Showing posts with label msaa. Show all posts
Showing posts with label msaa. Show all posts

Thursday, 2 December 2010

Tech Analysis: Gran Turismo 5

As a series that really excelled on Sony’s high bandwidth enabled PS2, Polyphony Digital have had to make some difficult choices in getting Gran Turismo 5 to work well with the RAM and bandwidth constraints of the PS3. Not compared to the PS2 as such, but against the EDRAM enabled set up of the Xbox 360 and the benefits it brings. Compromises can be seen in a variety of places, from the way shadowing is handled across the board, to actual particle effects resolution and even performance have all been part of a precise balancing act. But all that said, for the most part PD have made exactly the right choices for the game.

At times GT5 looks utterly amazing, truly gripping you in a way many other driving games rarely do. Although, at others, it looks decidedly a little rough – a telltale sign that the stunning technical achievement contained within didn’t come without a price. However, as a whole the game’s visual composition rarely falters to any degree in taking you out of the experience. Instead, it merely allows you to see just how massive an achievement GT5 really is.

Certain shortcomings aside, the game is a fantastic blend of artistic beauty and technical ambition. And although it is also one that doesn’t always manage to accomplish every goal flawlessly, it does more than enough convincingly in order to make it worthy of being called the ‘next-generation’ Gran Turismo.

Today we’ll be taking a huge in-depth look at the game as a whole, seeing how the whole graphics package holds up whilst also taking the time to delve deep into how, and why certain choices were made. We’ll be looking at both 720p and 1080p, along with the vast range of tech powering the game. We don’t have like for like screens for all our findings – and 1080p shots are unfortunately absent – but even then, just in 720p, there is much to discuss.

But without further ado, let’s get on with it.


Starting off as ever with the rendering resolution, and we can se that with Gran Turismo 5 Polyphony Digital are indeed aiming high. Higher you could say than any other driving game to date. Here we have a title that targets, and manages to frequently achieve 720p60, 1080p60, and even 3D, which I might add is no mean feat given the high levels of bandwidth and processing power this requires. And when you consider how limited RSX can be in some areas, you begin to realise just how much of an undertaking the developers were embarking on.

For 720p GT5 renders in 1280x720 with 4x MSAA (multi-sampling anti-aliasing), and in 1080p we have a horizontally upscaled 1280x1080 framebuffer with 2X QAA and use of 2X temporal AA (TAA).

Right off the bat, I can see that both approaches initially yield similar results in sharpness and overall texture detail. On first impressions GT5’s 1080p mode looks stunning. It appears slightly sharper on my HDTV than 720p despite the set’s top-end scaling capabilities. However, delve a little closer and things aren’t all they first appear to be.

Now before I compare the two, let’s go into 720p for a moment. The use of 4xMSAA is a pretty substantial inclusion to say the least. Here we are provided with some incredibly effective edge smoothing, with only the obvious sub-pixel and shader aliasing issues that stand out. But even they, from a regular viewing distance, don’t seem to be as pronounced compared to other games. Case in point: while fences, the steel barriers at the sides of the track, and thin/small pieces of geometry in the distance, and to a lesser extent up close, still shimmer and suffer from a few jaggies, this is hardly noticeable on all the game’s courses.

The older tracks – ported and mildly upgraded from GT4 – feature less in the way of details made up of small geometry, meaning generally less in the way of noticeable instances of sub-pixel issues. On the other hand, the more finely detail tracks created specifically for GT5 – or the ones from GT4 that have been given a full GT5 standard upgrade – have more in the way of jaggies appearing in the distance.

But even then, the use of 4xMSAA is largely successful in dealing with these issues in a general sense. That is, they take care of the most noticeable flaws leaving the rest to only mildly impact on what you are seeing.


Moving on to 1080p, and we can see a similar level of AA performance, courtesy of the combined use of QAA plus the temporal TAA solution. Usually an unwanted side effect of using QAA is the blurring of textures along with polygon edges, which often can noticeably reduced the level of detail available in any given scene. However, in the GT5 this doesn’t appear to be the case. Instead, texture detail comes off as being only a little bit blurrier compared to when the game is running in 720p.

The reason behind this boils down to the way the QAA algorithm has been specifically implemented across the 1080p framebuffer. Usually the QAA sampling pattern works across all pixels, smoothing over both high and low contrast areas, rather than looking for an edge. Here in GT5, with the greater pixel to texel ratio, the sampling pattern only affects neighbouring pixels, thus leading to less texture blurring as a result.

The comparative use of MSAA vs QAA also makes sense, as both 720 with 4xMSAA and 1080p with QAA have similar memory requirements with regards to the framebuffer. In fact, it is 720p that is slightly more memory heavy in GT5.

In terms of the TAA, this helps in mitigating some of the game’s sub-pixel shimmering issues, and looks to be present in 1080p only. You can turn off the use of temporal AA by changing the display option from ‘Normal’ to ‘Flicker Reduction’ to see the difference. ‘Flicker Reduction’ makes the overall image look slightly sharper, though introduces slightly more aliasing artefacts, whilst ‘Normal’ creates a smoother look, subtly improving image quality.

Another thing that is also apparent is that the use of TAA doesn’t blur the image when in motion. The technique used in GT5 doesn’t combine two separate frames together. There’s no unsightly frame blending, so no blur. Instead, the technique actually helps in reducing aliasing caused by alpha coverage, and parts of the game that uses A2C (alpha-to-coverage). This also means that it isn’t possible to capture it in frame-by-frame screen grabs.


Now, going back to our 720p vs 1080p observations. It is clear than the additional upscaling of the horizontal resolution and the implementation of QAA both impact the maximum level of sharpness present in the 1080p image. Whilst it still looks slightly sharper on a native 1080p display, the texture details don’t appears quite as clear and some of the polygon edges are not quite as clean. However, the trade off comes with a perceived level of better sharpness on the HDTV, as when looking at 720 vs 1080p FB grabs, the difference between the two is far more apparent – 720p appears to be clearly sharper.

The balance between having some kind of 1080p mode – even if the horizontal res is far short of that 1920 commonly quoted – with full use of AA, and a definitive 720p mode which maximises every drop of image quality, is an undeniably solid choice. Image quality seems to be well preserved across all modes – although we didn’t get to test out 3D – with less than the expected amount of aliasing being present. The fact that Polyphone have successfully implemented both high levels of AA whilst also managing to run at 1080p is particularly impressive.


