David Kirk on HDR+AA

[Laa-Yosh]

Even when rendering in layers, offline CG uses very high quality AA. There are some nontrivial issues with compositing overlapping objects, like, using premultiplied or unpremultiplied alpha, layer compositing math, etc. But you just have to get AA for the individual elements during rendering.

Also, shader AA and geometry/edge AA can and should be decoupled. Developers should write their own shader AA, and let the hardware handle multisample edge AA for them, I don't understand why Kirk wants to take it into another direction...

 

[nAo]

I don't understand why Kirk wants to take it into another direction...

Cause they can't support it now and they won't in the near future

 

[Geo]

Yes, CG is part of what irks me about that answer --Kirk appears to be throwing up his hands on the hardware side, but is there any evidence that they've taken a leadership position in trying to solve it on the software side, where he is pointing the solution must reside initially?

NV has often been a leader, yet here is one of the principle IQ issues of our time, and he's got the Pontius Pilate routine going. It seems to me that PS3 is going to shoot some extra juice into CG, so taking a leadership position there to get a software-side solution that is friendly for developers to use ought to be a high priority. It would also give you a much better chance (thru understanding the software solution better, and encouraging standardization of technique) to have a shot at appropriate hardware accel later.

 

[mr]

I don't understand why Kirk wants to take it into another direction...

Cause they can't support it now and they won't in the near future

Do you mean "near future" in a "RSX-near future" way?

From all the talk i gathered around this forum, and especially Dave Baumanns excellent Xenos article, I can only conclude that supporting hardware MSAA with HDRI is purely an implementation issue in the ROP's.
It "simply" requires logic (read transistors) to support the HDR image formats like FP16. Dave called Xenos ROP's fully orthogonal in that they can operate without restrictions on every supported output format.

The ROP's can handle several different formats, including a special FP10 mode. FP10 is a floating point precision mode in the format of 10-10-10-2 (bits for Red, Green, Blue, Alpha). The 10 bit colour storage has a 3 bit exponent and 7 bit mantissa, with an available range of -32.0 to 32.0. Whilst this mode does have some limitations it can offer HDR effects but at the same cost in performance and size as standard 32-bit (8-8-8-8) integer formats which will probably result in this format being used quite frequently on XBOX 360 titles. Other formats such as INT16 and FP16 are also available, but they obviously have space implications. Like the resolution of the MSAA samples, there is a conversion step to change the front buffer format to a displayable 8-8-8-8 format when moving the completed frame buffer portion from the eDRAM memory out to system RAM.

The ROP's are fully orthogonal so Multi-Sampling can operate with all pixel formats supported.

I sincerely hope that NV will find the die space to implement this in the near future as MSAA with sparse sampling positions is not an easy thing to give up to, even with resolutions as high as 1920x1080 (*cough* PS3).

I may be incorrect in my understanding of the inner workings of ROP's so please correct me if I'm wrong.

 

[OpenGL guy]

And what was the primary improvement of the rushed-forward 5900 over the cancelled-as-quickly-as-possible 5800? I can't believe we're even talking about this like it isn't a settled issue.

The 6600GT/9800P comparison was to show that a wider bus is not critical to performance superiority.

Pick a test that's bandwidth limited. Pointing to a case that's not bandwidth limited and saying "See? A 256-bit bus is nothing special." is pretty disingenious.

 

[nAo]

Do you mean "near future" in a "RSX-near future" way?

IMHO they will not suppor it till their WGF2.0 part is out (late 2006?)

I sincerely hope that NV will find the die space to implement this in the near future as MSAA with sparse sampling positions is not an easy thing to give up to, even with resolutions as high as 1920x1080 (*cough* PS3).

I think one of the reasons Sony/Nvidia are pushing this 1080p thing is cause RSX can't support MSAA on floating point targets and they know most people will just have SDTV or 720p TV sets so the majority of consumers will get a SSAAed picture. It would be nice if at least RSX supports some kind of rotated grid supersampling..

