RG/SGSSAA

[Gabbyjay]

This going nowhere, anyway.

Just to bring some new stuff into this discussion, here are two vids (which i already sent to shonumi by PM) showing a comparison of 4xSGSSAA vs 9xOGSSAA to show the difference in moiree effects.

Watch the black walls left and right of the track for moiree patterns.
With 9xOGSSAA they are quite disturbing:
http://www41.zippyshare.com/v/10507172/file.html

With 4xSGSSAA they are already gone:
http://www5.zippyshare.com/v/72685377/file.html

8xSGSSAA would be even more effective against more visible moiree patterns, of course. But on the other side, even 16xOGSSAA does not get rid of em.

Please notice that it's 4 subpixels vs 9 in this comparison.
9xOGSSAA has slightly better texture detail then 4xSGSSAA since it has a lower LOD bias, of course (but also comes with more then double the performance hit and has the worse EER).

[Shonumi]

xemnas Wrote:What do you mean by high DPIs?

Roughly anything Retina-level and above. I think I posted the DPI of the 13" or 15" here a bit ago. It's 227 and 220 respectively, notably higher than either 108.79 or 163.18.

Gabbyjay Wrote:You said it was not "part of the equation".
Make up your mind.

Go back and reread my posts. You were asking what the equation was. The equation was the overall value (all aspects) of RG/SGSSAA to users. I said that for some people, performance won't be part of the equation because they don't care about it, essentially stripping it from that said equation. Individual needs and interests are going to dictate changes to that balance, just like anything else, which is why I've been constantly reminding you of that. Broadly speaking, the "equation" is how worthwhile RG/SGSSAA on all levels is to all users, but when you look at individual cases or specific subsets of groups, you realize that their needs are going to alter what has to be taken into account and what can be ignored outright. We can't just look at RG/SGSSAA in general terms based strictly on it's overall value; we have to look at examples of how it would or wouldn't benefit others. If you claim performance is a bonus to some, I have to point out that conversely it won't be the case for others. If you point out that it provides a high IQ in comparison to other AA methods, I have to point out that conversely it won't make a difference to others.

Gabbyjay Wrote:And the reason for this could be that we already have it, as far as it is possible with today's hardware, couldn't it?

Are you talking about already having it (RG/SGSSAA) available through drivers? It doesn't change the fact that people have not been asking for this to be implemented natively in Dolphin. Most people simply haven't been wanting this enough to pester the developers.

Gabbyjay Wrote:Think about it this way:
There are many users who dont know about a certain feature as long as it's not included in the Emu.
But imagine it would be there: Many users notice it, and many of them will use it.

They could use it, or it's equally possible that they could just outright ignore it. Consider all of the features that get implemented in Dolphin already. When was the last time you checked Enable Wireframe, saved or loaded a state from a file manually (Load State... and Save State... respectively) or dabbled with TAS input? Consider a more mundane piece of software like MS Word, notorious for cramming features that most users scarcely ever used. Simply making it available does not ensure it will be used rather than passed over. It certainly increases the likelihood of its being used, but on the flipside does not mean it will be used.

Gabbyjay Wrote:So why do you think the same is not true for Dolphin?

We already get a lot of clueless people here like that who will flip every switch and knob on Dolphin just to see if they can get a better gaming experience (it's probably half the cause of issues in Support :/). Frankly we need less of them and more informed users, even if it's just about AA and not other options that affect Dolphin (like its hacks). If you want to go about educating users, I have no objection. We already try so hard with NaturalViolence's guides in the Hardware forum (and yet the clueless come forth...)

Gabbyjay Wrote:So as you lack even the basics, how can you determine, for example, if 9xAA would be sufficient in fighting moirees at high DPI resolutions?

How can you be so certain we will be able to sufficiently detect moirees with displays with high DPI and high resolutions, and how do you know that making the image finer by increasing the amount of pixel in a given spot and likewise shrinking said pixels won't be enough to deal with moirees without the need for RG/SGSSAA? It's as much speculation on my part as it is for you, since none of us has the displays necessary for testing (no I won't buy a MacBook Pro Retina either, I'm saving up for a Bay Trail NUC). I'm fairly confident 9xSSAA would be enough given second hand accounts from people who have done gaming on high DPI displays.

Gabbyjay Wrote:My answer to that was, of course they are not that concerned with it, as we already can use this (as far as possible).

