My RealFlight 7.5 Anti-aliasing Benchmarks

In another thread we've been talking about the frame rate hit you take when you turn on Anti-aliasing in RealFlight 3D fields, so I got curious and ran some benchmarks on my system. I used the recently released Zenjima (Island of Enlightenment) 3D field and ran 7 benchmarks in RealFlight 7.5. I uploaded a video of the 16xCSAA run with a chart at the end showing all of the CSAA results. The chart is the last 10 seconds of the video but you can pause the video to see it better. I created a video because I think it's best to see exactly what is displayed on the screen when the benchmark is taken. I'll put a link to the video at the bottom and also put the Avg, Min, Max results for all of the runs. The video is 1 minute and 16 seconds so it doesn't take much time if you're interested in seeing the chart. I only put the CSAA results in the chart because to me it gives about the same frame rate as a couple of steps down using straight AA but looks better. So I left the straight AA results out of the chart to make it easier to see the differences.

First my system specs for reference. (Nothing is Overclocked)

CPU: Intel I7 4770S Low Power 65 Watt (Clock 3.1 GHz - Max Turbo: 3.9 GHz)
MEMORY: 32 Gig Kingston Value Ram
GRAPHICS: EVGA GeForce GTX 760 (Reference Board Clocks)

CSAA Results

No AA - Avg: 98 Min: 65 Max: 158
8xCSAA - Avg: 95 Min: 62 Max: 142
16xCSAA - Avg: 91 Min: 61 Max: 137
32xCSAA - Avg: 87 Min: 55 Max: 127

Straight AA Results

No AA - Avg: 98 Min: 65 Max: 158
2xAA - Avg: 95 Min: 63 Max: 145
4xAA - Avg: 94 Min: 61 Max: 141
8xAA - Avg: 90 Min: 57 Max: 134

Link to Benchmark Video - The Chart shows the last 10 seconds of video.

I uploaded this video in 1080p but YouTube won't show it at that resolution unless you tell it to. Use the small 'gear' icon on the bottom right side of the video window to change the resolution. Click on it to get a small menu, then click on the drop down 'quality' entry and choose 1080p HD. It looks really good in 1080p, but YouTube's default resolution doesn't look very good. (At least on Desktop Monitors - Haven't seen it on a Cell Phone)

https://www.youtube.com/watch?v=lPxsmm5D4ig

I thought I better include how I did the benchmark, kind of an important thing. The first thing I did was create a recording of the flight using RealFlights recording feature. Then for each benchmark run I held my hand over the F3 key so I could change to the Chase View the moment the playback started. Then I hit the FRAPS benchmark key (F11) to start taking the benchmark at the moment the throttle was increased and stopped the benchmark the moment I touched down. So every benchmark was taken exactly the same. I exited RealFlight between each benchmark and adjusted the Anti-aliasing in the NVidia Control Panel. The control panel was set to override any application setting so the Anti-aliasing mode would always be correct. I used FRAPS 3.5.99 Build 15618 and RealFlight 7.50.011.

And Now The Rest Of The Story

When I created the first set of 3D Benchmarks Anisotropic Filtering was set to Application, and since I didn't see anything in RealFlight that set Anisotropic Filtering I assume there was NO Anisotropic Filtering applied in my first set of benchmarks. Since I'm a long time 3D Gamer I wanted to do some benchmarks with Anisotropic Filtering applied so I could see how much it changed the frame rate. Like most 3D Gamers to me frame rate is king so I always tailor my graphics settings for the best overall performance, not set everything to max and forget it. Of course on a lot of games these days you can do that with a fast CPU and graphics card, and mine is pretty fast. Not the fastest, but far from the slowest. Of course I've had it a while so it is showing it's age, and to be honest I'm looking hard at the new NVIDIA 970 or 980. I've always balanced price, performance, and the power the graphics board needed but with the 980 being so energy efficient it's just the price that's hard to swallow. But back to the point. I created another set of benchmarks to share with you.

I used the same benchmark recording as the first set to create new benchmarks that show what adding Anisotropic Filtering to the mix does. So here's some benchmarks with NO AA and NO AF, then adding just AF, then adding just AA, and last adding both AA and AF.

Zenjima 1 Minute Benchmark Recording

NoAA - NoAF - AVG: 97.87 - MIN: 65.00 - MAX: 144.33

NoAA - 16xAF - AVG: 90.60 - MIN: 61.33 - MAX: 133.00
16xCSAA - NoAF - AVG: 91.75 - MIN: 59.33 - MAX: 134.33

16xCSAA - 16xAF - AVG: 83.82 - MIN: 55.00 - MAX: 121.66

I separated the only AA and only AF benchmark numbers so you can see that either of them 'alone' takes about the same frame rate hit. And then the bottom set of numbers shows the frame rate with both AA and AF applied. Which still gives good numbers on my system. Of course I'd like the MIN number to be 60 or above but when I created a graph of the benchmark the area's below 60 fps were very small. So to me, on my system, running both together is OK. I will say that in 3D Games I could see a lot more difference applying 16x AF than in RealFlight. Of course I've only ran it on one 3D field and in other fields I've tried it may make more difference.