Of course, in order to maintain such high levels of IQ, compromises in other areas have had to have been made. The huge level of transparencies on offer in GT5 is somewhat unusual considering how ill-suited the PS3 usually is for handling such a task without the end product coming through unscathed. Though, like with the concessions made in order to have a solid 1080p mode in the game, a similar thing can also be seen here too. What’s also important to remember, is that PD are aiming to reach a near constant 60fps at all times, and this does eat away into bandwidth and processing time available per frame.

GT5 uses two distinct methods for allowing the engine – and the bandwidth starved PS3 – to handle such copious amounts of alpha effects - that is to render all such elements at a much lower resolution to the rest of the scene, whilst using the half-res, interlaced style alpha-to-coverage (A2C) for the foliage.

As we’ve discussed before at IQGamer, the use of low res effects can often be detrimental to the overall image, with things such as smoke, particles and water, all impacting on the clarity of other objects in the rendering pipeline. This is also the case in GT5. However, the extent to which this happens is largely contained within specific circumstances, and not through the entire game en-mass in anyway that really distracts.

First up are particle effects themselves. These are all rendered in 1/16th of the screen resolution, along with having depth buffers rendered at the same resolution. Despite their extremely low res nature, there is very little in the way of alisaing or shimmering artefacts to be found around any of the game’s smoke effects. Like in both Killzone 2&3 the individual layers of 2D sprites which make up the effect have been blended together nicely, resulting in a smooth, natural transition between layers.



However, the downside of using lower res effects comes in when there are either multiple cars on screen at once (affected by lighting) – in which the effects look slightly rougher – or when there are weather/time of day changes, where we see some obvious artefacts.

Taking a look at the two screenshots above (grabbed from my phone cam) we can see how some of these artefacts manifest themselves. These shots clearly shows how the extremely poor looking, low resolution water particles create some pixelated and jittery shadows on the cars using the external view. Although these two captures are far from ideal, they do at least highlight that the low res nature of some of the effects in GT5 are far from pretty – they can often be rather unsightly.

Another issue also arises when the low res particles cover the actual cars themselves, which causes some further unwanted aliasing artefacts not always seen in all circumstances. Anomalies like this only happen when the low res effects react with the high resolution car models and specular maps when the sun is directly shining on them. It’s hard to spot when racing due to the angle required for this to happen, along with specific viewpoints on the cars themselves. In replays however, this, along with some of the other issues are far more noticeable.


Secondly, we have the issue of rendering more transparencies with regards to the game’s foliage. Trees and surrounding plants are all made up of subtle transparent elements which can take up a lot of memory bandwidth when using traditional alpha coverage. So instead, the developers have mitigated this somewhat by rendering all foliage in the game using the cheaper alpha-to-coverage technique. A2C works by rendering transparencies in a half-res, interlaced style manner, and the result is that all foliage in the game features a subtle screen door effect.

You can see this clearly above. Looking at the trees we can notice what appears to be a dithered look to things – the screen door effect, a common trait of A2C. Thankfully, at regular viewing distances (say 5 to 6 feet from a 32” screen) the effect isn’t noticeable at all. Instead, you can only see it when going slightly closer to the screen. And even then, it isn’t a big deal.


Looking at the games use of shadowing on the cars and certain environment details, and we can see similar cutbacks in quality. The low quality shadow filtering (it’s simple 2x2 PCF that has a very small impact in performance) creates pixelated and jittering artefacts on shadow edges, most noticeably around cars during replays, when driving in the cockpit view, and when lots of alpha effects are present.

These side-effects show up clearly in replays, though are not quite so obvious when actually driving around the track. Other instances of jittering, and flickering shadows can also been seen around trackside details, such as next to the bystanders and other objects which have small shadowing elements to them. Unlike with some of the game’s filtering issues, these can be seen at all times and are not just confined to the replays.


Whilst pointing out all these low resolution and poor quality effects might make it seem that GT5 has some noticeably serious issues, that simply isn’t the case at all. Instead, in motion, and while playing the game, most of these things don’t really impact on the overall graphical look of the game. Sure, the jittery shadows do constantly stand out. But the other stuff blends in surprisingly well with the rest of the game’s rendering make up.

Another thing to remember, is that Polyphony Digital are aiming for an absolutely smooth 60fps update while pushing full 720p framebuffers with 4xMSAA, and 1080p with combined 2xQAA plus 2xTAA, which clearly uses up a lot of the memory bandwidth and pixel fill-rate available. In that case, the choice to go with lower res buffers and what amounts to basically free shadow filtering was the right thing to do.

The most important element in the GT series, like with Call Of Duty, is with maintaining its super responsive, ultra fluid refresh. Having this nailed down is absolutely key to the experience.


Looking at performance across both 720p and 1080p, it is apparent that Polyphony Digital has done a rather admirable job in maintaining it throughout, sans a few issues with dropped frames and some screen tearing.

The comprehensive replay option found in the game allows us to look at performance in exactly like for like situations across both 720p and 1080p. Saved replays can be viewed from all in-game viewpoints, thus allowing us to compare footage without needing to re-create the same conditions on the same track. For those who don’t know, all replays in their default view are rendered in real-time at 30fps, whilst playable viewpoints are rendered at 60fps. Just like when racing.

For the most part, and rather impressively, Gran Turismo 5 does run at a near constant 60fps, with only a few dips in places when the engine comes under stress. At these points the game suffers from some short bouts of screen tearing along with few drops in framerate lasting a few seconds – sometimes at the same time.

This happens in both 720p and 1080p, and our initial impressions found that overall performance is far less stable for the later. However, when playing the game across a range of different tracks with varying car counts, we can see that factors other than raw resolution play a much greater part in impacting overall smoothness. Obviously, when running in 1080p on some tracks – Rome and other city courses specifically – the framerate is frequently more unstable compared to the same scenario running in 720p. But on others, it remains remarkably close, if not identical.

