Is HDR "Free" for the Xenos?

[nAo]

Afaik the RSX has nothing comparable to the edram, so the frame buffer lies in the vram. This effectively halfes the memory bandwidth and makes AA cost alot of memory bandwith.

I think RSX, like any other modern GPU, should have some (very small) tile cache on board.
It would be nice if nvidia had used the die area devoted to the NV40/G70 videoprocessor to increase these buffers
Someone says videoprocessor takes something like 20 MTransistors..
Enough to put a 384x384x128 bits tile buffer on chip

Actually propably both GPUs even have to share memory bandwith with the CPU, but I am not sure about that.

That's the case..

 

[fouad]

Lets use another time, a little bit of some logic to analyse this :

((( "FiringSquad: You said earlier that EDRAM gives you AA for free. Is that 2xAA or 4x?

ATI: Both, and I would encourage all developers to use 4x FSAA. Well I should say there’s a slight penalty, but it’s not what you’d normally associate with 4x multisample AA. We’re at 95-99% efficiency, so it doesn’t degrade it much is what I should say, so I would encourage developers to use it. You’d be crazy not to do it." )))



Execuse me but I cant afford to not laugh when reading this...

((( I would encourage all developers to use 4x FSAA )))

Hey guys never asked the question : Why he would need to encourage developers to use 4X AA if its free AA ?!!!!!!!!!!!!!!!!!!!!!!!!!!!

there is 2 options :
1/ he think the developers are stupid, they dont know that 4X AA at 720P is free ( or as he said at 95-99% efficiency )
2/ He knows very well that 4X AA at 720P is not free, and that developers have the choice to use it or not.


I will let you find the correct answer using your brains who GOD created for you to think, and think, and think...
so please think at least a little bit...


Edit :
I think we could arrive to a compromise : Waiting xbox360 launch, so if we will have any game using only 2x AA at 720P, and not 4X AA..Than I am wright ! and you will just write me an apologize...
But if all games will have 4x AA at 720P than I am wrong...and I will write a big letter to apologize to everyone..

OK ?!! Deal ?

 

[London Geezer]

MS apparently is "forcing" developers to release all games at 720p and 4xAA. By forcing i guess they'll make it hard for developers to release their games until 720p and 4xAA are used. So whether it's "free" or not, which basically it is, it won't matter.

I think the same should be done for 60fps too, at least for ALL racing games and all games that are better when played at 60fps.

 

[Shifty Geezer]

At 4xAA if there's a 5% penalty (average, don't know about worst case scenarios) then that's still a choice devs have to make, though I think they'd be crazy not to use the benefit.

And regards apologies, how's about you make a sensible argument first? You haven't presented an intelligent reason why you regard AA as not being free. It seems your entire opinion is based on skepticism, not logic. But as I said before, it's not the argument that needs addressing but your attitude. Keep going around saying people are stupid, using lots of exclamation marks, and telling us to use 'our GOD given brains' and you'll be the second on my ignore list in record time

 

[nAo]

OK ?!! Deal ?

No, this is not correlated to our debate. you're wrong, period.

 

[scooby_dooby]

faud, maybe you should also use YOUR brain and realize that sometimes you're wrong about your assumptions.

when you've obviously been proved wrong, instead of grasping at strws and making arguments based on nothing but "ATI is lying", just accept that you were wrong. With the information we've showed you, and dave's own article you should be able to accept this fact.

It's really not so hard.

Simple:
2xAA = 0-1% hit.
4xAA = 1-5% hit
Tiling is always on.

 

[blakjedi]

OK ?!! Deal ?

No, this is not correlated to our debate. you're wrong, period.

Man that was funny.

Fouad.. the option is always present to do less with the system than what its capable of. Basically you're argument is:

"If 720p w/4xMSAA is truly "free" it should be the absolute minimum (or only) capability of the X360... to which we intuitively say "nope."

MS sees the 720p w/4xMSAA capability as the hardware's sweet spot... you get the best performance with the least cost at that resolution using an AA scheme.

