Welcome Stranger to OCC!Login | Register

F1 2017 Review

   -   
» Discuss this article (0)

Graphics:

Good. Now, can we move on to performance?

Okay, to be more serious and detailed, the graphics really are quite good. For me this is one of those examples where things look right so nothing stands out. The tracks have good detail, as do the cars, and there are small details to enjoy like blades of grass sticking to the tires, if you accidentally run off the track. When running a wet track, the ground does in fact look wet, and the sound the tires on it make is correct, too. If you look at the screenshots of the wet track I have, you will notice the tires are different from on a dry track, and if I had any images from when I stopped, you would see the tire tread is different, too.

The graphics look good and there are details that might be easily missed but are also easily appreciated when noticed, and I am glad I noticed them.

If I must give a criticism it is the limited anti-aliasing options, because there definitely is a degree of aliasing visible in the game. The only options are to use SMAA, TAA, or a combination of the two. For those of you unfamiliar, SMAA is Sub-pixel Morphological Anti-Aliasing, and is a post-processing AA method. That means it takes the final frame and tries to intelligently blur the edges it contains, as opposed to a method like MSAA, Multi-Sample Anti-Aliasing, that is applied while the frame is rendered, making it much more accurate, but also much more resource intensive. What TAA stands for is Temporal Anti-Aliasing, and it goes after what looks like shimmering as the edges of objects move across the screen.

A more powerful form of anti-aliasing would be nice to see as an option, but I am not upset by the lack of it. I would prioritize performance over visuals here, and that is exactly what SMAA does.

 

 

Now I will move over to talking about performance, which is, to my mind, the more interesting topic of this section. First up, here are my specs:

  • Processor: AMD A10-5800K @ 4.40 GHz (44.0x100)
  • Cooling: Corsair H110
  • Motherboard: ASUS F2A85-M PRO
  • GPU: NVIDIA GTX 1080 8 GB
  • PhysX: NVIDIA GTX 1070 8 GB
  • Memory: G.Skill Ripjaws 4x8 GB (32 GB total) at 1866 MHz 10-10-10-27
  • PSU: OCZ Fata1ty 750 W
  • OS: Windows 10 Pro 64-bit
  • Driver: NVIDIA GeForce Driver 385.28

I ran the game first at max settings, but to combat a little bit of stutter, which I will discuss in a little bit, I turned the options that had been marked at Ultra High down to just High. I somewhat doubt this change would have impacted the stutter, but I forgot to turn the options back up when I was then doing my frame time recordings. I also had motion blur turned all the way down, as a personal preference.

That stuttering I mentioned I believe comes from load on my CPU, because it not only had to handle the game, but also the Tobii Tracker 4C, which I have observed previously being enough to cause performance issues in other games. The impact seemed fairly minor, compared to other games, so I am willing to dismiss it and it is also possible optimizations will remove it. While the effect felt minor, a closer look at the frame time data shows it is not insignificant. (Now for my fun frame time graphs!)

I collected the frame time data in one of two ways: the full 300 second recordings I try to capture when reviewing games now; and in shorter lengths to match simultaneously recorded videos. First I want to cover the 300 second length recordings, of which I made three. Unless specifically noted, the recording was made with the Tobii tracking enabled.

To capture this data I just hit the button to start the recording and kept playing, keeping in mind to not go to the menu or do anything else that would create anomalous data.

 

 

The first recording I made on the Australia track under dry weather conditions. The average frame time for the 300 seconds was 14.57 ms, which translates to an average frame rate of 68.62 FPS. Here is a table of the percentiles:

  0.1% 1% 99% 99.9%
Frame Time (ms) 9.41 9.98 26.68 52.63
Frame Rate (FPS) 106.32 100.19 37.47 19.00

Nice numbers, but here are where some specific frame rates fall:

Frame Rate 60 50 30 20 15
Percentile 19.51 5.21 0.36 0.14 0.07

From this data, we can see that on this track I was only dropping below 60 FPS (16.7 ms) a bit less than 20% of the time. At one-in-five frames, that might seem like a lot, but looking at 50 FPS we see it drop to about 5% or one-in-twenty frames. At least I am comfortable in that 50-60 FPS range, but still, most of the time the framerate was above 60 FPS here.