This is more straightforward (to me) than what you wrote earlier. Essentially you're saying they have no reason to concern themselves with RG/SGSSAA because it can already be implemented through drivers, correct? At any rate, still consider that they are not concerned with it for other reasons I previously mentioned (lack of utility to the average Dolphin user).

Gabbyjay Wrote:I don't think so Bobby... i do not remember any emu dev wanting to have his emu have less functionality than an earlier version. If a feature is removed, there has to be a reason other than that, like bugs, compatibility, a lot of work involved in keeping it work over the revisions, and stuff. Plus, they could not even determine how many users use a feature that is already incloded and how many don't.

Alright, ignore the fact that I've been involved with this community longer than you have, and ignore the fact that Lock Threads to Cores was removed specifically because the devs did not see its usefulness (delroth or neobrain said it's not Dolphin's job to ensure its CPU affinity) and it was only needed by a limited number of mobile users and Windows could provide the same functionality as this feature. Disable Lighting was removed because it was useless to users and cluttered the GUI. The OpenCL Texture Decoder is going to suffer that same fate if no one bothers to improve it.

Gabbyjay Wrote:That is NOT really an option!
You will end up using a totally outdated version one day, and you will have tradeoffs to make, as newer versions might provide functions that you actually wanna use.

Again, you don't seem to have been part of this community for very long. For many people, it is their only option. They are unable to upgrade hardware, and they don't want to stop playing Dolphin. The solution, though not ideal, is to use outdated, and sometimes even unsupported revisions. It's a frequent enough phenomenon here, and it has long been the source of problems, both for users, devs, and forum staff.

Gabbyjay Wrote:Could this include the time when high DPI displays are common place?

Who knows. I'm still wanting for a display with infinite resolution

Gabbyjay Wrote:That way you could also say performance is not an issue at all, cause in 30 years there will be PCs capable of doing it.

My point was that GPU tech is advancing so fast that you'll probably have to wait more like 5-10, which is impressive given the scale of what's being done. GPUs are going to have to handle more (Intel has made a big deal about 4K for Bay Trail) once the display technology evolves.

Gabbyjay Wrote:I wanna see the system capable of handling high internal res/high DPI resolutions + 9xSSAA at playable framerates. 4K resolution also is a lot more demanding than today's 1080p standard. Thus not only the hardware evolves, but also the standard rises.

A user with a GTX Titan did something similar a few months ago (not high DPI, 4x IR + 9xSSAA, which, although smaller than common 4K resolution, is as best an estimate we have atm) with SMG2 (one of the most GPU intensive games in Dolphin). It was a constant 60 FPS and the results were quite impressive. It may seem hard to phathom but even the GTX Titan will be outdated eventually by GPUs a fraction of its price. The threshold of hardware that can achieve this is going to get lower as time goes by.

Gabbyjay Wrote:Plus, future revisions of an emu might as well be slower if they include another features and/or aim for more accuracy.

Some aspects speed up and other speed down. It all depends on what's changed. FastMem gave Linux users a free 10~15% boost in speed for example, and OpenGL performance has never been higher on my end. Dolphin, however, tends to get more intensive on the CPU than it does on the GPU. That is to say, in the time I've been playing Dolphin, I've noticed that as it gets more demanding, it tends to get more demanding for my CPU, not so much my GPU. What can run Dolphin at 4K + 9xSSAA probably will not greatly change from what can today. Of course, neobrain could come along and ruin all the fun by reverting everyone to the SoftRast...

Gabbyjay Wrote:There are also advantages of "saving" performance even if the hardware was capable of doing more.

That's not very "enthusiast". Push it to the MAX! :p

Gabbyjay Wrote:Same could be said for almost any feature in emulation: Faster loading times, savestates, customizable controls, improved graphics or sound and the likes are only great if you care about them.

All of those features were added nonetheless because the devs cared about them, saw their advantages, and made the assumption that others would as well. This is not the case with RG/SGSSAA.

Gabbyjay Wrote:Do you think Dolphin was created because they had a certain number of users in mind?

Yes. It was originally made for people who wanted to run GC games on their PCs; people wanting an automated hotdog maker were certainly not the focus of the early dev team.

Gabbyjay Wrote:And i consider the flexibility to adjust your IQ/performance ratio an advantage that makes a good emulator even better.