The RSX has to use brute force (which it might have) and/or assorted tricks to accomplish the same feat and maintain all the other aesthetics like framerate, draw distance, lighting, etc.

Its basically one less thing XB360 programmers have to worry about...

 

[ShootMyMonkey]

I think the same should be done for 60fps too, at least for ALL racing games and all games that are better when played at 60fps.

Essentially, you're saying, don't release any games when the console comes out. If you're expecting graphical miracles that will say "next-gen has arrived" and at the same time everything running at 60 fps on hardware that developers have laid eyes upon for no more than 3 months prior to release, well... I can't see that as anything short of being whiny. There's a pretty low percentage of games this gen are actually running at 30 fps. The only reason it was even remotely possible on Xbox the first time around is because there was nothing exotic about the system. It's not like people *want* to limit to 30 fps, but you've got a machine that is memory bandwidth-bound and is there to tell you "if you want to access memory, be prepared to hand over your firstborn child."

Also, I still debate the whole "free AA" thing. It's really not free when you basically consider potential for more fillrate -- if you're fillrate or pixel shader bound -- which is very possible with multipass renders, there's no way for AA to be free no matter how much bandwidth you have.

 

[SanGreal]

MS apparently is "forcing" developers to release all games at 720p and 4xAA. By forcing i guess they'll make it hard for developers to release their games until 720p and 4xAA are used. So whether it's "free" or not, which basically it is, it won't matter.

I think the same should be done for 60fps too, at least for ALL racing games and all games that are better when played at 60fps.

FWIW, Only 2x AA is required by Microsoft.

"We have four must-haves for Xbox 360 games," said the sunburnt, well-muscled Moore. "First, every game must have 720p minimum, and some developers want to go higher, which is fine by us. Second, every game must have 16x9 with no letterboxing allowed. Third, every game will support Dolby 5.1 surround sound, and fourth, every game will will be anti-aliased with our built-in 2X anti-aliasing tools."

http://xbox360.ign.com/articles/613/613257p1.html

 

[Acert93]

FWIW, Only 2x AA is required by Microsoft.

"We have four must-haves for Xbox 360 games," said the sunburnt, well-muscled Moore. "First, every game must have 720p minimum, and some developers want to go higher, which is fine by us. Second, every game must have 16x9 with no letterboxing allowed. Third, every game will support Dolby 5.1 surround sound, and fourth, every game will will be anti-aliased with our built-in 2X anti-aliasing tools."

http://xbox360.ign.com/articles/613/613257p1.html

I KNEW I read that before!

This may be why ATI says developers would be crazy not to use 4x AA... it has a minimal 1-5% hit to performance, and is not required, but the IQ advantage is enough that ~5% hit in overall framerate is negligable for the advantage. We are talking about the difference between 63 fps and 60 fps. 1.5 fps at the 30fps range.

 

[fouad]

you have free AA at 480i and 480p, all te 256 GB/s bandwidth is used, but than 2x and 4x AA at 720p you need to use tiling, and tiling doesent only affect the memory bandwidth of 256 GB/s, but also the main bandwidth of the system ( shared 22.4 GB/s ) so its impossible to use 3 tiles, when doing 4x AA at 720P and having only a hit performance of 5%.

 

[Jawed]

Also, I still debate the whole "free AA" thing. It's really not free when you basically consider potential for more fillrate -- if you're fillrate or pixel shader bound -- which is very possible with multipass renders, there's no way for AA to be free no matter how much bandwidth you have.

Xenos is never fill-rate bound, unless you consider 4GP/s fill-rate bound. The ROPs are in the daughter die with the EDRAM and they're designed to run at 4GP/s sustained, with the AA being an option that makes no difference in performance of the ROPs.

Turning off the AA doesn't increase the peak fill-rate in Xenos.

Jawed

 

[fouad]

Also, I still debate the whole "free AA" thing. It's really not free when you basically consider potential for more fillrate -- if you're fillrate or pixel shader bound -- which is very possible with multipass renders, there's no way for AA to be free no matter how much bandwidth you have.