I have more characteristic values than that though, including one that will be interesting when we look at the other two 300-second runs. The median of this data is 13.747 ms, so half of the time the frame time was longer than that, and half of the time it was less than that. The standard deviation was 7.82 ms, and this is the value to remember when looking at the next two sets of data.

I do not remember my reason for doing this, but the other two 300-second recordings are both from the same China track, under wet weather conditions, which is also where I did some of the video recording.

 

 

 

 

For one recording I got an average frame time of 20.33 ms, while the other got 22.15 ms, so pretty close to each other. Turning it into frame rate we get 49.20 FPS and 45.14 FPS. Looking at the percentiles we get this table:

  0.1% 1% 99% 99.9%
Frame Time (ms) Run 1 14.06 14.72 32.02 50.37
Frame Rate (FPS) Run 1 71.14 67.92 32.23 19.85
Frame Time (ms) Run 2 14.78 15.75 35.33 47.59
Frame Rate (FPS) Run 2 67.64 63.49 28.31 21.01

Now looking at where specific frame rates occur, we get this:

Frame Rate 60 50 30 20 15
Percentile - Run 1 84.48 48.32 0.73 0.11 0.05
Percentile - Run 2 95.42 64.92 1.71 0.09 0.02

Clearly we can see that China – Wet track is more challenging for my computer to run, and this is not surprising to me at all. I wanted to record some wet weather data because I assumed before starting it would be harder on my system than a dry track. However, there is something interesting about this data, as you will soon see.

The medians for these datasets are 19.824 ms and 21.674 ms, with standard deviations of 4.30 ms and 4.40 ms. This is interesting because while the frame times were consistently longer than on the Australia – Dry track, the frame times were also more consistent. I am not sure if this is something the graphs will display very well, especially as some of the axes will scale to match the data (although I have gone through to impose some limits on this), but that is why I am giving the numbers as well.

Now, how about we look at the frame time videos I made? I did my best to record just one lap on the two tracks I made these recordings on, starting at approximately the same time and entering the pause menu to stop recording at the same time as well. These are all from live runs, not the built-in benchmark tool, so there are going to be differences between them. The two tracks were China – Wet and Monaco – Dry, so I have both wet and dry. I also recorded with the game in a borderless window, in fullscreen, and in fullscreen without the Tobii eye tracker. I also made two recordings without OBS Studio, so only collecting frame time data along the same path, both with and without the Tobii eye tracker.

The reason for these different conditions, including without OBS Studio, is because both the Tobii tracker and OBS Studio put extra load on the CPU. This leads to stuttering in game, so I felt it best to also collect this data to compare, and at least I find it interesting. (In case you are wondering, I am using the NVENC ASIC on my GPU for the actual video encoding, but because of how OBS Studio works it must bounce frames off of the CPU before NVENC can encode them. NVIDIA's ShadowPlay/Share utility is able to use another hardware feature so this bouncing is not required, further reducing the performance impact, but I desire the depth of control OBS Studio offers.)

In case you are wondering why I bothered making recordings in a borderless window and in exclusive fullscreen, it is because I can. From what I have observed over the years, there is minimal, if any, performance impact. Also, it makes the overlay of the display time data (the time between frames being sent to the display) relevant, because these data are different between borderless and fullscreen.