And the devs don't. If you think it would make Dolphin better, you've yet to convince them. Like delroth said, it's only going to happen in the near-future if someone who wants it gets up and implements it.

Gabbyjay Wrote:So this is why people buy ferraris, live in expensive hotels, or play video games...

And usually != always.

[xemnas]

(10-04-2013, 01:45 PM)Shonumi Wrote: Roughly anything Retina-level and above. I think I posted the DPI of the 13" or 15" here a bit ago. It's 227 and 220 respectively, notably higher than either 108.79 or 163.18.

Thanks. However, I don't think many people use Dolphin mainly on MacBook Pro 13" or 15". The screen size is a bit too small to enjoy the games fully. Anyway, 163.18 PPI on a 27" monitor could be considered Retina-level given enough viewing distance (I think 20 inches or more). However, it will probably take 10 years or more from now until most people use 4K 27" monitors. It's unclear to me why you refer to that far future.

[pauldacheez]

»I don't think many people use Dolphin mainly on MacBook Pro 13" or 15".
comex and chpwn use Dolphin on 15" rMBPs. There are other users too, and while they're certainly not a majority, you can't discount 'em.

»The screen size is a bit too small to enjoy the games fully.
That's a generalization. I'm perfectly capable of enjoying a game on a small screen, and it's likely that plenty of others are too.

»However, it will probably take 10 years or more from now until most people use 4K 27" monitors.
People who care about image quality beyond the typical AA options will likely be using those much sooner. I mean, look at all the people with cheap Korean 27" 1440p screens.

[Gabbyjay]

(10-04-2013, 01:45 PM)Shonumi Wrote: Roughly anything Retina-level and above. I think I posted the DPI of the 13" or 15" here a bit ago. It's 227 and 220 respectively, notably higher than either 108.79 or 163.18.

It will take time till we see this on large screen devices such as TVs and even longer on projectors.
It is obvious that for the time until then, other AA forms make sense.
As for the time after that, you are only speculating:
Will 9xSSAA be enough?
I already had an example where this pattern is very ineffective: Moirees. Even on the highest resolutions, since it is a side-effect of the regular nature of the pixel grid, which is the same grid with higher resolutions.

' Wrote:I said that for some people, performance won't be part of the equation because they don't care about it, essentially stripping it from that said equation.

Today, this is not true even for the highest end systems.
I am using a Geforce Titan and cannot even get the highest possible setting (9xSSAA + auto window size on 3840x2160) to work. It will crash, which is logical. The next available setting is 9xSSAA with 2560x2112, which has unplayable frame rates. (And looks anything but perfect, btw.) And this is on a $1000 card.
It will take time till we see a graphic card of this level reach prices the average user is willing to pay. It will take even more time till we get a card in this price region that makes it possible at playable frame rates. And even more time will pass till the same is true for the much higher resolution of smartphone-like high DPIs on all screen sizes (as that resolution is also a lot more demanding).
So you're quadruple-speculating:

1) About the degree of aliasing it will remove.
Remember, aliasing does not only consist of pixelation, but also temporal effects, shimmer, moiree and the like, and it is also about precision with all it's side effects.
2) About the number of users that will make use of AA-Types other then 9xSSAA.
You don't know anything about that number, and you cannot judge about this number by looking at this forums. It is not representative.
3) About the date where "high DPI devices" are common for all screen sizes.
4) About the hardware and if it will be enough to make playable fps out of this, especially as the emu develompent progress can slow things down again.

Quote:We can't just look at RG/SGSSAA in general terms based strictly on it's overall value; we have to look at examples of how it would or wouldn't benefit others.

Why do we "have" to do this? Is there a rule that an emulator hase to be some kind of democratic model or what is this?
How many users have a 5.1 sound system on their PC? Definitely a minority, most people don't even have it in their living room.
Yet, there still is a "Use Dolby ProLogic II decoder"-option in Dolphin.
Should this option not have been included, cause only a few people use it?
Sould it get thrown out of Dolphin for this reason?
Both no! Cause it actually makes the emulator better.

What is the whole point of this majority/minority-thing anyway?
Why should percentages be more important then absolute numbers here?

If 10% out of 100.000 users make use of a feature, if is a minority.
Yet there are still more people using this than if 70% of 10.000 users would use it.

