Non sense... MLAA or not Sony would be better with a better GPU, Not that it was impossible with 330 millions of transistors.
Alternative AA methods and their comparison with traditional MSAA*
- Thread starter mitran
- Start date
Well then thank God Cell is in there. Just imagine a PS3 with RSX and a crappy CPU too!!
Good oneWell then thank God Cell is in there. Just imagine a PS3 with RSX and a crappy CPU too!!
Not only graphics. ^_^ Even if a modern day GPU can do MLAA (and they should be able to !), it doesn't mean Cell is useless. It just implies that the CPU is very versatile and can help the GPU where necessary. In the future, the line between CPU and GPU may blur too.
At the beginning of this gen, people also proclaimed that game AI doesn't suit/need Cell. Today, computer vision (PS Eye), motion gaming and other recognition algorithms are very suitable for Cell.
If you think about the Geo Hotz incident, it is the standalone SPU security that saved the day (or at least halt the advance in the mean time).
Also, the Japanese and Europeans can record TV in AVC while gaming. A lot of media tasks can be outsourced to the SPUs. This includes high quality in-game audio processing.
The benefit of a versatile and "abundant" CPU like Cell is you can use it anywhere where it makes sense.
(((interference)))
Veteran
So, is anyone implementing it on 360?
Why can't Bungie use it for Reach, instead of crappy temporal AA?
Imagine Alan Wake running at 720p using MLAA.
Why can't Bungie use it for Reach, instead of crappy temporal AA?
Imagine Alan Wake running at 720p using MLAA.
according to one of the experienced 360 developer
http://www.eurogamer.net/articles/digitalfoundry-blur-tech-interview?page=2
"I think there are certain techniques, such as God of War 3's MLAA that are just unfeasible on 360, but it works both ways, the 10MB of EDRAM in 360 makes 4x MSAA feasible."
http://www.eurogamer.net/articles/digitalfoundry-blur-tech-interview?page=2
"I think there are certain techniques, such as God of War 3's MLAA that are just unfeasible on 360, but it works both ways, the 10MB of EDRAM in 360 makes 4x MSAA feasible."
dragonelite
Veteran
Kinda old but didn't halo 3 odst have edge detection(visor mode).
Cant bungie implement edge based AA.
some test results from 2007
http://www.elitebastards.com/index.php?option=com_content&task=view&id=443&Itemid=29
Cant bungie implement edge based AA.
some test results from 2007
http://www.elitebastards.com/index.php?option=com_content&task=view&id=443&Itemid=29
This is a 'dumb' approach, in that it can only tell if a pixel's on an edge, but that's not enough information to apply reasonable AA post processing filters. It isn't aware of the actual polygon edges, their slope, direction, or whatever.
MLAA works by detecting patterns on the final image, even within objects so it will process more then just edge pixels.
Edge detection isn't totally useless though, it is common practice in movie VFX compositing to apply a slight edge blur on CG elements in order to better integrate them with the live background. I think Crysis 1 uses a similar technique as well.
MLAA works by detecting patterns on the final image, even within objects so it will process more then just edge pixels.
Edge detection isn't totally useless though, it is common practice in movie VFX compositing to apply a slight edge blur on CG elements in order to better integrate them with the live background. I think Crysis 1 uses a similar technique as well.
Billy Idol
Legend
Do you guys know if someone has ever tried a physical based AA method?
Let me explain: in physics you have a lot of funny equations, which model a "smear out" effect - such as the standard diffusion equation (used for instance to model heat transfer).
This equation have one important property: they affect first high frequency stuff, while low frequency stuff is affected less!
So you could construct an physcal equation which is based on high order viscosity terms (for instance square of Laplacian) to increase the effect that high frequency stuff is affected first, while keeping lower frequency stuff.
Amplying such an equation on a given image (interpreting pixel data as for instance a given temperature distribution) should in theory get rid of the high frequency stuff by using a smear out effect - while being super cheap...
Are there any AA methods based on this in use??
Let me explain: in physics you have a lot of funny equations, which model a "smear out" effect - such as the standard diffusion equation (used for instance to model heat transfer).
This equation have one important property: they affect first high frequency stuff, while low frequency stuff is affected less!
