Benchmarked: Rise of the Tomb Raider

A scenic overlook

Before we get to the benchmarks, let's take a quick detour off the beaten path to look at the scenery. As discussed in our optimization guide, Rise of the Tomb Raider comes equipped with five presets (plus "Custom") for graphics quality. We'll just bypass the Lowest preset, as frankly it looks pretty awful—though if we could travel back in time, our 2005 selves would likely be impressed. In fact, even for moderate systems, you hopefully won't need to stoop down to the Low preset, which again has a pretty noticeable drop in fidelity. Our focus will primarily be on the top three presets: Medium, High, and Very High. But if you're curious, here's what the five presets look like at 1080p:

Even the Very High preset doesn't actually represent the maximum image quality—you can still enable things like SSAA, along with higher quality shadows and hair. The penalty for going from Very High to Maximum (minus SSAA) looks to be around 25 percent, however, and the minor improvements in image quality generally aren't worth the trouble. In fact, even looking at the Very High vs. High vs. Medium screenshots, you might wonder if the drop in frame rates is worth the slightly better visuals. Bottom line here is that you shouldn't feel bad if you have to start at the Medium setting and start tweaking, as Rise of the Tomb Raider still looks quite nice.

Something else to mention is the game's use of HBAO+ for ambient occlusion, as opposed to the more pedestrian SSAO. Nvidia developed HBAO+, but unlike some previous titles, you can use the setting with both Nvidia and AMD GPUs. The catch is that it's optimized for Nvidia hardware, resulting in a larger hit to frame rates on AMD cards; for this reason, we've elected to test with HBAO+ turned off (using the "On" setting for ambient occlusion), even at the Very High preset. If you want to pixel hunt, there are differences between the two modes, and HBAO+ looks better, but in motion we feel most gamers are unlikely to notice or even appreciate the finer nuances of HBAO+.

Check Your Equipment

So what sort of settings will we test, and what hardware are we using? We've settled on the following five configurations, along with limited testing at the Low preset for a few specific cases that we'll get to later:

• 3840x2160, FXAA, High preset
• 2560x1440, TMAA, Very High preset (without HBAO+)
• 1920x1080, TMAA, Very High preset (without HBAO+)
• 1920x1080, TMAA, High preset
• 1920x1080, TMAA, Medium preset

If you happen to be familiar with the last Tomb Raider (2013), you might think you have a good idea of what to expect. At the highest quality settings, the patterns are pretty similar, but the 2013 release happened to scale very well to lower performance hardware. Sure, it looked pretty awful at the lower quality settings, but a single fast GPU could reach into the hundreds of fps. Rise of the Tomb Raider is not so forgiving, as we'll see in a moment. Fast graphics cards may have to opt for High or even Medium presets (with tweaking), while moderate hardware may struggle even at the Low preset. Ouch. Don't say we didn’t warn you!

For our test platform, we're using one system for all the discrete graphics cards, but we'll check out a few integrated graphics solutions later and update the article. For the time being, here's our standard GPU test system:

We've also included results for two 980 Ti cards in SLI for some of the more demanding settings. On the AMD side, we tried to test a pair of R9 290X GPUs in CrossFire, but things didn't go so well as one of our GPUs has gone belly up. Le sigh. Since that's the only pair of AMD GPUs we currently have for CrossFire testing, we don't have any results right now, but you'll see in a moment that there are other items that AMD needs to address.

A few final items before we depart. First, we're running the latest graphics drivers for both AMD and Nvidia GPUs. However, Rise of the Tomb Raider is an Nvidia title ("The Way It's Meant To Be Played") and Nvidia released a Game Ready driver a couple of days ago. AMD meanwhile reports that they are working on an optimized driver, but for the time being it is not ready—it may come out next week, or perhaps later in the month, and we'll see about testing and updating our findings when that happens. But let's be clear: If you're a gamer eagerly awaiting a new release, Nvidia's approach to drivers is far better; you get to play the game at launch with what should be a reasonably optimized experience. It may not always be perfect (see Batman: Arkham Knight), but more often than not, having a driver tuned for a new game helps a lot.

The second item we want to note is the choice of CPUs. Rather than trying to test multiple systems—which would be ideal if you want to know exactly how a particular configuration performs—we're electing to simulate slower processors using our i7-5930K. The Gigabyte BIOS allows us to disable cores and Hyper-Threading, and while the larger L3 cache is still a factor, at least we can get some idea of how mainstream parts like the i5-4690K and i3-4350 perform. Besides, testing every desirable configuration is a rabbit hole with no end in sight—we would have to look at A10-7850K, A8-7650K, FX-8350, FX-6300, and more to really check out the CPU side of the equation. The good news is that most games are far more dependent on GPU performance rather than CPU/APU performance, so our three test CPUs should at least give a good idea of what to expect.

Feeing Testy?

One final item to discuss before we get to the pretty graphs [Ed: I like pretty graphs!] is the benchmarking procedure. Rise of the Tomb Raider doesn't have a built-in benchmark mode, unlike its predecessor, which means we need to explore alternative means of benchmarking. This is good and bad—good because it's a better real-world look at the game's true performance, but bad because it's far more time consuming and it makes it hard for others to compare results with our numbers. So let's talk about what we're doing for our test sequence.

We use a save at the start of the Soviet Installation level, except we've already played through the level and taken care of all the enemies. This makes the test sequence more consistent, as engaging hostiles in a benchmark inevitably leads to increased variability. We follow the same path each time, as closely as possible; each test run takes about 52 seconds. While we're running the test path, we use FRAPS to log frame rates, which we then analyze to find the average as well as 97 percentile performance.

If you'd like to test your own rig using our benchmark, you can download our save file and put a copy in the appropriate folder (default is C:\Program Files (x86)\Steam\Userdata\[Unique Steam ID]\391220\remote)—don't forget to back up your own save first, if you have one. As for the benchmark run itself, just follow our path shown in the video below. Then feel free to share your results in the comments—and if you want to calculate the 97 percentile, you'll have to do that by opening the CSV file in Excel (or some other spreadsheet program).