So whats the point? Should they stop to care about that feature if it's user percentage would get low, even if with time the total number is actually rising? You don't even know anything about both numbers, the percentage and the total...

Did they even have a certain number of users in mind when they created the whole Dolphin project? And if they did, for what reason? It's written for a minority of users anyway, and no money outta that...

Quote:If you point out that it provides a high IQ in comparison to other AA methods, I have to point out that conversely it won't make a difference to others.

No. You can only say they probably don't care about the difference, based on speculation about all involved factors, see above.

Funny thing is that i remember the same discussion about the 1080p-standard. Lots of people said "AA would be unnecessary since the high resolution" (compared to the former SD). We all know that this is not true, and this also shows another important thing:

The personal demands someone has on quality often rises over time.

When i was younger, i was satisfied with things which i today wonder about: How could i ever accept this?

Quote:Are you talking about already having it (RG/SGSSAA) available through drivers?

Yes.

Quote:It doesn't change the fact that people have not been asking for this to be implemented natively in Dolphin.

What was that thread here about again?
The opening posting says: "RG/SGSSAA - will it ever be an option?".

Anyhow - take me for example. You said I was some kind of "AA enthusiast" - and even as such I don't care if it's included in Dolphin, as long as i can make it work through the drivers. As long as this method works and does not have any disadvantages compared to if it would be included in the drivers, I'm quite fine with it. In fact, I like it even more this way.
So why should I ask to have it implemented directly?

Quote:They could use it, or it's equally possible that they could just outright ignore it. Consider all of the features that get implemented in Dolphin already. When was the last time you checked Enable Wireframe, saved or loaded a state from a file manually (Load State... and Save State... respectively) or dabbled with TAS input?

And yet, they are still included (for a reason). You are contradicting yourself here. As you can see it's not only about how many users will actually use it.

Quote:Simply making it available does not ensure it will be used rather than passed over. It certainly increases the likelihood of its being used, but on the flipside does not mean it will be used.

And that was my whole point, which you just confirmed:
You cannot look at number of people speaking out for it in the forums, extrapolate a certain number of users who would want such a feature out of this, and then again determine the number of people actually using it from this. Your "equation" does not work this way. This is not a basis from where a representative analysis can be made.

And then, even if you could know the numbers: An emulator is not necessarily a project whose decisions are based on the analysis of the userbase.

Quote:How can you be so certain we won't be able to sufficiently detect moirees with displays with high DPI and high resolutions, and how do you know that making the image finer by increasing the amount of pixel in a given spot and likewise shrinking said pixels won't be enough to deal with moirees without the need for RG/SGSSAA?

Imagine a 1 pixel broad line crawling across your high DPI smartphone screen. Do you think you won't notice this one just because it is thin?
That is basically what moiree effects produce, only that it is curves crawling instead of lines.
To fight this, you need irregularity: Bring some irregularity into a regular pattern, so the human brain cannot detect it as a continuing line/pattern anymore. Higher resolutions are as regular as smaller resolutions are, and so is an ordered grid.

Quote:I'm fairly confident 9xSSAA would be enough given second hand account from people who have done gaming on high DPI displays.

I can only tell from my experience at 3840x2160 @ 4x, 9x and even 16x (!) OGSSAA, admittedly. If this is not enough, I'm sorry, but please remember that the original statement was "you people won't give a shit". Which implicts: It WILL be enough. My reply to that was that [even at high DPI there are scenarios that] "still might take advantage of further refined anti aliasing methods.". You just cannot rule out the possibility.

Quote:Alright, ignore the fact that I've been involved with this community longer than you have

Can you be certain about this, just because my nickname is not registered a long time?

Well, i already waited for this to come... "I am an authority!" - as if it would be of any significance in that context.

Quote:Ignore the fact that Lock Threads to Cores was removed specifically because the devs did not see it's usefulness

Now what does that mean for this thread?
It has already been made clear that SGSSAA is not possible in the DX9-plugin, as MSAA collides with the rendering technique. In DX11, it's already possible by forcing it with the drivers. The same is true for OpenGL (which has even more options), you only need to enable Transparency Supersampling in the driver instead of Transparency Sparse Grid Supersampling. This is a driver bug, of course, but it enables SG(!)SSAA for the whole image.
So it eems like we already got everything that is possible at the moment, the only feature I am asking for to be added is a custom internal resolution (and SSAA in DX11), which has not been implemented (only started) yet cause it causes GUI probs and cause the work that needs to be done.
The rest of this thread was a discussion about the advantages/disadvantages of OG vs SGSSAA, and our little discussion here.