Thanks GOD at least finally someone convinced that there is no free 4x AA at 720P on xbox360... ( even though his reasons arent true...)

 

[fouad]

That's the case..

No its not the case...In PS3 sharing bandwidth is optional, in XBOX360, its not optional...

 

[Mefisutoferesu]

Not to add fuel to Fouad's little fire, but so far the 95% efficiency at 4xAA has been a number only ATi has substantiated... meaning to say I haven't seen anyone work the math out. Perhaps, someone has and I forgot (bad memory), but if someone hasn't shouldn't we? I'd do it if I could, but I can't.

I did a quick reread through dave's article and he too only just quotes ATi... perhaps for what we know it's currently impossible to calculate the plenalty? Still, if we haven't done the math ourself; we can't really adhere to it as if it were the gospel truth, right?

Sorry, if this turns out to be nothing more than lighter fulid.

 

[Mariner]

Shhh. Quiet. If you listen very carefully you'll hear the sound of many people banging their heads against a brick wall.

Seriously, Fouad obviously isn't going to change his (gloriously uninformed) opinion so why continue to argue?

 

[Shifty Geezer]

you have free AA at 480i and 480p, all te 256 GB/s bandwidth is used, but than 2x and 4x AA at 720p you need to use tiling, and tiling doesent only affect the memory bandwidth of 256 GB/s, but also the main bandwidth of the system ( shared 22.4 GB/s ) so its impossible to use 3 tiles, when doing 4x AA at 720P and having only a hit performance of 5%.

How much bandwidth does writing 3 tiles to RAM take? At a guess, 10 MBs eDRAM is gonna consume...um...10 MB bandwidth per write. 10 MB write x 3 tiles = 30MB/s per frame x 60fps = 1,800 MB/s = 1.8 GB/s; well within the capabilities of the RAM bandwidth.

And regards the performance hit of only 5%, that's for the shaders, which aren't affected by RAM BW consumption (unless of course that's all taken up with backbuffer work so there's no data being delivered to Xenos, which it isn't).

 

[therealskywolf]

Fouad is funny

The man is really begging for people to believe him, and give support to his claims of Ps3 superiority in every possible way.

Give it a rest man.

 

[aaaaa00]

How much bandwidth does writing 3 tiles to RAM take? At a guess, 10 MBs eDRAM is gonna consume...um...10 MB bandwidth per write. 10 MB write x 3 tiles = 30MB/s per frame x 60fps = 1,800 MB/s = 1.8 GB/s; well within the capabilities of the RAM bandwidth.

Even better, it's a single long linear write of 10MB. This means the memory can be running at near peak efficiency for this transfer.

 

[Jawed]

How much bandwidth does writing 3 tiles to RAM take? At a guess, 10 MBs eDRAM is gonna consume...um...10 MB bandwidth per write. 10 MB write x 3 tiles = 30MB/s per frame x 60fps = 1,800 MB/s = 1.8 GB/s; well within the capabilities of the RAM bandwidth.

Well, a 720p frame is actually 2.6MB, so 60fps equates to 159MB/s.

This bandwidth cost is fixed because Xenos always copies the back buffer to the front buffer, whether tiling for AA is in use or not.

And regards the performance hit of only 5%, that's for the shaders, which aren't affected by RAM BW consumption (unless of course that's all taken up with backbuffer work so there's no data being delivered to Xenos, which it isn't).

Well, that 159MB/s will have a very minor effect.

The 5% seems to relate to the count of triangles that are transformed-and-lit in vertex shader code. I'm guessing that ATI estimates that upto 5% of the triangles in a scene will cross either of the 2 tile boundaries in a 3-tile frame-buffer. That would cost both in terms of the bandwidth consumed in reading the vertex data for those triangles and in executing the vertex shaders and running the triangles through the various set-up and rasteriser engines.

Jawed