Another factor to consider is the number of cars on screen at once. Not necessarily the total number of cars in the race, but how many can become bunched up in a particular area when racing. This seems to be the root cause of most drops in performance - when the screen tears massively and the framerate gets cut down in half to 30fps. Quite often tearing is accompanied by dropped frames, thus resulting in a noticeable reduction in controller responsiveness and what looks like a juddering of the image on screen.

Thankfully, these dips in performance aren’t a domineering presence, and even in the densely detailed city tracks, so long as there isn’t a whole group of cars bunmched together, the frame holds steady for the most part. Tearing on the other hand usually comes with a screen stuttering effect regardless of whether or not frames are dropped. But even this isn’t a frequent occurrence.

What is impressive, is that PD have managed to get GT5 running at a mostly stable 60fps across both display modes whilst pushing around a large amount of alpha on screen, along with several highly complex, high poly car models. The use of a full 720p FB with MSAA, and 1080p with QAA plus TAA at 60fps is surely pushing the RSX and PS3’s pixel fill-rate through the roof. So, seeing such consistent levels of performance is a real testament to PD’s coding team and the engine they’ve managed to create in working within such tight constraints.


Moving away from performance and back to the make up of the visuals instead, we can see that GT5’s lighting and shadowing is made up firmly of both small real-time elements, along with plentiful use of pre-baked techniques. Like with GT’s 3&4 on the PS2, the bulk of GT5’s lighting and shadowing effects are pre-baked onto the surrounding environments through the use of shadow maps and light maps.

Whilst this does mean that there doesn’t seen to be a great deal of on-the-fly dynamic changes to the overall look of the game’s lighting conditions, the effect is mostly very convincing. The reflections on the cars for example, are as beautifully implemented as they are elegantly done, whilst the environmental lighting and shadowing model show how good art design can often take the place of having the most advanced technical solution. They do a good job of blending the two together when concentrating on driving around the track, and not when deliberately picking out technical details.

However, GT5 also features at least one noticeable light source which changes the amount of intensity and direction of the lighting on the cars whilst driving along the track. This appears to be implemented in a very similar way to the PS2 GT games, in the sense that it doesn’t effect the lighting or shadow intensity on all parts of the environment – it doesn’t seem to change the composition of the pre-baked parts of the engine, only the cars depending on their direction from the light source.

On the other hand, some racks feature day to night time changes, and in these parts of the game the impact of GT5’s lighting engine is felt far more strongly. Inclusion of extremely intense bloom lighting when driving out of certain tunnels into the sunlight also delivers a solid faking of proper HDR effects.

This use of what actually looks like HDR lighting (the bloom) is in fact a more limited, compressed version of the technique. Quite which one, I’m not sure. But certainly we’re not looking at full FP16 goodness here. Instead something approaching a wider range of bloom that complements the baked nature of a large amount of the lighting and shadowing on show.

On the whole, whilst looking a little static, PD’s choice of lighting schemes was clearly the right one for the game. The use of baked and subtler real-time elements works very well, while this also allowing for the cars to shine. The way they are lit and shaded in general is incredibly impressive, and the environment reflections play a strong part in delivering the feeling of realism that is essential to the series.


The same can also be said about the graphical make of the environments as well. Most are made up of a simple combination of multi-textured, shaded and lit geometry, with some basic use of shader effects (specular, diffuse etc) and some transparent alpha coverage and A2C. The lack of large amounts of multi-layered transparent polygons in making up the scenery (most of the foliage for example are simple 2D sprites) can occasionally break the realistic illusion PD are going for, with some parts of the track appearing rather flat and quite dated.

However, on others such as, and in particular Monza, the trackside detail can be pretty convincing despite not using a full 3D solution. Photo realistic texturing, and the baked lighting also makes a large difference. The City tracks of course benefit from having loads of fairly low poly, but beautifully textured and filtered buildings.

Obviously, there is a noticeable difference between legacy courses and the ones created specifically for GT5. In particular, we can see the likes of Deep Forest using a large amount of old untouched textures mixed in with new or reworked objects. The old stuff looks pretty ropey at times, although parts of it still appear to be nicely blended with the rest of the scenery. The reduced amount of sub-pixel and shader aliasing in these tracks do go some way to making up for that, even if more work could have been done.

On the other hand, the courses made specifically for GT5 looks far, far more impressive. We still see the inclusion of some low res texturing and flat 2D imagery used in quite prominent places, although the overall composition is noticeably superior. Photo-realism is well preserved, and the overall look is a little more natural in its appearance.

Ultimately, more of the engine’s budget (processing power and rendering time) could have been spent on the environments, but at the expense of having as exquisitely detailed car models. Some of the stuff like the foliage itself would have required heavy use of multiple layers of transparent geometry – something which greatly eats into the RSX’s available bandwidth, which would impact strongly on other parts of the game. Instead, PD’s focus has been on the cars; creating the most realistic looking vehicles seen in any game so far, whilst also delivering a fair compromise with the environments where they race.

The balance between having more detailed scenery and less impressive car renderings was always going to be the subject of debate. However, for GT5 this was another required compromise in seeing the team’s vision come to life. The cars were always the star of the show when it comes to GT, and quite rightly that’s where a lot of resources have gone.


Speaking of the cars themselves, and GT5 has just over a thousand of them. 200 of these are high-end premium models designed and modelled specifically for this game, whilst the other 800 or so have been ported directly from GT4, known as standard vehicles.

The premium models have been lavished with a degree of detail and a level of care and attention rarely seen in other comparable racing titles. The bodywork has finite amounts of detail in general, and the shader effects used to create their beautifully shiny exteriors are simply sublime. Seeing these racing around the track with environmental reflections is one thing, but in photo mode their true level of precision becomes apparent.

Sadly, the same cannot be said when talking about the ‘standard’ model variety. Seeing as these have simply been ported over from GT4 with nothing else been done to them, other than being rendered in HD, they look decidedly poor in comparison. Quality seems to vary between cars though, with some looking pretty good (even when viewed in replays) while others are pretty awful in general. You can clearly see this in the shot of a standard car below.


Other than what looks like better specular highlighting across the cars, nothing else has been done to improve their appearance at all. Here we see plenty of low resolution textures and no bump mapping of any kind. Even the environmental reflections are of a simpler standard. There’s no damage modelling either. While it might be a bit of a stretch to see full blown normal mapped exteriors helping to make up for the deficit in geometry, some reworked textures would have been nice in lessening the overall gap between the two sets of cars.