Of course after seeing those numbers I wondered what I'd get if I used 8xCSAA and 8xAF so I ran the benchmark again to get those numbers. I'll put both of them below for you to check out.

Zenjima 1 Minute Benchmark Recording

8xCSAA - 8xAF - AVG: 86.34 - MIN: 57.33 - MAX: 126.66
16xCSAA - 16xAF - AVG: 83.82 - MIN: 55.00 - MAX: 121.66

I'm glad I added 8xCSAA and 8xAF because there's not that much difference between them, but there is a noticeable difference when you watch the plane in chase view. So far I always fly in chase view on the 3D fields. I tried first person view in the cockpit but for me watching the plane fly through the fields is more enjoyable.

Okay, one last thing before I go. The truth is benchmarks are nice because they give us a way to measure our system but I wondered what I'd get if I just flew the field for enjoyment, like I normally do. So I did a second benchmark recording and flew the field like normal. The recording was a very fun fly, just what I generally do when flying Zenjima. So using the long recording gives me what I really get flying the field. I'll put the numbers below and wrap this up.

Zenjima long benchmark recording (7.6 Minute Flight).

NoAA - NoAF - AVG: 102.850 - MIN: 65 - MAX: 168
8xCSAA - 8xAF - AVG: 91.147 - MIN: 58 - MAX: 153
16xCSAA - 16xAF - AVG: 88.175 - MIN: 56 - MAX: 155

Of course the long flight is the best numbers because they show what I'm getting during a real flight, and they also echo what the benchmark shows. Of course that's what we want, but sometimes that's not the way it works. For those of you that looked close you'll be saying that the MAX number for the 16x flight is higher than the MAX number for the 8x flight, what's the deal? Well the deal is, that's what was in the FRAPS log file so I left it alone. I will say the first set of benchmark numbers at the top of this post is an average of 3 runs and I seen that the numbers change from run to run a little, and because of that I left the bottom set of numbers (the long benchmark) alone. I was too lazy to take three sets of benchmarks for the long recording because it's 7.6 minutes for each run. So the final long benchmark numbers are only for one run, not three averaged.

Now You Know The Rest Of The Story. (Yes I know - Paul Harvey)

It is interesting how far along hardware AA has come.

In your benchmarks even high levels of AA have little effect when using the newer cards.

Median and lower end cards fair well too. The biggest impact on FPS on those cards are depth of field, particles and VRAM/Texture space availability, if the CPU can keep up with the card's demands.

e.g. on an Nvidia GeForce 750 or 750ti that are pretty good cards for the money.

Adding Bloom and Depth of Field Benchmarks

After reading opjose's post about how much bloom and depth of field would affect the frame rate I thought I'd add one more set of benchmarks to see how much they changed the frame rate on my system. And since I'm considering upgrading my graphics card adding bloom and depth of field seemed like a good idea since I knew they would definitely hit the frame rate pretty hard, so adding them would give me a better picture of my current performance in RealFlight. Then I would be able to compare everything again if I upgrade my graphics board.

These benchmarks are not compatible with my previous benchmarks for two reasons. First I lost the Short Benchmark recording I used for the previous benchmarks and had to re-create it, and second I quit using one of the settings in the NVIDIA drivers I used before. The setting is called FXAA and it turns On and Off a shader based technique to improve aliasing, but it's not hardware AA just a fast shader technique. And since it lowers the average frame rate about 10 fps and I can't really see much difference so I stopped using it.

Like before every time I made a change to either the RealFlight settings or the NVIDIA driver settings I exited RealFlight and restarted it before each benchmark run. I also did 2 consecutive benchmark runs for each setting and listed both of them. All benchmarks started the moment I increased the throttle and stopped the moment I touched down, so they should be very close for each run. I also listed every setting in the benchmark results so I could make sure I had the same settings later if I get a new graphics card. So enough talk I'll paste the contents of the text file I created with the benchmark results below.

----

Hardware Used For Benchmark

CPU: Intel I7-4770S (Low Power 65 Watt Version 3.1 Ghz - Max Boost 3.9 Ghz)
RAM: 32 GB Kingston Value Ram
Graphics Board: EVGA 02G-P4-2763-KR GeForce GTX 760 (Reference Board Clocks)
Nothing Overclocked

NVIDIA Graphics Driver: 344.48
Display Resolution: 1920x1200
FRAPS: 3.5.99 Build 15618
RealFlight: 7.50.011 Public Release

NOTE:
FXAA set to OFF in NVIDIA Driver Program Settings Dialog Box for realflight.exe
FXAA is a fast shader based post processing technique, not hardware AA, but it
lowers the average frame rate about 10 fps and I see very little improvement.

