Gears of War 4 Benchmark

[moonbogg]

This game needed no testing from me. It was perfect from day 1. Awesome game. Carfax told me that it looked arguably better than DOOM and I called blasphemy, but now I'm not so sure. Gears looks stunning and runs amazing. DOOM and Gears both did us PC gamers right. Both are very rare kinds of games these days. The story was great too and I felt like I was participating in an action movie. I really thought it was great and a unique game.
The only thing they got wrong was the ending. What was wrong with the ending you ask? The game ended! That's what, and screw them for that. It should have kept going for like 50 more hours and I'd still be playing it. I'm buying the next one and I hope there will be a next one.

 

[Carfax83]

Both Doom and Gears 4 represent the pinnacle of PC gaming in many ways. Both games look fantastic, and more importantly, run like Usain Bolt with rocket boosters attached to his backside. The reason I give Gears 4 the edge though, is because there is a lot more going on in every scene...

Doom has superior particle effects, but Gears has a wider range of effects, including some very advanced ones like real time single bounce global illumination. Gears also makes handy use of weather and destruction effects which are physically driven..

 

[Red Hawk]

So picked up the game yesterday, played it for a bit, looks fun...but actually playing the game will have to wait because the tech geek part of me demands BENCHMARKS! And good gravy, what a benchmark tool they have. Much more useful, and fun, than the flat min/max/avg provided by benchmark tools in games like Deus Ex. I appreciate when games provide internal metrics for measuring framerates, frametimes, memory usage, etc., because I feel like tools like MSI Afterburner are prone to inaccuracy after seeing in-game metrics contradict those tools.

So a simple first run, using the system in my signature, running on the ultra preset (which maxes everything out except for depth of field and reflections) with no vsync. First number is frame rate per second, second number is frametime in milliseconds.

Avg: 73.0 (13.7)
Avg GPU: 73.7 (13.6)
Avg CPU: 193.0 (5.2)
Avg CPU Render: 158 (6.3)
Avg Min: 54.3
GPU Bound: 96.9 %

So, starting out...yeah, pretty well optimized for my purposes, since I'm playing on a plain old 60 Hz monitor. Definitely GPU bound as opposed to CPU bound -- lots of CPU headroom, it seems like.

So what happens when I turn on the Insane settings for depth of field and reflections? I hear they're very performance heavy. Results running the benchmark for that:

Avg: 52.3 (19.1)
Avg GPU: 52.4 (19.1)
Avg CPU: 196.0 (5.1)
Avg CPU Render: 170.8 (5.9)
Avg Min: 41.3 (24.2)
GPU Bound: 100%

Yep, definitely performance heavy, bringing it significantly below 60 FPS. I'm fine playing with those settings turned off, thanks.

I was curious what difference asynchronous compute makes, so I ran the benchmark on ultra with asynchronous compute disabled...

Avg: 65.9 (15.2)
Avg GPU: 66.0 (15.1)
Avg CPU: 187.4 (5.3)
Avg CPU Render: 159.2 (6.3)
Avg Min: 47.9 (20.9)
GPU Bound: 98.3%

That's a good 10% performance increase thanks to asynchronous compute...not bad, not bad at all.

That amount of headroom on CPU performance makes me wonder just how the game performs on a slow CPU...will be back with benchmarks from the other PC I have access to, stock Core 2 Quad Q6600 @ 2.4 GHz, Radeon 270X 2 GB @ 1080 MHz, 8 GB DDR2 RAM @ 800 MHz.

 

[Red Hawk]

So, ran some benchmarks on that Q6600 PC. Got some interesting results, particularly when it comes to asynchronous compute. I normally test this PC at the resolution of the monitor it's hooked up to, which is 1440x900. The odd thing with GOW4 is that it couldn't be set to that resolution. The resolution slider would not stop on 100% of the monitor's native resolution -- it would jump between 98% and 102%, but never land on 100%. It's not that the game refuses to run at a 16:10 aspect ratio -- with VSR enabled I could bump the resolution up to 1920x1200 no problem. In any case, I settled on 98%, which is 1408x880. It's worth noting that it's been reported that the game seems to automatically decrease level of detail on cards with low amounts of VRAM, rendering results on higher settings somewhat suspect. But in any case, here are the results from the in-game benchmark:

Results: Framerate (Frametime)

Ultra Preset:
Avg Framerate: 38.1 (26.2)
Avg GPU: 44.1 (22.7)
Avg CPU: 63.9 (15.7)
Avg CPU Render: 49.3 (20.3)
Avg Min: 26.4 (37.9)
GPU Bound: 74.8%

High Preset:
Avg Framerate: 37.5 (26.6)
Avg GPU: 47.3 (21.2)
Avg CPU: 64.5 (15.5)
Avg CPU Render: 47.3 (21.1)
Avg Min: 25.3 (39.5)
GPU Bound: 51.6%

Medium Preset:
Avg Framerate: 42.8 (23.4)
Avg GPU: 57.3 (17.4)
Avg CPU: 69.6 (14.4)
Avg CPU Render: 54.7 (18.3)
Avg Min: 32.5 (30.7)
GPU Bound: 43.2%

At first, it seems like there's not much difference between the Ultra preset and the High preset. Some numbers come in a bit higher on the Ultra preset, even. Things don't really seem to improve until you go down to Medium. I tested actual gameplay using the Ultra and High presets though, and there's definitely a difference. Things are smoother on the High preset, for sure. Note the difference in the GPU bound percentage between the two -- even though the other numbers are similar, Ultra hits the graphics card harder. High still has problems, though. It can swing all the way from 60 to down into the teens sometimes during intense moments. You can make it more consistent by enabling a 30 FPS cap, locking it down to console level performance. I'd definitely say it's playable, but maybe not as good of an experience as playing on an Xbox One.

What of asynchronous compute, then? I ran the benchmark on the High preset with async disabled, and...

Avg Framerate: 39.7 (25.2)
Avg GPU: 48.8 (20.5)
Avg CPU: 69.6 (14.4)
Avg CPU Render: 50.9 (19.6)
Avg Min: 27.0 (37.1)
GPU Bound: 61.1%

Like with 3DMark Timespy, there doesn't seem to be a benefit from asynchronous compute on this system. In fact, turning off asynchronous compute resulted in some gains to framerates/frametimes. I am a bit skeptical given how Ultra had some better numbers than High despite feeling worse in practice, but whatever the case, there's no significant gain when using asynchronous compute. I would say it might be because it's just too CPU bottlenecked to make a difference, but the individual GPU frametime is still better without async compute than with. It doesn't make a whole lot of sense -- I thought asynchronous compute helps on consoles, where both the Xbox One and PS4 have significantly weaker GPUs than the 270X? Xbox One coming in behind the 260X in raw performance potential. Maybe asynchronous compute can only be tuned that much on a console? I don't know.

 

[PigSkinWytBoy]

[redacted]

rasism comments are not allowed
Markfw
anandtech moderator.