So you could construct an physcal equation which is based on high order viscosity terms (for instance square of Laplacian) to increase the effect that high frequency stuff is affected first, while keeping lower frequency stuff.
Amplying such an equation on a given image (interpreting pixel data as for instance a given temperature distribution) should in theory get rid of the high frequency stuff by using a smear out effect - while being super cheap...
Are there any AA methods based on this in use??
Yes, edge blur from the sounds of it. The problem is blur isn't the same as antialiasing, and though you lose the jaggies, you also introduce a feeling of blur. You can detect high-frequency image data through an effective edge filter that finds those jaggedy edges, and then do what you want to them, which isn't new. MLAA tries to reconstruct the edges which is where it's new and exciting.
Kameradschaft
Regular
So MLAA is not possible on 360 because of it's weak CPU? or because the 360's CPU is already pushed and used for other things in games? I was under the impression that Xenon is still under-utilized and most developers are relying on the easy and relatively strong Xenos - IIRC Capcom's MT Framework based games is an exception but they still manage to add dynamic 2-4xAA in their games resulting in very good IQ.
It's sad to not have a first-party like ND or Guerilla on the 360 to finally see how good a game can look with an engine specifically built around 360...Gears 1/2 and the upcoming third game look awesome but it's still UE3.5+ and Bungie's effort with Reach is kinda disappointing considering the game is again not running at 720p and having temporal AA especially on the system that it's well known for it's easy AA implementation.
Hmmmm. In terms of processing time, if we consider one of Xenon's VMX units equivalent to a SPU, GWAA takes up 20ms on one SPU. You'd need ~7ms per frame of 3 VMX units on XB360 to achieve the same, if they can work as effectively which is dependent on cache activity. So whatever game would implement it would need to find that much spare time, which is over 20% of total availble processing. Just for MLAA, whereas XB360 can do MSAA quite comfortably.
So it'd unsurprising that no dev would try it, but I am curious how well the VMX units could handle the technique, and if the SPU architecture is providing grand efficiencies that make possible a technique that doesn't fit well with conventional architectures.
Shifty sorry about of topic and please correct me if I'm wrong, but I've heard (dev talks) Xenon/x360 CPU does not have exactly the power of three PPU, but something like two at most(despite 6 threads not so efficient),because have in theory cache issues(L2 1MB for all 3 core and sharing with Gpu C1 include), memory controller, latencies among others that much affect the overall performance.
So if the above statement is true,spend a core or part of this process(cycles) in MLAA would impact substantially higher and maybe too heavy.
(sorry my bad english)
Last edited by a moderator:
Lucid_Dreamer
Veteran
Why would we consider one of Xenon's VMX units to be the equivalent of an SPU? The VMX unit is a part of each core, right? In other words, when someone says an Xenon core, it is understood that the VMX unit is a part of what makes up that core. An SPU is suppose to be a lot faster than an XeCPU core, according to nAo.
Might be wrong here but I was under the impression each core does 2 threads with one being not fulyl fledged and alike Intels hyper-threading. IIRC Metro 2033 dev said it could boost perfomance 30% or 50% under optimal conditions for each core.
Lucid_Dreamer
Veteran
http://forum.beyond3d.com/showpost.php?p=1124263&postcount=116
nAo said:
There are, also, several studies on PPU performance in the Cell versus an SPU. These tests didn't cover the whole gambit. In these particular tests, an SPU was several times faster. I can find those studies, if you want.
Yes but PPU in PS3 is not of same architecture as the 360s ones. And then "run circles around" dont tell much as to what type of code. How is the efficiency with different types of code?
Lucid_Dreamer
Veteran
The PPUs are suppose to be almost the same. The differences are suppose to be the size of the VMX register and the L2 cache. Per PPU performance should be faster in Xenon, but not by a large amount (especially with all PPUs in Xenon sharing L2 cache). Your questions can be answered with this studies. I guess that's your way of requesting them. I will relocate those studies and add them to this post.
Edit: Tree Sync - http://www.ll.mit.edu/HPEC/agendas/proc07/Day2/08_Agarwal_Pres.pdf
Biomolecular Simulations - http://arxiv.org/PS_cache/physics/pdf/0611/0611201v2.pdf
Cloth Simulation - http://www.research.ibm.com/cell/whitepapers/alias_cloth.pdf
Last edited by a moderator:
Similar threads
