bloom or slight under-resolution rendering and scaling up with a really soft sampling kernel -- basically things which muddle the sharpness enough that it can pass for looking like it has AA most of the time.

Is it even possible for 360 to do normal AA, aka like any normal GPU? Bypass the hardware AA?

I don't follow what you mean by "normal AA"... Hardware AA IS the "normal" AA, but there's no rule that it has to be turned on. Having to use tiling is not inherent of Xenos' AA, but that enabling hardware AA means needing more space than the eDRAM provides, so you end up having to do it by tiles.

And I thought all 360 titles used tiling? Isn't it necessary at 720P?

Absolutely not. It's only necessary if you enable AA at 720p. Without any level of AA, there's enough room in 10 MB to hold full-size color and Z at 720p (which occupies about 8 MB assuming 32-bit color).

You do not need tiling for neither 720p nor 1080p. You need tiling if your going to have MSAA.

Ummm... you do need it for 1080p. 1080 means 2 mpixels for a full framebuffer, meaning framebuffer + Z takes up 16 MB. Even with Z-compression, you still need 12 MB. So tiling is a given for 1080p. Anything that pushes you over the size of the 10 MB of eDRAM. e.g. 720p with an FP16 framebuffer and 32-bit Z will also demand tiling.

It really seems to me that tiling is the real shame of Xenos. It seems to be performing at par with RSX anyway, so one wonder if it couldn't gain even more perfomance if it didn't have to rely on tiling. What a tidy little piece of kit Xenos is. 170mm^2 or something like that.

Well, it's kind of the inconvenience that occurs when you can't make tiling transparent to the user. When Microsoft tells you "5% hit," they're basically just talking about the hardware side (e.g. re-processing some small number of verts for triangles that cross tile boundaries)... and that side of the statement is probably true. Of course, that's assuming that everything else you might have to do doesn't incur any cost whatsoever.

Or you can use it to do other heavy bandwidth operations.

I'm not entirely sure what you're thinking of here... Alpha blending?

 

[Graham]

Actually, when we actually begin to see games that use tiling, it will be a performance gain. Forza 2 will be the first game that will use tiling for the X360. (Maybe im missing a game or something), anyways, the eDRAM if used correctly, is brilliant, it basically means "free" AA (or AA at a VERY low cost). Or you can use it to do other heavy bandwidth operations.

I'm pretty sure there have been a load of games that already do use antialiasing at 720.

Ummm... you do need it for 1080p. 1080 means 2 mpixels for a full framebuffer, meaning framebuffer + Z takes up 16 MB. Even with Z-compression, you still need 12 MB. So tiling is a given for 1080p. Anything that pushes you over the size of the 10 MB of eDRAM. e.g. 720p with an FP16 framebuffer and 32-bit Z will also demand tiling.

Just wondering... Z compression,
my understanding is Z/Colour compression is done on traditional gpu's to reduce bandwidth demands, by sending the MSAA samples in a chunk, so if they are all the same, only one gets sent (hence 4:1 compression for 4xaa?) ... ?
Or have I just read something wrong and assumed something wrong? Is there another form of z compression?

I was under the impression xenos had no z/colour compression hardware as the bandwidth to edram was enough to not bottleneck...

I'm confused now

[edit]

or are you simply talking about a 16bit z buffer? ;-)

 

[djskribbles]

I do belive, RSX will have a BIG problem doing 4x AA @60fps, because its to low on bandwidth.

i'm not really much of a tech geek, but doesn't the 10mb of edram on the sister core of Xenos limit the nearly free 4xAA to 720p resolution? or in other words, wouldn't the 10mb not suffice for 4xAA on resolutions higher than 1280x720? and wouldn't RSX not really have a problem with 4x AA @60fps at 1280x720? maybe at 1080p resolution, but (if correct) because the edram is limited to 720p, Xenos would struggle with 4xAA at 1080p as well, wouldnt it? and doesn't the fact that RSX can offload things to Cell account for something?