RealFlight Default Graphics Settings

Clouds: Yes
Particles: Yes
Scenery Objects: Yes
Shadows Enabled: Yes
Soft Shadows Enabled: No
Terrain Detail: Yes
Trees: Yes
Wind Field Indicators: Yes
Post-processing: Bloom: No
Post-processing: Night Flying Glow: Yes
Post-processing: Depth of Field: None
Foliage Density (%): 5
Normal Maps: No
Water Quality: Medium
Water Particle Quality: Medium
Water Shadow Quality: Off
Shadow Quality: Low
Shadow Map Quality: Low
Particle Quality: Highest
Screenshot Quality: Low
Streamer Quality: Highest
Texture Quality: Highest
Misc. Graphics Quality: Highest

Zenjima Short Benchmark V2 68 Second Flight Time (2 Consecutive Benchmark Runs)

RealFlight Default Graphics Settings (No AA - NoAF)
Frames: 12601 - Time: 67674ms - Avg: 186.201 - Min: 118 - Max: 288
Frames: 12675 - Time: 67751ms - Avg: 187.082 - Min: 118 - Max: 288

Add Normal Maps: Yes - Water Quality: Highest - Water Particle Quality: High - Shadow Quality: Highest - Shadow Map Quality: Highest (No AA - NoAF)
Frames: 7144 - Time: 67610ms - Avg: 105.665 - Min: 70 - Max: 181
Frames: 7208 - Time: 67704ms - Avg: 106.463 - Min: 70 - Max: 182

Add 16xCSAA - 16xAF
Frames: 6642 - Time: 67548ms - Avg: 98.330 - Min: 67 - Max: 161
Frames: 6665 - Time: 67689ms - Avg: 98.465 - Min: 67 - Max: 161

Add Post Processing Bloom: Yes - Post Processing Depth of Field: Medium
Frames: 5241 - Time: 67720ms - Avg: 77.392 - Min: 55 - Max: 116
Frames: 5225 - Time: 67798ms - Avg: 77.067 - Min: 54 - Max: 116

Add Post Processing Bloom: Yes - Post Processing Depth of Field: Highest
Frames: 4318 - Time: 67689ms - Avg: 63.792 - Min: 47 - Max: 86
Frames: 4191 - Time: 67673ms - Avg: 61.930 - Min: 45 - Max: 82

Pretty big hits!

I ran some benchmarks on Shadow of Mordor using a 660ti w/ 2GB of VRAM on an iCore7 w/16gb of RAM & Win7 x64. A typical median one or two year old or so computer....

That game uses depth of field extensively and even with everything cranked up with far more details than that displayed in Realflight, I'm well over 90 FPS on a worst case everywhere.

That leads me to believe the DOF hits in Realflight are not due to the program engine but the API instead.

DX9 should have been put to sleep a few years ago.

Since I only did a benchmark for two of the four Depth of Field settings my curiosity got the best of me so I ran a full set to see what the difference was for each step. I'm glad I did because it was interesting to see that the first three settings took almost the same frame rate hit, with only the Highest setting having a significant change. I'll put all four benchmarks below.

These were done just like my first set with all the modifications to the default graphics settings and 16xCSAA and 16xAF applied. I restarted RealFlight after every change to make sure the setting was correct and did 2 consecutive runs. I put all changes to the default settings below and the starting point numbers with all the settings applied. This is a new set of benchmarks so the numbers are slightly different from the first set.

16xCSAA
16xAF
Normal Maps: Yes
Water Quality: Highest
Water Particle Quality: High
Shadow Quality: Highest
Shadow Map Quality: Highest

Starting Point Of Benchmark with All Changes except Depth of Field
Frames: 6622 - Time: 67704ms - Avg: 97.808 - Min: 66 - Max: 162
Frames: 6636 - Time: 67704ms - Avg: 98.015 - Min: 66 - Max: 162

Here's the full set of benchmarks for Depth of Field from FRAPS log.

Post Processing Bloom: Yes - Post Processing Depth of Field: Low
Frames: 5251 - Time: 67626ms - Avg: 77.648 - Min: 55 - Max: 117
Frames: 5201 - Time: 67548ms - Avg: 76.997 - Min: 55 - Max: 113

Post Processing Bloom: Yes - Post Processing Depth of Field: Medium
Frames: 5243 - Time: 67767ms - Avg: 77.368 - Min: 54 - Max: 119
Frames: 5171 - Time: 67642ms - Avg: 76.447 - Min: 53 - Max: 116

Post Processing Bloom: Yes - Post Processing Depth of Field: High
Frames: 5237 - Time: 67548ms - Avg: 77.530 - Min: 55 - Max: 113
Frames: 5146 - Time: 67595ms - Avg: 76.130 - Min: 53 - Max: 111

Post Processing Bloom: Yes - Post Processing Depth of Field: Highest
Frames: 4314 - Time: 68079ms - Avg: 63.368 - Min: 46 - Max: 87
Frames: 4133 - Time: 67595ms - Avg: 61.144 - Min: 43 - Max: 81

I doubt that anyone would use the highest setting because it's definitely on the blurry side. But I will say that if the program would automatically step from low, to mid, to high as a fast plane accelerated it might be a nice effect. But as a static effect it's not something I'd want.