Graphics and IQ versus Performance

GTA V has excellent graphics. It is perhaps not quite up to the level of realism that Crysis 3 or even Battlefield 4 commands, but it is definitely next generation open world console graphics sufficient to give superb visuals when everything is maxed out. We benchmarked and played the game at 1920×1080, 2560×1600, and at 3840×2160 using 6 of Nvidia’s’ top video cards including TITAN X and GTX 980 SLI, using Core i7-4790K which turbos to 4.4GHz.  We also played using 16GB of Kingston’s 2133MHz Beast HyperX DDR3.

One thing to note is that GTA V uses a lot of video memory and 3GB is the minimum for playing with ultra or even very high settings at 1920×1080, never mind at higher resolutions.  If you have a 2GB vRAM-equipped card, you will be limited to mostly a combination of high and very high settings with modest AA at 1920×1080 even if you have an otherwise very fast card like the dual-GPU GTX 690.  It is not pretty when the game runs out of video memory and has to access system RAM or the hard drive/SSD.  We saw serious artifacting, intense chugging, and late loading pop-in textures when a card’s vRAM limit was exceeded.  Playing at 4K mostly requires 4GB as a minimum and the 6GB and 12GB vRAM equipped TITANs are ideal for this game.

The implementation of FXAA in GTA V is decent and the text is not blurred.  Setting AA to Off offers the highest performance but the jaggies and shimmering from aliasing in motion is really irritating, especially in the city where there are a lot of buildings. 2x or 4xMSAA can be used but the performance hit is high and 4xAA actually does not look as good as 2xTXAA which does a great job of eliminating shimmering at a noticeable cost of blur which the player may decide is cinematic, or not.  MFAA is an improvement over TXAA in terms of much less blur, and the temporal aliasing (texture crawling and shimmering) may also be slightly mitigated by 4xMFAA and to a greater extent by 8xMFAA.

The Performance Hit of using varying AA

First up we look at the GTX TITAN X running at just 1920×1080, but with every single setting at Ultra or the Very Highest chosen, with the exception of “2xMSAA Reflections” and Motion Blur.  Using 8xMSAA is barely playable and the improvement in IQ is less than 8xMFAA which is 10fps faster.GTA5-TitanX-Chart

Here is the main chart comparing our top six Nvidia GeForce cards using ultra settings and only varying the AA:GTA5-Main-ChartAs you can see a 3GB vRAM-equipped GTX 780/Ti needs to play at 1920×1080 with up to 4xAA, and the performance results will be OK with ultra settings.  A 3GB GTX 780 or GTX 780 Ti can manage FXAA at 2560×1600, but adding MSAA impairs fluid performance that a 6GB TITAN can handle – but don’t even think about 4K with a 3GB card.  A GTX 980 can handle 4K with mostly ultra settings with FXAA, but adding MSAA tanks performance.  It takes a TITAN X or GTX 980 SLI to handle 4K with ultra settings with FXAA on; but adding 4xMSAA to GTX 980 SLI will cause chugging and swapping to disk that the TITAN X doesn’t have issues with, although a single TITAN X’ performance is too low for fluidity.

IQ comparisons of Anti-aliasing at 1920×1080 with maxed out settings

GTA V is a real challenge to benchmark while at the same time to show near-identical IQ settings.  Fortunately, we found a very nice workaround which lets us compare the IQ in motion as well as benchmark the game very accurately using the last (long) benchmark.

We will show Fraps captured screenshots from the game.  Unfortunately, due to the dynamic nature of the game, they are not identical although they illustrate the differences in AA methods.  And since MFAA cannot be imaged perfectly in a Fraps screenshot (the MFAA screens posted here use Fraps which miss the last bit of temporal filtering, so they will look slightly better on the actual display).

Fortunately ShadowPlay can capture the differences between the AA settings accurately, including MFAA and even DSR.  Here is a YouTube video which illustrates the varying AA levels in GTA V, including FXAA, MSAA, TXAA, and MFAA.  When we look at our performance, we see that FXAA is almost “free” performance-wise and we enable it for all of our IQ testing except for “No FXAA” as the blur is quite mild yet it can effectively combine with other forms of AA including with MFAA.

Please note the Fraps frame rate counter in the upper left corner of the video as different levels of AA are applied.  You can see that FXAA takes a small hit over no AA, and that MSAA takes a proportionally larger hit than anything else.  4xMFAA takes about the same performance hit as 2xMSAA but it looks better, much like 4xMSAA; and 8xMFAA takes a smaller performance hit than 4xMSAA, but it looks arguably better.

All below Fraps captured screenshots are at 1920×1080 with completely maxed-out Ultra settings including PCS Shadows and Ultra Grass.  We even enable 2xMSAA “Reflections”.  It might be a crime for a hardware enthusiast to see aliased reflections of buildings in your Mammoth’s polished exterior as you flee from police.


Here are two similar shots – one with MSAA reflections off, and the other is set to 8X;

Mammoth1 Mammoth2Make sure to open each screenshot up in a separate window to compare them to each other at full size. First up, with no anti-aliasing, it is pretty easy to pick out the jaggies. Focus in on the fence and also on the mailbox door.








 * MFAA cannot be captured perfectly by a Fraps screenshot as it misses the last filter which is applied after the capture.  Please see the comparison video for a ShadowPlay comparison at 1920×1080 which will capture it as it goes to the display output.

From viewing the screenshots, TXAA looks the best as far as eliminating jaggies (and aliasing in motion), but at a cost of significantly more blur than FXAA.   However, with the camera in motion, it appears that 2xTXAA is sufficient to provide a similar amount of anti-aliasing to 4xMSAA but with less of a performance hit and with much less crawling and shimmering.

FXAA is a solid improvement over no AA.  2xMSAA takes a big performance hit but does little for the jaggies.  Using 2xMSAA plus MFAA (4xMFAA) gives the performance hit of 2xMSAA, but improves the jaggies almost much as 4xMSAA does.  And using MFAA with 4xMSAA (8xMFAA) gives a very nice anti-aliasing effect almost equal to TXAA, but without much of the blur!  And playing with 8xMSAA loses fluidity.

Let’s wrap it up and head for our our conclusion


Comments are closed.