 

 

 

 

First up are the China – Wet Borderless and Fullscreen runs, and if we look at the course graphs, we see a similar spread between them, supporting that there is no performance difference between these two display modes. Also, as you can see from the graphs, there is a lot more spread to the data, which is pretty bad and why I collected data without the tracker and/or OBS Studio enabled. The borderless run had an average frame time of 31.23 ms (32.00 FPS) while fullscreen had an average of 30.40 ms (32.88 FPS), which is effectively identical. Percentile table next:

Percentile 0.1% 1% 99% 99.9%
Frame Time (Borderless) 15.37 17.86 60.77 253.11
Frame Rate (Borderless) 65.06 55.98 16.45 3.95
Frame Time (Fullscreen) 14.49 17.34 60.79 323.17
Frame Rate (Fullscreen) 69.03 57.68 16.45 3.09

 

Looking at the percentiles for specific frame rates we get this:

Frame Rate 60 50 30 20 15
Percentile (Borderless) 99.46 95.62 32.68 3.38 0.58
Percentile (Fullscreen) 99.39 91.94 27.61 3.91 0.51

Definitely a performance drop compared to the 300-second recordings, and this is further confirmed with the medians and standard deviations: 29.491 ms and 27.896 ms medians; and 14.479 ms and 16.803 ms for the standard deviations. Pretty bad, but next up are the recordings without either the Tobii tracker enabled or OBS Studio. Obviously, for the latter, I do not have any videos.

 

 

 

 

Since the pattern has been set, I will cut back on the words.

  Frame Time (ms) Frame Rate (FPS)
Average (no Tobii) 19.75 50.61
Average (no OBS Studio) 20.05 49.89

And the percentiles:

  0.1% 1% 99% 99.9%
Frame Time (no Tobii) 13.27 14.65 30.57 168.81
Frame Rate (no Tobii) 75.35 68.24 32.71 5.92
Frame Time (no OBS Studio) 13.77 14.59 33.15 49.90
Frame Rate (no OBS Studio) 72.64 68.54 30.17 20.04

Specific frame rate percentiles:

Frame Rate 60 50 30 20 15
Percentile (no Tobii) 80.15 36.65 0.61 0.13 0.12
Percentile (no OBS Studio) 81.35 44.80 0.94 0.11 0.03

Finally the median and standard deviation:

  Median Standard Deviation
No Tobii (ms) 18.544 12.139
No OBS Studio (ms) 19.51 4.207

As we can see from this data, just removing one of these extra CPU loads significantly improved performance. I am not sure why the standard deviation without the Tobii tracker here is so high, but it looks like some kind of anomaly. Here are the percentiles for the consecutive difference values I have (meaning the difference in frame time between two consecutive frames, which is ideally near zero):

Percentile 0.1% 1% 99% 99.9%
No Tobii (ms) -143.74 -11.58 12.30 147.96
No OBS Studio (ms) -25.71 -12.33 12.33 26.77

That center 98% of data roughly matches between the two runs, so it looks like it is that outer one percent of data that might be throwing off the standard deviation.

I also did a recording with both the Tobii tracker and OBS Studio disabled and again we see performance improvements, which might be more representative of what you can expect (the Tobii Tracker 4C is pretty awesome, but I suspect it is an uncommon peripheral still).

 

 

  Frame Time (ms) Frame Rate (FPS)
Average 14.69 68.04

The percentiles:

  0.1% 1% 99% 99.9%
Frame Time (ms) 10.57 11.72 21.55 40.99
Frame Rate (FPS) 94.61 85.32 46.39 24.40

Specific frame rate percentiles:

Frame Rate 60 50 30 20 15
Percentile 20.37 2.26 0.17 0.05 0.03

Median and standard deviation:

  Median Standard Deviation
Milliseconds 13.777 4.600

Consecutive difference percentiles

  0.1% 1% 99% 99.9%
Milliseconds -14.311 -6.616 6.648 12.746

Even though the standard deviation here matches the No OBS Studio data I have above, the actual spread for the consecutive difference is very different. For the center 98% of the data, the frame time changed by 6.6 ms, in one direction or another, which is half what we saw before. The center 99.98% of the data here roughly matches the center 98% before, so apparently just one of these extra CPU loads is enough to throw off frame time consistency.

Now we can look at the Monaco, dry weather runs and first up are the borderless and fullscreen runs.