 

[inefficient]

Doesn't the 10mb of edram on the sister core of Xenos limit the nearly free 4xAA to 720p resolution?

Both 2xAA and 4xAA at 720p require tiling on the 360.

There is an official B3D article right here on this:
http://www.beyond3d.com/articles/xenos/index.php?p=05#tiled

2xAA can be done at 1024x600 and fit within the 10MB eDRAM. Many early games with AA are rendering at this res.

 

[Jesus2006]

2xAA can be done at 1024x600 and fit within the 10MB eDRAM. Many early games with AA are rendering at this res.

Not only early but also some recent ones seem to be doing this (like CoD3, Tony Hawks).

I do belive, RSX will have a BIG problem doing 4x AA @60fps, because its to low on bandwidth. (Forza 2, the first game that uses tiling on the X360, will do just this)

Isn't RR7 already rendering at 1080p with 60fps + AA (not sure what level though)? It depends on the game if you can reach this AA level imo - regardless of the console and it's definately easier for a straight forward racing game than a first person shooter.

 

[Cheezdoodles]

Ummm... you do need it for 1080p. 1080 means 2 mpixels for a full framebuffer, meaning framebuffer + Z takes up 16 MB. Even with Z-compression, you still need 12 MB. So tiling is a given for 1080p. Anything that pushes you over the size of the 10 MB of eDRAM. e.g. 720p with an FP16 framebuffer and 32-bit Z will also demand tiling.

My bad, the Xenos article, is ofcourse talking about the framebuffer being used for 1080i not 1080p!! I forgot people use interlaced ghetto resolutions still

I'm pretty sure there have been a load of games that already do use antialiasing at 720.

I can think of oblivion. Thats it

 

[djskribbles]

Both 2xAA and 4xAA at 720p require tiling on the 360.

There is an official B3D article right here on this:
http://www.beyond3d.com/articles/xenos/index.php?p=05#tiled

2xAA can be done at 1024x600 and fit within the 10MB eDRAM. Many early games with AA are rendering at this res.

forgive me for my ignorance, but what is tiling?

 

[inefficient]

Isn't RR7 already rendering at 1080p with 60fps + AA (not sure what level though)? It depends on the game if you can reach this AA level imo - regardless of the console and it's definately easier for a straight forward racing game than a first person shooter.

There is 0x AA in RR7 at both 1080p and 720p. It's mostly 60fps but it does drop at times due to some pathologically bad situations involving particle effects.

Forgive me for my ignorance, but what is tiling?

Tiling is when you only fully render a portion of a frame/screen at a time.

If you don't have enough eDRAM to fit the whole AA frame in memory, then you can try only rendering half. That half of a screen is called a tile. Once the first half is completed, you transfer it to main memory, then you render the other half. It might be necessary to use 3 or 4 tiles depending on the resolution and level of AA. It sounds simple, but it has significant consequences.

 

[djskribbles]

thanks.

kind of off topic, but can RSX do HDR + AA? or only 1 or the other? i also heard that the 360 can only do FP10 HDR whereas the PS3 can do HDR16; is this true?

edit: nm, i think i found my answers with the search function =].

another question though...can Xenos do FP16 HDR + AA?

 

[Acert93]

2xAA can be done at 1024x600 and fit within the 10MB eDRAM. Many early games with AA are rendering at this res.

​

How many is many?

What games?

As far as I know of the early games only PGR3 has ever been identified as using this technique.

Not only early but also some recent ones seem to be doing this (like CoD3, Tony Hawks).

This still hasn't been verified one way or the other, so it is nice to see you qualifying this with "seem" :smile: Browsing all the media sites you will find many various screen sizes. While certain resolutions are fishy due to the framebuffer issue, this isn't absolute proof either. You would think that someone who cared would take a grab to substantiate this.

Seeing as how PGR3 sticks out like a sore thumb with 2xMSAA I do wonder about the other titles as nothing I have seen in them (albeit little) sticks out as jarringly as some of the tell tale signs seen in PGR3.