 

[trinibwoy]

Pick a test that's bandwidth limited. Pointing to a case that's not bandwidth limited and saying "See? A 256-bit bus is nothing special." is pretty disingenious.

Well in a bandwidth limited test, a card with more bandwidth will win no? And nobody is saying that 256-bit buses are useless (where'd you get that from) - just that there is evidence of a 128-bit card keeping pace with a 256-bit one.

f there are games that are bandwidth limited at playable settings then the 9800p will probably come through but I don't think there were any examples of that in those particular links (although I didn't look very closely). Even at 16x12 4xAA the GT was slightly ahead in Far Cry.

 

[Bob]

Pick a test that's bandwidth limited. Pointing to a case that's not bandwidth limited and saying "See? A 256-bit bus is nothing special." is pretty disingenious.

Which game and at what resolution would that be? I can only find the 6600GT noticibly slower than 9800 Pro/XT on Jedi Knight at 1600x1200, 4xAA 8xAF. But then again, the 6600 GT is still 2x faster than the 9700 Pro on those same settings.

Link

 

[mr]

Do you mean "near future" in a "RSX-near future" way?

IMHO they will not suppor it till their WGF2.0 part is out (late 2006?)

I sincerely hope that NV will find the die space to implement this in the near future as MSAA with sparse sampling positions is not an easy thing to give up to, even with resolutions as high as 1920x1080 (*cough* PS3).

I think one of the reasons Sony/Nvidia are pushing this 1080p thing is cause RSX can't support MSAA on floating point targets and they know most people will just have SDTV or 720p TV sets so the majority of consumers will get a SSAAed picture. It would be nice if at least RSX supports some kind of rotated grid supersampling..

Wow, thanks for the great answer. That's as straight out as I could ever hoped for.
It's true that downsampling to SDTV resolutions may look pretty nice, but I 'm not convinced that downsampling to 720p is going to look good enough compared to a native 4xMSAA'd 720p image. Ahh I can hear the disussions already.

Anyway enough of consoles.
While downsampling may be an option for PS3 (atleast for most users) it won't be for the PC. Lacking AA combined with HDR rendering has been the most annoying "feature" of NV40 for me despite the already low HDR performance.
I wanted G70 to fix that but it looks like we'll have to wait a little longer for that to happen.

 

[OpenGL guy]

Pick a test that's bandwidth limited. Pointing to a case that's not bandwidth limited and saying "See? A 256-bit bus is nothing special." is pretty disingenious.

Which game and at what resolution would that be? I can only find the 6600GT noticibly slower than 9800 Pro/XT on Jedi Knight at 1600x1200, 4xAA 8xAF. But then again, the 6600 GT is still 2x faster than the 9700 Pro on those same settings.

Link

And what's your point? The 6600 GT has a pretty significant mutlitexturing fillrate advantage compared to the 9700 or 9800. Why don't you look at the single texture fillrate from 3DMark? Those tests are very bandwidth limited.

 

[Bob]

And what's your point? The 6600 GT has a pretty significant mutlitexturing fillrate advantage compared to the 9700 or 9800.

How so? 8 texture pipes vs 8 texture pipes. Both the 6600 GT and FX 5800 Ultra are clocked the same.

If anything, the 9700/9800 has the advantage of decoupled texture/shader!

Why don't you look at the single texture fillrate from 3DMark? Those tests are very bandwidth limited.

I don't know about you, but I tend to not play 3DMark very often.

 

[Ailuros]

IMHO they will not suppor it till their WGF2.0 part is out (late 2006?)

While a reasonable assumption, there's still no guarantee that it'll happen with the first incarnations of WGF2.0 GPUs. The way Kirk illustrates the situation he doesn't make it sound like it'll be possible within one generation. Either that or he's deliberately exaggerating.

By the way where are the rumours that "G90" or beyond will be TBDRs? I heard according rumours for NV30,40,50,60....errrr wait I lost track...

I think one of the reasons Sony/Nvidia are pushing this 1080p thing is cause RSX can't support MSAA on floating point targets and they know most people will just have SDTV or 720p TV sets so the majority of consumers will get a SSAAed picture. It would be nice if at least RSX supports some kind of rotated grid supersampling..