Quote:Again, you don't seem to have been part of this community for very long.

"I am the autority here!" - you might as well say you're right cause you're older. Statement such as this dont help anyone in a forum.

Quote:For many people, it is their only option. They are unable to upgrade hardware, and they don't want to stop playing Dolphin. The solution, though not ideal, is to use outdated, and sometimes even unsupported revisions. It's a frequent enough phenomenon here, and it has long been the source of problems, both for users, devs, and forum staff.

Ah i see, you're nitpicking about my "this is not an option" here... yeah, of course you can use older revisions, but this is not a reason to remove a feature from an emulator, since as i said you will have many disadvantages or at some point you might be even forced to upgrade.

Quote:Who knows. I'm still wanting for a display with infinite resolution

Go play Vectrex.

Quote:My point was that GPU tech is advancing so fast that you'll probably have to wait more like 5-10 years

I think 5 to 10 years (!) is already enough time for certain AA-techniques to be worth keeping included, and even after that it still has advantages if the hardware power needed can be lowered.

Quote:A user with a GTX Titan did something similar a few months ago (not high DPI, 4x IR + 9xSSAA, which, although smaller than common 4K resolution, is as best an estimate we have atm) with SMG2 (one of the most GPU intensive games in Dolphin). It was a constant 60 FPS and the results were quite impressive.

Oh really?
I just happen to have a GTX Titan highly overclocked and at a custom watercooling using two aquariums dedicated to cooling.
Plus, i got SMG2 on my disk and yesterday, i did a vast amount of tests in 1920x1080, 2880x1620 and 3480x2160.
It runs at barely half the speed!
Higher internal resolutions are not even possible.

' Wrote:Some aspects speed up and other speed down.

So, again, lots of speculation involved in your part.
I mean look at PSX-Emulation, for example.
The system was released in 1994 (!). Lots of hardware upgrading since then, it's been 20 years!
Yet again, emu development still progresses and now that a certain russion coder found a method for more accurate internal hardware rendering and even included a Z-Buffer (!!), today's hardware barely manges 2x it's internal resolution and 3x won't even run in his plugin.
It's full of surprises!

So it's always good to have flexibility, so you can customize your IQ/performance-ratio in a way that you can have the best available quality at the optimal framerate, instead of making a compromise: Sacrificing a lot of IQ, or play at choppy FPS.

Quote:That's not very "enthusiast". Push it to the MAX! :p

In my country, there are many enthusiasts about silent computing, which is quite as challanging as high OC or the like can be.

Look at me, for example.
I got an highly overclocked i7 and a highly overclocked GTX titan in my system. Before i had two GTX 580 highly overclocked, which can consume even more power.
Yet, i still manage not only to keep GPU- and CPU-temps below the 30 degrees (celsius) mark, i'm also doing this completely passive with no active cooling.
Why do i say this here?
I'm using a storage solution as described above, so I am happy for every wattage saved while no compromises in performance/IQ should be made (so i keep overclocking). Because this lowers the amount of water i need, it keeps the water temperature low and also increases the hours it stays cool.
So I am happy if i can limit the amount of power consumption where ever possible. Frame limiters are a good example for this! You essentially are not using more hardware power than you need.

So, if there is an option where i can get the result i want with lower hardware usage, it is a good choice even from an "enthusiast" point of view.

For the average user, it also makes sense, since this one might not even want to afford high end hardware and/or wants to keep his hardware cool and quiet. Air cooling can make a lot of noise.

And what about devices like laptops or smartphones? What about older computers? It still makes sense to keep flexibility.

Quote:All of those features were added nonetheless because the devs cared about them, saw their advantages, and made the assumption that others would as well. This is not the case with RG/SGSSAA.

It seems some of em did not even really know about it, as you can see in this or in the other thread. It's a complex topic and the differences between an OG and a SG are sometimes hard to get. Even people who know a lot about this topic like naturalviolence, often have difficulties with the difference and similarities between an RG and a SG, for example. No i'm not saying I know everything or devs are not informed or anything, I just want to say there's a difference if there is no advantage - or if you just don't know about it or don't realize it. This is why a thead like this make sense, a lot of explanations have been given. I tried to contribute myself by posting an example about the moiree patterns and the differences an OG VS a SG makes in this context.
Plus, as i said, there is not much the devs can do about it now, anyway. All they could do would be to include it directly into Dolphin instead of making it available in the drivers, as i said. This works in OGL and DX11, in DX9 it is not possible due to Dolpin's rendering anyway.

