Nvidia's MFAA demo (BF4)

[Silverforce11]

Its a question of what you think NV is doing with that -1n frame data in motion.

Is the driver flushing it when its not static & vary versus the n frame? If so, its default 2x MSAA without the blur during motion.

If they somehow use that moving -1n frame, compare & extrapolate the edges to filter in the n frame, that by definition, would cause motion blurring.

 

[Mand]

Err, if they're combining data from the previous frame with the current frame in whatever capacity, that's the definition of a blend.

It's not even remotely close to that simple.

Blending means you take pixel n-1, pixel n, calculate the average, and display as the new pixel n.

That's not what they're doing.

 

[Silverforce11]

AT covered it very well here, and it IS just a form of temporal AA, with clever driver optimizations to discard frames that vary too much causing massive blurring.

http://www.anandtech.com/show/8526/nvidia-geforce-gtx-980-review/6
------------------------------

"In any case, Multi-Frame Sampled Anti-Aliasing is based on the concept of changing the MSAA sample pattern in every frame, in practice using a 2x (2 sample) MSAA pattern and combining the results from multiple frames to mimic a 4x (4 sample) MSAA pattern. If it’s done right then you should receive results comparable to 4x MSAA with the cost of 2x MSAA."

"Once you can grasp the concept of changing sample patterns, the idea is actually relatively simple. And in fact like DSR it has been done before in a lesser form by none other than AMD (or at the time, ATI). In 2004 with their X800 series of cards, AMD launched their Temporal Anti-Aliasing technology, which was based on the same sampling concept but importantly without any kind of frame combining/blending. Over the years Temporal AA never did see much use, and was ultimately discontinued by AMD."

"What sets Multi-Frame Sampled Anti-Aliasing apart from Temporal AA and similar efforts – and why NVIDIA thinks they will succeed where AMD failed – is the concept of temporal reprojection, or as NVIDIA calls it their temporal synthesis filter. By reusing pixels from a previous frame (to use them as pseudo-MSAA samples), the resulting frame can more closely match true 4x MSAA thanks to the presence of multiple samples. The trick is that you can’t simply reuse the entire last frame, as this would result in a much less jagged image that also suffered from incredible motion blur. For this reason the proper/best form of temporal reprojection requires figuring out which specific pixels to reproject and which to discard.

From an image quality standpoint, in the ideal case of a static image this would correctly result in image quality rivaling 4x MSAA. As a lack of camera motion means that the pixels being sampled never changed, the samples would line up perfectly and would fully emulate 4x MSAA. However once in motion the overall image quality is going to be heavily reliant on the quality of the temporal reprojection. In the best case scenario for motion Multi-Frame Sampled Anti-Aliasing still will not perfectly match 4x MSAA, and in the worst case scenario for motion it could still result in either 2x MSAA-like anti-aliasing, significant blurring, or even both outcomes."

I hope you guys understand it better now, its not magical.

All NV did was took temporal AA, but made their drivers analyze and flush the -1n when it would lead to motion blur or entirely losing the edge data, as during fast motion. This in essence, switches from 2x MSAA when its moving to 4x MSAA equivalent when static.

It IS ADAPTIVE MSAA. It's quite clever actually!

Also as I posted earlier, their Dynamic Super Resolution is a form of SSAA. Interestingly, its not as accurate as SSAA & uses Gaussian filtering so its not as sharp, but its advantage is wider compatibility.

 

[BFG10K]

If they somehow use that moving -1n frame, compare & extrapolate the edges to filter in the n frame, that by definition, would cause motion blurring.

It looks like something similar. If this was just "Plain Jane" MSAA in both frames with different samples and no blending, then transparent surfaces wouldn't be affected, and it would be exactly like AMD's TAA.

It's not even remotely close to that simple.

Blending means you take pixel n-1, pixel n, calculate the average, and display as the new pixel n.

That's not what they're doing.

http://www.thefreedictionary.com/blend

blend (blnd)
v. blend·ed or blent (blnt), blend·ing, blends
v.tr.

2. To combine (varieties or grades) to obtain a mixture of a particular character, quality, or consistency: blend tobaccos.

You can blend data in many different ways including simple averages, weighted mixing, nearest neighbour, etc. So in reality it's not as simple as you make it out to be.

They're using a filter to combine those two frames, plain and simple. "Temporal Reprojection", as shown by Silverforce11's post above.

 

[SirPauly]

It's gong to be nice for GeForce owners over-all for flexibility if they choose and for potentially more quality for less performance! If ya don't like it noone is forcing ya to potentially use it.

The setting that interests me is what nVidia is talking about here:

https://www.youtube.com/watch?v=Nef6yWYu0-I

A basic free x2 setting and may do wonders for improving image quality for higher resolutions and stereo 3d gaming -- curious if one could add this to DSR.

 

[Grooveriding]

It's unfortunate this is not a working available feature and there are no reviews, just marketers trumpeting for it. When I hear filter related to AA and nvidia I start to expect blur and image quality reduction. After the way TXAA was hyped for Kepler it ended up a disaster that looked awful and ruins image quality that was barely implemented.

I hope reviewers revisit this MFAA when/if it becomes available to take a look at the IQ it produces and how much work has been put in to get compatibility in games. This is why I never care much about most of the new features tossed around with video card launches, they generally go nowhere and don't live up to the hype and just become bullet points to put on the packaging.

 

[Mand]

It looks like something similar. If this was just "Plain Jane" MSAA in both frames with different samples and no blending, then transparent surfaces wouldn't be affected, and it would be exactly like AMD's TAA.


http://www.thefreedictionary.com/blend

blend (blnd)
v. blend·ed or blent (blnt), blend·ing, blends
v.tr.