I don't seem to follow exactly with that one. If there are going to be PS3 games that support 1080p then they'd have to be pretty simplistic in a relative sense and I wouldn't even think of float HDR in such a high resolution.

As for Supersampling in 720p, how many samples exactly? Sounds equally optimistic to me. IMHO the majority of games on either console will be more like 720p/4x or 2xMSAA depending on how demanding the game will be.

 

[OpenGL guy]

And what's your point? The 6600 GT has a pretty significant mutlitexturing fillrate advantage compared to the 9700 or 9800.

How so? 8 texture pipes vs 8 texture pipes. Both the 6600 GT and FX 5800 Ultra are clocked the same.

If anything, the 9700/9800 has the advantage of decoupled texture/shader!

You brought up the 9700 and 9800, now you want to confuse things be bringing in the 5800 Ultra? The 5800 Ultra was a 128-bit part, so if the 6600 GT is faster, might that not be because of the more efficient shaders?

The 9700 and 9800 are clocked far lower than the 6600 GT.

Why don't you look at the single texture fillrate from 3DMark? Those tests are very bandwidth limited.

I don't know about you, but I tend to not play 3DMark very often.

Whatever.

I can only find the 6600GT noticibly slower than 9800 Pro/XT on Jedi Knight at 1600x1200, 4xAA 8xAF. But then again, the 6600 GT is still 2x faster than the 9700 Pro on those same settings.

I checked out those results. Looks like something is very wrong with the test. The 9800 XT is 3x faster than the 9700 Pro, that doesn't make sense. Who cares if the 6600 GT is double the speed of a board with bad results?

 

[jvd]

Isn't the 9800pro over a year older than the 6600gt ?

I would expect for a valid comparision to compare cards of the same time .


So whats faster the x800xt pe or the 6600gt ?

About as valid as the 6600gt vs 9800pro comparisions

 

[digitalwanderer]

Pointing to a case that's not bandwidth limited and saying "See? A 256-bit bus is nothing special." is pretty disingenious.

I'd say that's a hell of a fair point and it absolutely slayed me for some reason.

 

[Ailuros]

Uhmmm gentlemen could we please come back to the float HDR + MSAA topic?

 

[nAo]

I 'm not convinced that downsampling to 720p is going to look good enough compared to a native 4xMSAA'd 720p image. Ahh I can hear the disussions already.

4X MSAA (as Xenos supports it on 32bit fp and 64bit fp RT) will look great on primitives edges, 1080p -> 720p -> 480p via SSAA will look great on pimitives interiors..
Too bad I can't see why XBOX460 titles can't afford some SSAA too if PS3 titles will render to 1080p resolution.

 

[_xxx_]

Well, one thing's for sure, Digi and WaltC are both gonna have fun today.

Which means that we're gonna have fun reading that, so that's nice

Although I'll be @ Napalm Death concert today, so I guess I'll have even more fun...

Napalm Death, that is a name I havent heard in years.

Just got back. They kicked some serious ass, it was a real joy. My ears are still bleeding...

 

[_xxx_]

Another question: the in-game AA settings in HL2 or Doom3 - are these just tickling the HW or is it something done on the SW side of things?

 

[nAo]

If there are going to be PS3 games that support 1080p then they'd have to be pretty simplistic in a relative sense and I wouldn't even think of float HDR in such a high resolution.

Think about running at half the frame rate (30 fps) but with per pixel motion blur techniques to fake a higher frame rate..
1080p -> 2 MPixels at 30 fps -> 60 Mpixel to shade per second
So we have 550*48/60 = 440 DP4 per pixel per frame (ovedraw 1)
So we can do a lot of work per pixel too..

As for Supersampling in 720p, how many samples exactly? Sounds equally optimistic to me.

1080p -> 720P = 1.5*1.5 = 2.25x AA

IMHO the majority of games on either console will be more like 720p/4x or 2xMSAA depending on how demanding the game will be.

Are you talking about XBOX360, PS3 or both?