(Personally I don't really care as what ends up on the screen is the most important aspect, but that is irrelevant to your point).

I can think of oblivion. Thats it

There are quite a few games using tiling as well as MSAA at 720p.

NBA Street and Viva Pinata immediately come to mind and if I remember correctly Splinter Cell was also confirmed right here on B3D to as well.

 

[Fran]

Has any developer used already memexport ?

John Carmack started to use it but it seems he finished giving it up. Maybe Fable 2 will make use of any exhotic way of programming, Frank ?

It's Fran

Yup, I know of devs using memexport. It has some bugs though and it's not always straighforward to use, but it can potentially be a win in certain scenarios.

You can use memexport when a title is doing PTR, just make sure you don't issue a memexport primitive inside a PTR bracket, just before or after, depending on what you are trying to do.

In fact, a typical frame with MSAA and memexport would be rendered something like this on 360:

- Submit all memexport primitives
- Render all shadow buffers
- Render all offscreen textures
- Begin predicated tiling
- Submit the main scene
- End predicated tiling (actual tiled rendering happens here)
- Render particles (?)
- Post processing (tonemapping, bloom, dof, whatever)
- Swap
- Pub

It's important to notice that "begin predicated tiling" is a kind of barrier that logically divides the engine flow, hence the shadow buffers are all prepared before rendering the scene for example. This introduces several architectural issues to deal with when using PTR. On another platform you would probably render a shadow buffer, render the part of the scene using it, then render another shadow buffer and so on, in order to reuse shadow buffer memory. Or any other combination that suites the engine. On the 360 you are pretty much forced to that scheme if you want to use PTR. The main benefit is you get MSAA more or less for free if you solve the problems related to resubmitting the same primitive for each tile.

Fran

 

[Rangers]

​

How many is many?

What games?

As far as I know of the early games only PGR3 has ever been identified as using this technique.



This still hasn't been verified one way or the other, so it is nice to see you qualifying this with "seem" :smile: Browsing all the media sites you will find many various screen sizes. While certain resolutions are fishy due to the framebuffer issue, this isn't absolute proof either. You would think that someone who cared would take a grab to substantiate this.

Seeing as how PGR3 sticks out like a sore thumb with 2xMSAA I do wonder about the other titles as nothing I have seen in them (albeit little) sticks out as jarringly as some of the tell tale signs seen in PGR3.

(Personally I don't really care as what ends up on the screen is the most important aspect, but that is irrelevant to your point).



There are quite a few games using tiling as well as MSAA at 720p.

NBA Street and Viva Pinata immediately come to mind and if I remember correctly Splinter Cell was also confirmed right here on B3D to as well.

I am pretty sure the two Activision games (Tony Hawk, COD3) are all but confirmed to be at 600p (although it didn't stop most review from calling it prettier than the PS3 rev). As well as PGR3.

 

[Jesus2006]

This still hasn't been verified one way or the other, so it is nice to see you qualifying this with "seem" :smile: Browsing all the media sites you will find many various screen sizes. While certain resolutions are fishy due to the framebuffer issue, this isn't absolute proof either. You would think that someone who cared would take a grab to substantiate this.

Well i don't think these reports especially for these 2 games came out all of a sudden (and not for other games despite many had downscaled screenshtos available). There were several reports about this "issue" around the web.

But maybe someone dares just to count pixels in one of those and calc the actutal (non-scaled) resolution? Shouldn't be much of a problem (unfortunately i cannot do it now so maybe someone else will? )

 

[scooby_dooby]

Well i don't think these reports especially for these 2 games came out all of a sudden (and not for other games despite many had downscaled screenshtos available). There were several reports about this "issue" around the web.

But maybe someone dares just to count pixels in one of those and calc the actutal (non-scaled) resolution? Shouldn't be much of a problem (unfortunately i cannot do it now so maybe someone else will? )

So we have 3 games out of 110 to put that into perspective, one was a launch game and the other two activision ports which both seem to run and look better on 360 than the PS3 version.

Hopefully we can turn this discussion back to Xenos now??