2. To combine (varieties or grades) to obtain a mixture of a particular character, quality, or consistency: blend tobaccos.

You can blend data in many different ways including simple averages, weighted mixing, nearest neighbour, etc. So in reality it's not as simple as you make it out to be.

They're using a filter to combine those two frames, plain and simple. "Temporal Reprojection", as shown by Silverforce11's post above.

I love how I make a post telling you your assessment was too simple, and your response is telling me I'm making it out to be too simple, and then tell me it's simple.

This is not a simple blending, and using the term does not fully describe what is actually being used. Dictionary definitions of common words often fail to fully capture highly technical information.

Yes they're combining information from two frames, but it's not "plain and simple" blending.

 

[f1sherman]

All NV did was took temporal AA, but made their drivers analyze and flush the -1n when it would lead to motion blur or entirely losing the edge data, as during fast motion. This in essence, switches from 2x MSAA when its moving to 4x MSAA equivalent when static.

They are not discarding entire n-1 frame.
They are running test on n-1 frame, test which decides what pixels are usable in sampling and combining with n frame.
Clever part is constructing this test/filter.

 

[Silverforce11]

They are not discarding entire n-1 frame.
They are running test on n-1 frame, test which decides what pixels are usable in sampling and combining with n frame.
Clever part is constructing this test/filter.

Just imagine a fan blade that has a sharp edge and is moving. If it moves slow, you can still see its sharp edge, the faster it moves (where each frame will differ greater in the position of our object of interest), the more blurred it gets.

What you think they are doing, is analyze that frame, flush the data relating to the fan but keep the other static samples to combine with the n frame. Sure, but I'd think that approach is very computational heavy to do on each frame... but we won't know how they do it unless there's a detailed paper on it.

As it is, Ryan covered it very well. Depending on how its implemented, it could simply be 2x MSAA during motion or worse with major motion blur. I doubt NV would accept the blurring, or maybe they will, they can claim it adds "cinematic IQ & experience"..

 

[antihelten]

What you think they are doing, is analyze that frame, flush the data relating to the fan but keep the other static samples to combine with the n frame. Sure, but I'd think that approach is very computational heavy to do on each frame... but we won't know how they do it unless there's a detailed paper on it.

Actually it's not really that computationally heavy to do so depending upon the method used. I would imagine that Nvidia simply uses the velocity buffer (which should be present in any game which features motion blur), since this buffer already contains all the data you need.

Also this whole thing is old hat anyway, since Intel came up with the same thing 3 years ago, including the blending of frames (which was missing from AMDs implementation):
https://software.intel.com/en-us/articles/dynamic-resolution-rendering-article

 

[SirPauly]

And to add another advantage is smaller memory footprints.

 

[SirPauly]

We took a look at this up closely in demos, and indeed 4X MSAA image quality is achieved, for only the cost of 2X MSAA. This also scales up, it can achieve 8X MSAA image quality for the performance of 4X MSAA.

A third party did at least look closely

If you were to see this in motion, the sample grid is actually moving, changing sample patterns in real-time. It has the added benefit of providing image quality improvements on objects that you wouldn't necessarily think would receive image quality improvements with MSAA. For example, some alpha textures or transparencies can benefit from MFAA versus MSAA. We will have to take a closer look at that once the technology is ready.

Neat to hear that it may improve transparency as well at no additional cost in performance, imho.

MFAA's implementation is not finalized yet, but again we discussed user control with NVIDIA and the goal is to retain user control over using MFAA or MSAA in your game as you want to. Right now, in its early iteration, 2X MSAA had to be set in the game, and then 4X MFAA enabled from GFE or NV Control Panel. Final implementation is not set yet. This technology has the potential to give you 4X and 8X MSAA at the cost of 2X and 4X MSAA respectively, which will allow you to raise your in-game settings and still have a high amount of AA sampling

Touches on compatibility -- enhance the application. Good points from Hardocp


http://www.hardocp.com/article/2014...orce_gtx_980_video_card_review/2#.VCB5GxbRWqJ

 

[njdevilsfan87]

Will this work on Kepler or this a Maxwell+ feature?

They are not discarding entire n-1 frame.
They are running test on n-1 frame, test which decides what pixels are usable in sampling and combining with n frame.
Clever part is constructing this test/filter.

My guess is using it to calculate a pixel change and then summing it up over a region, and then using that value to determine the filter to apply. Storing one frame would hardly be taxing on the memory. And then less AA being used in the end would end up with less memory usage overall.

Very clever Nvidia engineers. Keep up the good work!

 

[videogames101]

msaa and ssaa are still king, no matter how many letters nvidia or ati throw in front of "aa"

 

[Dylanharvey999]

MSAA and SSAA for the win!

 

[f1sherman]

SSAA is ded

Long live DSR

SGSSAA may still end up better looking vs DSR alone
,although that's not entirely certain due to DSR quality downsampling filter + you can still combine it with MSAA same like SGSSAA requires.

No need for driver support and trickery, not requiring a profile and not being tied to any particular API is simply huge.
Can't believe it took them this long.

 

[mindbomb]

all these forms of AA are dead. FXAA and it's variants are the future. DSR is throwing performance out the window, MSAA doesn't work in too many scenarios.

 

[CakeMonster]

So.. will this "replace" MSAA meaning that in the games that support MSAA it will use MFAA instead? Does it have to be added to the game to actually support it? In that case there's not really much point. And the games with MSAA are getting fewer and fewer anyway....