Thankfully, most of these differences can only be seen when viewing the standard vehicles in replays, and not so much to when racing. Driving while using the chase-cam external view, the lower quality nature of the ‘standard’ cars is barely noticeable. The difference tends to stand out the most when up next to premium models, or in a scene whereby the game’s lighting has a greater effect on the looks of the cars.

However, as these standard cars are only available via the used car shop, chances are you’ll rarely be using one as anything other than a starting vehicle. With a varied selection of 200 premium cars from which to choose from, it’s unlikely that you’ll want to go back given the range available.


Still, one can’t help to wonder why PD thought that it would be a good idea to have such a discrepancy in car quality marking down the overall focus on perfection and visual beauty seen throughout the game. 200 highly detailed cars is more than enough given that most players will only use a handful of these on offer. Usually, what you’ll find in a racing game is that the quality of the cars on offer is what counts, and not the sheer number available as a whole.

Another option – and one that is entirely plausible, if not almost guranteed to be the case for GT6 - is to re-use the 200 premium models again for the next instalment of the series. Maybe even if that means next-gen. PD already have a large variety of obscenely detailed cars which look beautiful in 1080p – and we’re not likely to seeing an increase in resolution beyond that for some time – so it certainly makes sense to take advantage of this fact. Perhaps they could spend more time, and indeed shader power, on getting the environments and trackside scenery up to the same level.


Lastly, one area we don’t often cover in our analyses, is the impact of loading and install time on the overall play experience. But in GT5, seeing as the amount of loading present definitely intrudes on the whole, it definitely had to be mentioned.

GT5 comes with an optional 6.5GB install function. And while you don’t need to take advantage of this fact, I strongly suggest that you do. Installing the game on the PS3’s HDD not only cuts down load times significantly, but also allows for shorter, permanent installs to take place instead of longer temporary ones when playing without.

Without performing the core 6.5GB install, the game will need to both load and install data for a large variety of components. The game’s menu screens need to be installed, as does car preview information (cached off the hard drive) and the courses themselves… every time you access them. This brings the experience down to painfully slow levels, in which it takes easily over a minute to load up a track, let alone enter a mode, pick a car etc, on top of that. The fact that these installs are only temporary means that while you save on HDD space, loading times are permanently high.

Installing the game then, is the only realistic option. This can take between 35 and 50 minutes, depending on which model PS3 you have (HDD cache amongst other things apparently affect this), but does significantly reduce the overall time it takes for parts of the game to load.

Like before, GT5 needs to install data for various elements independently from the main install, although it only needs to do this once. And this takes only a mere two or three seconds, without impacting on general loading times. But unlike before, loading times are dramatically cut down: we’re looking at around half the time to loading in the courses, and the cars are now quickly streamed off the HDD when going to view them.

Arguably, PD’s system for managing game data on the PS3 is somewhat disappointing. Even after the main install we still see the occasional install messages pop up from time to time while accessing the same course we’ve already been on several times before. Plus, the menu screens still take quite a noticeable amount of time to load up. Quite why PD didn’t decide to have these completely installed on the hard drive is beyond me. It would certainly give the whole experience a more natural flow, instead of feeling a little disjointed.

Either way, the bulk 6.5GB install does indeed work well. Although, it has to be said that it doesn’t completely mitigate the constant loading issues still present in the game. But to be fair, without changing the make up of the game, the menus, and how data is handled as a whole, little else realistically could have been done. A full game install would indeed sort this. But that is hardly a feasible option, considering the vast nature of the content available and the size it would take up.


In conclusion, after over five longs years of waiting and an estimated $60 million plus in terms of development costs, Gran Turismo 5 is a great example of technical mastery and artistic vision blended beautifully together. Although, that is not to say there aren’t any problems – the poor quality shadowing and low res effects spoil things somewhat, while the baked nature of the lighting doesn’t impress as much as competing titles dynamic solutions. However, these tradeoffs can’t, and indeed don’t take away from the incredible feat the team at Polyphony Digital have performed with regards to achieving 720p and 1080p performance at 60fps, with good use of anti-aliasing to boot.

Given the massive bandwidth issues with then PS3 architecture as a whole, the range of compromises seen throughout simply help to highlight the mammoth feat that has been performed within. The individual components of the rendering engine as a whole may well seem lacking, or slightly behind the times. However, in combination, when put together to form the final product, the final image, they work exceptionally well, whilst also delivering a game that at times is simply gorgeous to look at.

Of course, the cars themselves are virtually flawless. Most of our qualms come to down to the way the environments are rendered, how the effects are handled etc. But nothing impacts on the visual splendour of the work done by PD on the premium vehicle modelling. The sheen on the near-perfect, better looking than real life exteriors, to the intricately detailed stylings of the bodywork - everything has been delivered with a finesse and polish rarely seen outside of high-end CG, let alone in any other game to date.

Visually, the GT series has always handled specific elements in a certain way, and GT5 is no exception to this rule. The developers at PD have carved a fine line between technical compromise and artistic creativity, perhaps eschewing purely advanced rendering techniques for something a little simpler. But here, as before… it works. Sometimes exceptionally well, sometimes not. Either way, it seems like the right choices were made at the right time for this particular title.


Outside of graphical concerns and achievements Gran Turismo 5 delivers a truly in-depth, accomplished experience on and off the track. The realistic, albeit stylised handling feels incredibly good, with plenty of customisation available from tyre changes to suspension tweaks, whilst the huge range of events and different things on offer provide plenty to see and do. The special events are great for earning money, and the B-Spec mode provides an ample way out for those looking for looking to cut down the grind.

Also, the levelling system and the unlocks it provides keeps drip-feeding you new features and things to try out, thus keeping things interesting for anyone looking for more stuff to do. And there are lots: from Go-Karting to Nascar, to Rallying and Drift Challenges, there’s plenty on offer for everyone.

The only downsides come in the form of GT’s own style of convoluted menu system, and RPG-like grind required in order to gain entry to specific events (without using B-Spec mode). In which case, more could have been done to bring this somewhat dated system up to scratch for 2010, because it feels positively archaic – like it belongs as much in a PS2 instalment of the series more than it does here. The long loading times as well, often break up the action in a way that makes the game feel disjointed.