' Wrote:Do you think Dolphin was created because they had a certain number of users in mind?
---
Yes. It was originally made for people who wanted to run GC games on their PCs; people wanting an automated hotdog maker were certainly not the focus of the early dev team.

That was not my question and you know that.
My question was about the number of people, not the kind of people.
They could not know if 100 or 10000 people use their program, yet they still wrote it.
So it seems there is a lot more involved here than just pleasing a certain number of people.

Maybe there are some people who just LIKE what they are doing here or it is their personal hobby to just write the best emu available?

' Wrote:And i consider the flexibility to adjust your IQ/performance ratio an advantage that makes a good emulator even better.

And the devs don't.

So why did they include the flexibility to choose between 2x, 4x, 8x multisampling, each even with 16 (!) quality levels, 8 options for different internal resolutions, 4x or 9xSSAA, 4 types of CSAA and more, if it wasn't about the flexibility to adjust your IQ to your system?

[xemnas]

(10-05-2013, 02:24 AM)pauldacheez Wrote: »I don't think many people use Dolphin mainly on MacBook Pro 13" or 15".
comex and chpwn use Dolphin on 15" rMBPs. There are other users too, and while they're certainly not a majority, you can't discount 'em.

No, I don't discount them. But 2 users or even 100 users are not many people compared to the entire Dolphin user base.

(10-05-2013, 02:24 AM)pauldacheez Wrote: »The screen size is a bit too small to enjoy the games fully.
That's a generalization. I'm perfectly capable of enjoying a game on a small screen, and it's likely that plenty of others are too.

Yes. That is a generalization. Surely, some people enjoy playing a game on a small display. However, I think between playing games on a 15" retina display and a 27" non-retina display, more people prefer the latter. Thus, a 13" or 15" retina display isn't a perfect solution for most people in this case.

(10-05-2013, 02:24 AM)pauldacheez Wrote: »However, it will probably take 10 years or more from now until most people use 4K 27" monitors.
People who care about image quality beyond the typical AA options will likely be using those much sooner. I mean, look at all the people with cheap Korean 27" 1440p screens.

I think we have different assumptions. I talked about most people but you talked about some people. AA is a solution for everybody. Everyone can use AA regardless of the resolution they are using. On the other hand, a high PPI display would be considered a solution for everybody if and only if everybody (or at least most people) used it.

[Gabbyjay]

Someone said in this thread that there is not much difference about OG/SG sample patterns.

I made some screenshots to demonstrate that the opposite is the case in regard to edge aliasing.

First is 4xSSAA, second is 9xSSAA and third is 4xSGSSAA.

Notice how 9xSSAA is working barely any better on the edges then 4xSSAA.
But with 4xSGSSAA, it's a LOT better - at less then halve the performance hit!

(The reason for this is that 4xSGSSAA has 4x4 EER, while 4xOGSSAA has only 2x2 and 9xOGSSAA only 3x3 EER; plus SG works better on the most aliased angles then OG.)

Look for yourself:

4xOGSSAA:
http://www7.pic-upload.de/04.10.13/8lt3skceeix.jpg

9xOGSSAA:
http://www7.pic-upload.de/04.10.13/3mcp5smgkjf.jpg

4xSGSSAA:
http://www7.pic-upload.de/04.10.13/7ypgqhrn9lj.jpg

[xemnas]

4x SGSSAA looks clearly nicer. Thanks for sharing.

[Gabbyjay]

So we got a new supporter?


Sadly, the image upload service converts the images to JPG. In the BMPs i originally created the difference is even more visible.

[AnyOldName3]

Y U No Use imgur? Also, why bitmap, not png? The compression's much better, and it's still lossless. Also, most things won't bother converting it, as it doesn't waste as much space as a bitmap.

Also, I'm pretty sure Razius posted video evidence of SMG2 at 4x IR with 9x SSAA. Maybe the issue is your weird AA not being as resource efficient as you claim (although I'm no expert on SG/OG variants of SSAA, so can't say this for sure).