Dragon Age Inquisition: The Highs and Lows of AMD’s Mantle

The latest game in the Dragon Age series launched last year to critical and popular acclaim. Dragon Age Inquisition from developers BioWare and published by Electronic Arts won plenty of awards for 2014, even though it came out towards the end of November, and it is undoubtedly one of the best RPGs of recent history. With a current Metacritic rating of 85%, it's by no means perfect, but if you're a fan of the genre or series picking the latest release up is a no-brainer.
Except you might be wondering if you actually have enough graphics power to handle the game in all its glory. BioWare also updated the engine from their own Lycium/Eclipse engine used in the earlier titles to DICE's FrostBite 3 engine, of Battlefield 4 fame. Besides supporting advanced rendering features, FrostBite 3 is one of the few engines to fully support AMD's Mantle API, and that means most games that use the engine will by default support Mantle. So how much does Mantle help with performance, and what sort of hardware do you need to run Dragon Age Inquisition? The answers end up being a bit more complex than you might suspect.

Dragon Age Inquisition - What's Going on with Mantle vs. DX11

Before we get to the actual results, we need to clear the air a bit. Dragon Age Inquisition actually ships with a built-in benchmark of sorts -- or at least a benchmark mode was added with one of the early patches. That was a pleasant surprise initially...except the built-in benchmark is complete crap. The actual scenes are fine, but it appears that the benchmark mode renders a set number of frames as fast as possible. The problem with that approach is that where the benchmark might take 20 seconds on moderate hardware, it's over in about 5 seconds on a pair of fast GPUs, and that's simply not long enough to give us good information. This is exacerbated by the fact that the game tends to take about 10 seconds for performance to normalize on certain GPUs (CrossFire DX11 mode in particular), which means the benchmark is done before frame rates are anywhere near stabilized! If you use the built-in benchmark to compare DX11 and Mantle performance, Mantle will incorrectly appear to be over twice as fast.
If that were the only issue with Mantle, however, things wouldn't be too bad; unfortunately there are plenty of other problems. First, we test with a 4K display for our gaming benchmarks -- the Acer XB280HK specifically. The first issue we encounter with Mantle is that for whatever reason, resolution scaling appears to be broken, so if you run at 1080p instead of 4K you end up with black borders occupying three quarters of your display area. It looks like this:
After seeing this problem and thinking back, we recalled seeing similar problems on other games. A quick spot check of Sniper Elite 3 revealed the same issue when using Mantle. It could be something that AMD simply needs to fix in their latest Catalyst Omega drivers, or it might be something the software developers need to take charge of fixing as Mantle basically puts them in direct control of much of the graphics work; either way, it really needs to be addressed -- especially since gaming at 4K resolutions is generally not possible with a lot of titles right now (unless you drop to very low quality settings elsewhere).
So that's big problem number one with Mantle, but there's another issue we noticed with Dragon Age Inquisition: load times for the game and the initial level take quite a bit longer when Mantle is enabled. It's not horrible when you're only running a single GPU, but for CrossFire it reaches the point where we'd call it completely unacceptable. We timed the game load time on CrossFire R9 290X as an example: over three minutes to get to the main screen, and another five minutes to load the save game. We were ready to disable Mantle after two minutes, but we persevered for our benchmarks at least.
Needless to say, that's not a very good beginning for the game as far as benchmarking is concerned. A broken built-in benchmark, Mantle not scaling properly at non-native resolutions, and horribly long load times with Mantle. Lions and tigers and bears, oh my! But we were able to work around most of these issues at least, so let's move on to the benchmarks.

Dragon Age Inquisition - Performance and Analysis

The Dragon Age series has always been pretty demanding when it comes to graphics hardware, and at least at higher quality settings Dragon Age Inquisition is no different. The game has graphics presets which we've used for testing: Ultra, High, Medium, and Low. Interestingly, Ultra doesn't actually max out all settings. If you have the hardware, you can increase the tessellation quality one more notch, and there's a "Fade Touched" texture quality setting, which likely makes very little difference to quality but just doubles the texture size; you can also bump anti-aliasing up to 4xMSAA. For our purposes, however, we've stuck with presets.
Our benchmark consists of running through a scene with a bunch of NPCs visible, which tends to be one of the more demanding areas in the game and should be a good representation of expected performance. One final caveat: testing was completed in January 2015, using the AMD Omega and NVIDIA 347.09 drivers with the current version of Dragon Age Inquisition at the time; patches get pushed out periodically, and the NVIDIA 347.52 drivers just came out, so potentially things have improved a bit in the past few weeks.