Otherwise, GT5 is a superb, expertly crafted game, one that would perhaps have benefited from being developed outside of its own protective bubble and into the wider world instead. That is to say, it is most certainly well worth picking up, representing another true generational leap for the series… even if it doesn’t quite make it through completely unscathed.

Many thanks go out to both AlStrong and Quaz51 for their pixel counting skills and aaditional insight into TAA. Plus Cynamite.de, Gamekult, and Taxi Gamer for the screens.

Monday, 22 November 2010

Tech Analysis: Assassin's Creed: Brotherhood (PS3 vs 360)

There has always been quite a considerable gap in between the PS3 and 360 versions of previous Assassin’s Creed titles. Whilst most of the core make-up of each game was in fact pretty much identical, sans occasional differences, both performance and image quality lagged behind on the PS3. So much so that Ubisoft themselves ublicly recognised this, with Level Design Content Director, Phillipe Bergeron, acknowledging that more could be done in a recent interview with IGN.

"At the end of ACII we realised that the PS3 was sort of an afterthought – or, not that it was an afterthought, but we hadn't fully debugged it until the very end, and we had a bunch of frame rate issues and quality issues. This time around we knew that, because we went through it once, so we decided to attack it from the beginning and I think the final product is much more on the level, and even on some parts, the frame rate is probably sometimes better on the PS3 than it is on 360."

The question is though. Did Ubisoft Montreal actually follow through, or are we left with another disappointing PS3 port, complete with noticeably worse performance and a smeary Vaseline-styled look? The answer in fact, may just surprise you. As although Assassin’s Creed Brotherhood shares much with its predecessors, it is certainly a more polished affair in which the PS3 version stands up remarkably well with its 360 counterpart. It is for the most part, bar some texture blurring and a contrast/gamma difference, identical.


360


PS3

As expected Assassin’s Creed Brotherhood is rendered in 720p (1280x720) on both formats, with 360 getting the standard issue 2x multi-sampling anti-aliasing (MSAA) as usually seen in titles on the platform, and the PS3 once again receiving the alternative quincunx (QAA) solution.

As we’ve mentioned before in previous tech analyses, the very nature of how QAA works in smoothening jagged lines means that the entire image, including textures, gets blurred to some extend. Unlike with MSAA, QAA works on applying the smoothening algorithm to every pixel and not just specific edges.

Essentially edge-based pixels are still sampled in a similar way, however QAA uses a five-point sampling pattern which inconveniently works on all areas of the image regardless of whether an edge is present or not – pixels in both low and high contrast areas are equally affected, which is the main cause of textures becoming blurred as a result.

The choice of using QAA over traditional MSAA then, is rather strange to say the least, considering the technique comes with roughly the same processing and memory cost as 2xMSAA. But the advantage it seems, is with it being able to deliver ample edge smoothing closer to that of 4xMSA, though at the expense of overall scene clarity.

The most obvious reason for its use probably stems from an artistic decision rather than a purely technical one – there’s no reason why MSAA couldn’t have been implemented, so it’s likely that the developers actually wanted to have 4xMSAA type levels of edge smoothing on both platforms, but without any easy way to do so on 360 (you would have to use tiling). The PS3, naturally, has QAA as a standard form of anti-aliasing not included in the 360’s GPU feature-set (it’s an NVIDIA thing), so represents an obvious compromise.

However, compared to some games that use the technique, the QAA in ACB has less of an initial impact in overall image quality than you might expect. Especially seeing as the PS3 build’s 720p output remains fairly sharp and continuously crisp despite additional texture blurring.



Unfortunately, most of our comparison screens for ACB are rather compressed whilst suffering from obvious lack of proper gamma adjustment on the console end, which makes showing the clean appearance of the PS3 build and the extra sharpness of the 360 game rather difficult. As a result we’ve left all the screens untouched - free from additional compression induced labelling, whilst also bringing you two much higher quality PS3 screens in which to demonstrate out findings.

With the two shots above you can clearly see just how sharp the PS3 version really is, with the use of QAA providing a decent level of jaggies reduction without compromising the clarity of polygon edges. Some texture blurring is apparent, which is perhaps the biggest bugbear I have about the technique. But as you can see its affects aren’t especially displeasing - not in every instance - and as a result ACB still looks incredibly good on the PS3.


360


PS3

Despite all our images showcasing what looks like reduced quality texturing in the PS3 build ACB, you can see that the main reason for this is a combination of both the additional blur provided by the use of QAA and the drastic difference in gamma curves for both versions.

This gamma difference is half of what makes the PS3 game look less detailed on first impressions, with textures that could be misconstrued as being in a lower resolution to those not aware of how QAA imapcts on the final image. However, the actual assets used in the game are actually like-for-like, and you can see this when both versions closely scrutinised.

Instead, it is the initially higher contrast and washed out nature of the PS3 game, in combination with the QAA which helps in hiding texture details, and making the filtering come across as looking worse, which in fact, is actually identical (same levels of AF present on both). Calibrating both the brightness through the game’s own menu, and gamma on the HDTV itself practically solves the problem, with the PS3 version looking clean and sharp with more visible detail being present after this is done.


360


PS3


360


PS3

You can clearly see this above: the in-game brightness setting has been adjusted in order to provide a more uniform look across both consoles. HDTV settings haven’t been touched in these two phone captures, instead showing that a similar level of brightness can be obtain by simply changing an option in the game’s menu.

For the most part, like with Need For Speed: Hot Pursuit and indeed past Assassin’s Creed titles, both versions of Brotherhood look pretty much alike in the vast majority of areas, with any differences coming across more as mere quirks and rendering oddities than anything else. Other than the use of QAA and varying gamma curves, only performance really separates them in any meaningful way. And even in this regard, ACB has seen some noticeable improvements on the PS3 side.


360


PS3

In terms of performance past Assassin’s Creed titles have always been behind on the PS3, with more instances of slowdown and much greater amounts of screen tearing. Although in Brotherhood the gap has indeed been closed significantly, with less in the way of either taking place. Naturally, it is the 360 build which still commands an advantage, but both at times, feel and look very similar in this regard.