 

 

 

 

  Frame Time (ms) Frame Rate (FPS)
Average (Borderless) 27.23 36.70
Average (Fullscreen) 23.68 42.19

The percentiles:

Percentile 0.1% 1% 99% 99.9%
Frame Time (Borderless) 12.39 13.74 60.24 160.52
Frame Rate (Borderless) 80.70 72.79 16.60 6.23
Frame Time (Fullscreen) 11.84 13.10 49.48 70.95
Frame Rate (Fullscreen) 84.45 76.36 20.21 14.09

Specific frame rate percentiles:

Frame Rate 60 50 30 20 15
Percentile (Borderless) 86.78 66.06 25.09 3.74 0.53
Percentile (Fullscreen) 81.30 59.48 11.45 0.91 0.16

Median and standard deviation:

  Median Standard Deviation
Milliseconds (Borderless) 23.617 13.614
Milliseconds (Fullscreen) 21.62 9.763

Not all that dissimilar, which is hardly surprising, but we do see better performance here than in the wet weather conditions from before. (It is my belief the weather conditions are the cause of this difference, and not that track itself.)

Moving on to without either the Tobii tracker or OBS Studio and here's the data:

 

 

 

 

  Frame Time (ms) Frame Rate (FPS)
Average (no Tobii) 16.52 60.51
Average (no OBS Studio) 17.24 58.0

Percentiles:

Percentile 0.1% 1% 99% 99.9%
Frame Time (Borderless) 10.14 11.77 27.23 44.11
Frame Rate (Borderless) 98.63 84.95 36.72 22.67
Frame Time (Fullscreen) 9.79 11.60 30.54 68.18
Frame Rate (Fullscreen) 102.12 86.24 32.74 14.67

Specific frame rate percentiles:

Frame Rate 60 50 30 20 15
Percentile (no Tobii) 39.57 10.49 0.32 0.07 0.04
Percentile (no OBS Studio) 48.79 16.60 0.78 0.24 0.11

Median and standard deviation:

  Median Standard Deviation
No Tobii (ms) 15.956 5.367
No OBS Studio (ms) 16.57 5.305

Fairly similar between them, until we look at the percentiles for specific frame rates, so we are still seeing that reducing that CPU load has an impact.

Finally, the data without neither the Tobii tracker nor OBS Studio going:

 

 

  Frame Time (ms) Frame Rate (FPS)
Average 12.53 79.79

 

The percentiles:

  0.1% 1% 99% 99.9%
Frame Time (ms) 8.58 9.39 19.15 25.92
Frame Rate (FPS) 116.57 106.52 52.22 38.58

Specific frame rate percentiles:

Frame Rate 60 50 30 20 15
Percentile 5.17 0.66 0.04 0.1 0

Median and standard deviation:

  Median Standard Deviation
Milliseconds 11.861 2.265

Much higher and tighter data here, suggesting it is best to not play with either extra loads, at least if you have a four-thread CPU and am playing at the high settings I was.

So what does all of this add up to? The game looks good and still performs well, but not great if you are loading extras onto the CPU, like video recording software. I assume a CPU with more threads will not be so impacted, but since that is not what I have, cannot exactly test that, but I think the guess is reasonable. I will also say right now, even though it is a Gameplay section point, that I would continue playing with the Tobii tracking features enabled. The impact is not that bad and what it adds is awesome, so for me the tradeoff is worth it. (By the way, I do not play games I am reviewing on a "clean" machine. I have applications running in the background, like Firefox and Thunderbird, and there is an anti-virus going, so I have never given you such best-case numbers, but real-life numbers, in case you were curious.)

 




  1. F1 2017 Review - Introduction
  2. F1 2017 Review - Graphics
  3. F1 2017 Review - Gameplay
  4. F1 2017 Review - Additional Gameplay Media
  5. F1 2017 Review - Conclusion
Related Products
Random Pic
© 2001-2017 Overclockers Club ® Privacy Policy
Elapsed: 0.0587930679   (xlweb1)