Starting at the top with 4K Ultra gaming, right now nothing can handle Dragon Age Inquisition while delivering frame rates anywhere near 60 FPS. However, if you're willing to drop from Ultra to High settings, 4K High performs nearly the same as QHD Ultra in our limited testing, which is definitely playable. It's also a bit interesting that DX11 actually does better than Mantle at 4K Ultra; the built-in benchmark wouldn't have shown such a result, and it's still a bit odd that DX11 CrossFire performance starts low for 10 seconds and then improves by nearly 50%, but regardless you will probably want to stick with DX11 for multiple GPUs until/unless BioWare can fix the load time issues with Mantle.
Continuing on to QHD Ultra, here we have both GTX 970 SLI and R9 290X CrossFire hitting 60+ FPS, with NVIDIA taking a slight lead over the Mantle R9 290X CrossFire, which in turn leads by a small marging over DX11 290X CrossFire. Yes, you read the correctly: the supposedly gimped GTX 970 that "only" has 3.5GB of high speed GDDR5 performs perfectly well. Similarly, R9 280 CrossFire with Mantle has a moderate lead over the DX11 code path. While 4K gaming didn't seem to benefit from Mantle, the same cannot be said of QHD and lower resolutions. In fact, even the single AMD GPUs all show rather impressive ~15% gains at a resolution that is normally a GPU bottleneck.
Drop down to 1080p Ultra and the gains from Mantle become even more substantial -- the R9 290X is nearly 30% faster while R9 280X and R9 280 (R9 285) are ~20% faster. In this case, it's often the difference between stuttering and a 30+ FPS experience, though only the R9 290X is able to get above 60 FPS. Not bad for just an API switch, and it looks like DirectX 12 will deliver similar results later this year in a GPU agnostic fashion. NVIDIA's best single GPU meanwhile is playable at QHD Ultra, but the GTX 980 is only able to break 60 FPS at 1080p Ultra and lower settings.
Once we get to 1080p High settings and below, while Mantle still delivers a good increase in performance it becomes less necessary. The AMD GPUs are all hitting around 106 FPS averages with Mantle compared to 60-80 FPS with DX11, but both are totally playable so unless you have a display that can run at more than 60Hz it's mostly an academic question. The same holds for our 1600x900 Medium and 1366x768 Low settings -- and since resolution scaling doesn't work with Mantle, you wouldn't want to use these regardless. Besides the earlier issues with Mantle, however, it's not all roses and sunshine.
Look at the R7 250X, which we have in the 1080p Ultra and lower settings. Mantle helps the card a bit at 1080p Ultra and High, but 5% faster than "too slow" is still very much in the "too slow" classification. Interestingly, we also see a drop in minimum frame rates at 1600x900 Medium with the R7 250X, though average FPS improves by 10%. It's only at 1366x768 Low that the 250X really reaches fully playable levels, and Mantle provides a 25% performance increase there even though it ends up being less necessary. Of course, 1080p Low should still run well on the R7 250X, which is more than we can say from something like Intel's HD Graphics 4600.
Dragon-Age-Inquisition-IQ1-1-Max Dragon-Age-Inquisition-IQ1-2-Ultra Dragon-Age-Inquisition-IQ1-3-High Dragon-Age-Inquisition-IQ1-4-Medium Dragon-Age-Inquisition-IQ1-5-Low
Dragon-Age-Inquisition-IQ2-1-Max Dragon-Age-Inquisition-IQ2-2-Ultra Dragon-Age-Inquisition-IQ2-3-High Dragon-Age-Inquisition-IQ2-4-Medium Dragon-Age-Inquisition-IQ2-5-Low
Dragon-Age-Inquisition-IQ3-1-Max Dragon-Age-Inquisition-IQ3-2-Ultra Dragon-Age-Inquisition-IQ3-3-High Dragon-Age-Inquisition-IQ3-4-Medium Dragon-Age-Inquisition-IQ3-5-Low
Wrapping up with a look at image quality and settings, it has to be said that even at lower quality settings Dragon Age Inquisition is a nice looking game. It doesn't necessarily have all the fanciest options, but tessellation at least is present and adds a nice sense of realism to certain surfaces, while the lighting model is good as well. The difference between maximum, Ultra, and High quality are relatively minor, with the major change between Ultra and High being a loss in shadow quality and a reduced amount of tessellation. At Medium, tessellation is turned off completely and shadow accuracy drops again, but there also seems to be an increase in vegetation to make up for the drop -- look at the number of distant trees in the Medium settings, for example. Finally, Low as you'd expect turns down just about everything, so SSAO is completely disabled and models start to look flat, there's lots of aliasing visible, and texture quality on distant objects drops off a lot. Still, if you don't have high-end hardware Dragon Age Inquisition is still plenty of fun to play even at lower quality settings.
Ultimately, Dragon Age Inquisition is a good game, but the launch was a bit rough and even now at higher quality settings it can really pound on most systems. Mantle in particular is a mixed bag, with higher performance in many cases but far worse load times -- and if your GPU can't handle your native resolution, at least for now it looks like you'd want to stick with the DX11 path. This is the blessing and curse of a low level API like Mantle: it allows the game developers full access to the hardware, but if the developers don't take care of every last detail (i.e. resolution scaling), things can go bad in a hurry.
CrossFire support works in terms of frame rates, which is something that can't be said for Sniper Elite 3, but whatever is going on with level loads on Mantle makes CrossFire basically useless. Every major area transition results in a massive delay of several minutes while the game loads; it's slower on single GPUs as well, but the delay increases exponentially with CrossFire. Hopefully that gets fixed in a patch, but then we've also seen Mantle break and require periodic fixing (e.g. Thief) with later game/driver updates. If that continues to happen, Mantle will fade away in a hurry. But that might just be for the best once DirectX 12 actually becomes available -- for Windows 10 only, mind you, but then legitimate Windows 7 and Windows 8 users will be able to update for free.

Leave a Reply