When looking at both 360 and PS3 versions of most games, it is clear that developers usually try to balance out the use of v-sync with trying to maintain a smooth framerate. Normally, PS3 owners are privy to a near solidly v-synced experience at the expense of a large increase in slowdown. Whilst on the 360, developers usually choose the opposite: ditching v-sync in order to allow for a smoother experience, but with noticeable amounts of screen tearing.

For ACB, like with past AC titles, Ubisoft Montreal have favoured the latter, in which case it is apparent that neither version employs v-sync but both can run relatively smoothly on many occasions. ACB targets a 30fps update, and the framerate is capped at that level – it never goes higher than this, but it does drop below.

However, both versions actually maintain a reasonably solid 30fps most of the time when load isn't being pushed – noticeable drops only really occur in situations where long draw distances are visible, or in areas in the city where crowds converse together. In these sections the 360 version does run smoother, featuring less prolonged dips in framerate (sometimes only by a few seconds or so) and less screen tearing. Although, on most occasions the two games operate near identically, with both dropping frames and tearing terribly at similar points. Bar perhaps the odd point in which the PS3 version felt a little smoother for a brief moment in time.

One thing that is apparent, is that the game on both platforms can suffer from regular, and continuous bouts of screen tearing even when the framerate appears to be mostly solid during the experience. In that respect, it is all too obvious that this concern from the first two AC titles hasn’t been fixed at all. The PS3 version tears more often than the 360 one, and both tear regularly in heavy load situations. Though perhaps this is something that we simply have to accept in order to gain better performance via a smoother overall framerate.

In the end ACB does display an improvement in this area on the PS3, although not quite to the extent we expected given Ubisoft publically released statement. Tearing is still an issue – more torn frames on the PS3 – and the framerate at times still struggles in a scenes with high detail and many characters on screen – on both formats no less. In which case it is obvious that despite some upgrades and optimisations, the PS3 version is still a little behind in terms of overall performance.

Saying that, there are often times where both versions are basically close to being like-for-like, and the differences during gameplay can be so subtle that they can regularly go unnoticed (between both formats). Screen tearing aside, both versions are reasonable performers, with the engine obviously struggling in situations where the overall load exceeds the capacity for it to be resolved.


360


PS3

Moving on, and we can see that the engine powering ACB on both platforms has seen a few steady improvements in the lighting and shadowing departments, along with other additional effects – the water for example, in some places, now looks to be made up of more than just a few texture changes.

Dynamic lighting and shadowing has been expanded upon, and the game features noticeable cloud coverage providing moving shadows which seem to affect the lighting and shadowing on the ground. All of this is done in real time, and actually accounts for some of the shadowing differences you can see in some of the screens.


360


PS3

The most obvious improvement comes with the inclusion of screen-space ambient occlusion (SSAO), which adds an extra level of depth to the scene. Use of SSAO clearly expands the shadowing properties used throughout most of the game, and can be found noticeably on characters, and some parts of the environment. Along with the use of both dynamic and static shadowing, the use of SSAO helps to bring a more defining, realistic quality to the entire scene, with ample balance between areas of the environment with and without the effect.

Also in terms of shadowing, one thing we did notice was that certain shadows feature a slightly dithered look to them, much like what we were seeing in Mafia II. The effect stands out a little more on the 360, along with the shadowmaps themselves, which look sharper as a result of no QAA blurring. However, it also appears that the PS3’s use of QAA actually provides better blending with regards to the dithering effect - it becomes less obvious as a result.


360


PS3

In motion it is also possible to see some LOD issues with regards to the game’s use of shadowmaps - whereby shadowmaps feature a transitional change from lower to higher quality as you get closer to them - along with LOD issues on both platforms in general. Parts of the environment (textures, geometry, and shadows) in both the far distance and from a few feet away, tend to pop up noticeably on occasion as the engine struggles to load them in time. Though given the large draw distances it has to handle, this is understandable.

There are a few other differences to be found, but nothing major, or even anything that would really account for a clear rendering choice difference. Some shadowing and lighting oddities occasionally pop up – such as missing baked lighting on the PS3, or shadows appearing and popping in and out where they shouldn’t be. But this stuff isn’t noticeable whilst simply playing the game.


360


PS3

In the end Assassin’s Creed Brotherhood is pretty solid on both platforms with the 360 version maintaining that technical edge in certain areas. On the other hand, with the exception of more screen tearing and the inclusion of the texture blurring QAA, the PS3 build is almost a good, having the same baseline assets and tech powering the game.

Even the use of QAA is no real blemish on the experience – calibrating both the game’s brightness level and the gamma on the HDTV makes overall image sharpness fairly close to that of the 360 game in this regard. The extra edge smoothing that quincunx provides can also create a more organic look to the overall visual make up of the game as a whole, with only some scenes looking noticeably worse off than others in terms of texture blur. In any case the PS3 version can be almost equally attractive if set up properly, even if the 360 game's additional sharpness is preferable.

Pretty much every aspect of the game in other areas is a like-for-like match, with only the odd rendering bug to separate them, and of course the 360 version’s lead in overall performance. But even that isn’t quite as commanding as with previous instalments – despite the PS3 build suffering from more noticeable bouts of screen tearing, the general framerate is pretty much in the same ballpark for both versions, with the 360 only fairing a little better in most cases from what we’ve played.

As to whether Ubisoft have delivered on their promise to provide a thoroughly more optimised, de-bugged PS3 experience. I think that as a whole they have. While perhaps not eradicating all of the problems found in previous titles, the overall result is far, far closer than before, with the 360 code no longer having a significant advantage. In short, there could be more work done to improve performance, but otherwise what we have here is a clear step forward in the right direction - a solid result in delivering a decent multi-platform outcome.

Ultimately, ACB can be comfortably recommended across both formats with your purchasing decision more likely coming down to which controller you prefer to use, or which format your friends play online with. The added inclusion of exclusive DLC for PS3 owners is yet another thing to consider, if a choice is indeed available to you. Either way, I’m sure most users will be happy whichever version they opt to go for.

As always, many thanks go out to AlStrong for the pixel counting, and to Cynamite.de for most of the screens. The full gallery can be found here.

Sunday, 30 May 2010

Tech Analysis: Red Dead Redemption (360 vs PS3)

There’s no doubt that Rockstar are onto another success story with Red Dead Redemption. Piece together the same kind of open worldliness as in Grand Theft Auto IV, along with some traditionally styled old west action, and you have another finely crafted and life-consuming experience. If there were ever a game to make your Clint Eastwood fantasies come true, then RDR would be it.

The underlying engine behind the game is based on the same impressive tech that powered 2008’s GTA IV, expanded upon, tweaked and refined, perhaps pushing the most it can get out of the current consoles. Well, without a complete rewrite that is. And like with the GTA IV conversion, there are many similarities between the two games. Everything from the rendering resolution and the use of anti-aliasing can be plucked right out of Rockstar’s last multiplatform title, so it’s no surprise that some of what you’ll be hearing today is at least vaguely familiar.

Starting off, as always with the rendering resolution. RDR is presented in 720p (1280x720) on the Xbox 360 using 2xMSAA (multi-sampling anti-aliasing), whilst on PS3 is rendered in a slightly lesser 1152x640 and using the alternative QAA (quincunx anti-aliasing).

Straight away you can see that Red Dead is exactly the same as GTAIV in this regard, with the 360 benefiting from decent edge smoothing, and the PS3 having to make do with the blurrier quincunx solution. This choice appears a little odd, as both 2xMSAA and the standard QAA take up roughly the same amount of memory and processing requirements. So why use the poorer QAA option in the first place?



Well, the most obvious explanation is that the QAA in theory delivers more edge smoothing for the same memory cost as the conventional MSAA, so thus there should be less in the way of jagged edges. However the consequence is that QAA not only smoothens polygon edges, but also blurs general pixel details in the entire scene. So textures along with geometry become blurred, creating a softer overall image. Bizarrely the use of this QAA seems to be selective, with some edges getting clear levels of jaggies reduction, and others without.

Unlike in GTAIV, in which the blur effect actually added an ambience and a layer of atmosphere to the proceedings, in Red Dead it does nothing of the sort. The gritty environments of GTA made the QAA look fit in well with the game world, and the 360’s dithered textures made the loss of sharpness a worthy compromise on PS3. However RDR on 360 doesn’t really suffer from any dithering textures. Or rather, that the effect has been greatly reduced, and the bright and dusty open world nature of the game benefits from a sharper image, meaning that the blur effect simply heightens the differences between the two versions.

The 360 build however, isn’t free from criticism either. The high contrasting edges between the buildings with the bright blue sky often leads to noticeable shimmering, sometimes as bad as seen in the PS3 build. Most of the time these jaggies are in fact smoother than the ones seen on its PS3 counterpart but shimmer in the same way.

Essentially MSAA works by taking samples from two adjacent pixels, and then combines those samples with the final image to form an anti-aliased one. With edges in high contrast areas, there isn’t enough information to create decent samples to act as in betweens where the two pixels meet, thus the image receives either no AA altogether, or a significantly reduced amount.



In terms of texture detail and filtering, both versions are identical. Texture resolution seems to be the same on either platform with the lower 640p resolution of the PS3 game resulting in some additional blurring of the visible detail. This is caused by both the upscaled nature of the framebuffer, and the use of QAA, in which we’ve already discussed the technique’s unwanted side effects.

Anisotropic filtering is evident in the screenshots below, as is the blurred nature of the PS3 version’s textures, and general screen composition. The levels of AF don’t appear to be much higher than what is usually available with the older trilinear method of filtering. So we are looking at perhaps 4x AF for Red Read on both platforms.



One area of the game that has been noticeably paired back on the PS3, is the heavy use of alpha-based foliage. This seems to be the main reason for the lower resolution of the PS3 version and it’s use of a more aggressive LOD (level of detail) system in certain scenarios.

As we’ve discussed before at IQGamer, the use of alpha heavy transparency effects for objects like hair and foliage use up a huge amount of bandwidth, a commodity which is particularly limited in home consoles, but especially on the PS3 with its lack of high speed EDRAM. It’s this bandwidth advantage which allows the 360 to not only render more foliage on screen, but also render it at a higher resolution as well.

These two shots show exactly how much of the foliage has been cut back on the PS3 game. Notice how it is mostly only the smaller plants and grasses that have been culled, which actually has less of an impact during play than you might expect. However it’s also pretty clear that the environments in the 360 version look visibly more dense as a result of having loads more minor pieces of foliage.

Some major parts of the foliage have also been cut on PS3 too. When this happens, the Sony version of RDR looks much blander as a result, taking away some of that ‘living world’ look that the game at times possesses.



Whilst both versions use A2C (alpha-to-coverage) for the surrounding foliage - a memory saving technique for rendering half resolution transparencies in an interlaced type manner - the 360 version clearly benefits from better alpha blending through the use of A2C in conjunction with 2xMSAA.

On the other hand, the PS3’s use of QAA and low-resolution foliage makes for a poorer blend overall, and some unsightly shimmering artifacts which don’t look too great with the blurred nature of the framebuffer.

As regular readers of this blog might know, A2C also creates a screen door effect on all transparencies and textures that use it. MSAA is used to blend away this unwanted effect, and the more AA used, and the higher the resolution of the A2C textures, the better the overall result will be. For RDR the PS3 game and its use of low-res foliage and QAA simply makes the screen door effect stand out far more than it does on the 360. By contrast, it’s barely noticeable in motion in the 360 build.



Along with a reduced foliage count, the PS3 game appears to feature a more aggressive LOD system as well. It’s not so much the issue of objects being cut back on the PS3 build, but also of objects popping into view later than in the 360 one. Geometry changes, shadow pop, and object pop in are all more noticeable on the PS3. Although these do occur on 360, but to a far lesser extent.

Below you can see the differences between the two versions. Notice how the water effects and shadowing have been cut back on the PS3, with entire shadows missing altogether. Detail in the distance has been deduced, with buildings featuring simpler levels of geometry, and with some objects being completely omitted from the scene.

The PS3 version also suffers from some other LOD issues as well. In towns and small outposts, as you approach the buildings the shadows pop in a lot later in the PS3 build compared to the 360. They appear more erratically on the PS3, whereas on 360 they appear smoothly on screen from further into the distance, and in a far more succinct manner.

The reason why, on occasion, that so much detail has been reduced in the PS3 build, is due partially to the game’s lower resolution in combination with the blur inducing QAA, and what appears to be a more forceful LOD system.

Either way, the Sony game suffers far more from LOD issues and the overall more aggressive LOD system in place for RDR. Unlike in GTAIV, the whole game has been designed around sprawling vistas, and wondrous views into the far distance. So its no surprise that some large compromises had to be made for the PS3 build, especially where trying to keep a more consistant frame rate is concerned.



Red Dead Redemption aims to keep a steady 30 frames per-second framerate at all times, but both versions deal with doing this a little differently. Like with many multi-platform titles, the PS3 build is focused on eliminating screen tear at the expense of maximum smoothness, being v-synced that is, and the framerate also capped at 30fps maximum to ensure it never goes above this level.

The 360 build on the other hand is more concerned with maintaining as smoothest framerate as possible. And to that end the framerate hasn’t been capped at all, with the game running between 40 to 50fps in certain scenarios. This mainly happens when on foot, and in enclosed locations without much in the way of foliage and environment detail.

Most noticeably is the fact that the 360 game isn’t v-synced, or v-locked at all, meaning that it is prone to screen tearing in stressful situations or fast camera pans. The tearing only ever appears at the top of the screen, and stops halfway across the screen, almost like the game has caught up with the problem.

By contrast the PS3 game never tears any frames at all, remaining completely free of the problem regardless of what is happening on screen. However, it does drop framerate badly in heavy load situations, far more than the 360 game. When this happens, the screen can crawl with jagged edges and shimmering foliage in areas in which the QAA isn’t applied, and where the A2C blend fails to work successfully.

Both versions do massively drop their framerate in heavy load situations, but it’s the PS3 one that seems to be more greatly affected by such dips in performance.

So it is safe to say that performance wise the 360 version has the edge, and that the screen tearing present on that version of the game isn’t much of an issue, being barely noticeable for much of the time. By contrast, the PS3 build features a small drop in IQ over the already worse 640p resolution and use of QAA whenever the framerate drops. That said, neither version maintains the target 30fps for long, with constant dips below all throughout the game.



Moving on, we can see that in terms of lighting and shading there are differences in both versions of the game. For the most part they are both largely identical, but in certain circumstances the PS3 build actually features more light reflections and extra shadows cast in indoor areas, and the 360 build gets self-shadowing on all characters, which are absent for some on PS3.

The lighting differences don’t always seem to be technical achievements, but rather technical anomalies, in which it seems more likely that there is a rendering error on the 360 build compared to the PS3 one. Although we cannot be sure as to why this is happening without knowing the ins and outs of how Rockstar’s engine renders its shadows and lighting in detail.

In the screenshots below you can see how the PS3 version is casting a light source from outside and through the window, into the scene indoors. Whereas on 360 it is clear that the only light source affecting the characters is the one coming from the wall-mounted lamp.

The light source that is cast through the doors and windows on the PS3 game also casts shadows from the characters and onto the grown. Something that is also absent from the 360 build, which again only gets shadows from indoor lighting.

Thankfully these scenes are few and far in between as most of the game is set outside, in which there are only a few minor cloud shadow oddities in the 360 build. However it does mean that in indoor scenes the PS3 clearly demonstrates better use of lighting, whether that be due to an error or otherwise.



The 360 build however, does have the benefit of having both sharper and higher quality shadows for character and environmental objects. All characters on 360 have the benefits of using self-shadowing – shadows that are cast upon characters by themselves - which gives them an extra depth and three-dimensional look.

On PS3 characters generally look flatter than their 360 counterparts, with some lacking self-shadowing altogether, and others simply having the effect paired back over the 360 build.

It’s pretty obvious in the screenshots below how superior the shadowing can be in the 360 game. Notice how much extra in the way of depth the self-shadows can add over environment shadowing and lighting.



In the end, despite the differences we can still see that what Rockstar’s engine has achieved on both platforms is pretty impressive, with it’s ability to render miles of detailed scenery and still keep up a decent framerate outside of strenuous encounters. Red Dead Redemption may not be artistically pleasing to everyone, but is still a mean technical achievement. It is however, one which favours the strengths of Microsoft’s console, with the 360’s superior vertex processing capabilities and greater memory bandwidth.

From what we’ve uncovered today, it’s pretty clear that the 360 demonstrates superior technical prowess when it comes to handling the wide and open-world nature of Red Dead Redemption. Unlike in GTAIV, a far-reaching field of view is absolutely required in order for a game like RDR to accurately represent the time and period in which it’s set. And it’s also exactly the type of engine that is ill suited to PS3’s lack of bandwidth and vertex shader power.

It comes as no surprise then, to see that without large buildings and natural structures hiding far off areas, that Rockstar had little option but to noticeably cut back on the sheer amount of geometry present in the PS3 game. After all, you can’t start culling much in the way of unseen polygons when the game requires most of this detail to visible from far into the distance. Instead the only thing left to do is cut away at the foliage, sap away some of the resolution and try to make the best of what you’ve got left.

However most of these differences aren’t all that visible unless you’ve seen one version, and then moved on to playing the other. Someone whose only ever seen the PS3 game is more than likely not to notice most of the issues we’ve highlighted in our tech analysis, and will still enjoy the many superb experiences Red Dead has to offer. Perhaps the only thing that IS noticeable is the blurred nature of the game owning to the use of QAA. In that respect it is the game’s largest issue, and the one which takes away the most from the experience.

In conclusion, our recommendation rests with the 360 version. There’s simply no doubt that it’s use of a higher resolution, proper MSAA, better LOD detail, better shadowing system, plus more stable overall performance makes it the better of the two. This isn’t a case of GTAIV, in which the slightly blurred look adds to the visual style of the game world, but rather that every part of RDR benefits from having a generally cleaner and clearer look about it, with more detail and a smoother framerate all gelling it together.

However, that said there’s no reason for people with access to only a PS3 not to pick up Rockstar’s latest. As we’ve stated above, that unless you’ve seen the two versions running you’ll unlikely to notice most of the differences, and the game itself is the same on both platforms, being an utterly absorbing experience at times, and one of the best open world games to date.

Thanks to MazingerDUDE for the majority of our comparison screenshots, and as always, to Quaz51 for his exemplary pixel counting skills.