After yesterday’s settings tests I was excited to try the sim in VR today. I loaded up at Palm Springs, which with my “daytime” settings yesterday was giving me about 50 FPS on the ramp and about 30-35 FPS at takeoff in the non-VR home cockpit.
Today in VR I was getting 70 (!!) FPS in the Oculus Rift with those same settings. I increased the objects to max, and was still getting 50-60 FPS. I increased the super sampling to 1.3, and was still getting high frames (40-50). I ended up with visual effects medium, max textures, 8xAA, shadows off, max objects, reflections low, parked aircraft on, 1.3 super sampling, and had no problem holding 40 FPS with spacewarp enabled. And this is with the vFlyte SR20, which also takes a few frames away vs. other aircraft.
Unbelievable. Total game-changer for VR in Vulkan, at least for me.
Yesterday I committed the morning to testing graphics settings in X-Plane 11.5beta6 with three objectives: (1) understanding the performance difference Vulkan gives me over OpenGL, (2) understanding the performance implications of the various graphics settings under Vulkan on my setup, and (3) finding my optimal Vulkan graphics settings under the current beta.
As a reminder, my setup is an Intel i7-6700K 4.0 overclocked to about 4.6, liquid cooled, with 64G RAM and an EVGA GTX 1080ti Black Edition GPU with 11G VRAM. The GPU is running three screens in X-Plane, and four overall, with the motherboard GPU running a fifth screen. Full details are here. All testing was done sitting on the ramp at KVNY in the Jason Chandler SR20 I use in the sim, with Ortho scenery and no plugins. My testing setup was to compare settings in clear skies and in scattered clouds so I could understand their impact. I had a number of crashes along the way — it’s a beta, so this isn’t unexpected.
Overall it was a very interesting morning. Here’s what I learned.
First, regarding OpenGL vs Vulkan, my prior Vulkan settings (HDR, 4xSSAA+FXAA, shadows on, high objects, reflections on first notch, draw aircraft) were producing about 25 FPS on the ramp at KVNY. These same settings in OpenGL produced 13 FPS, and then quickly crashed the sim. So call it a nearly 100% increase in performance, and a big increase in stability. That’s all the OpenGL testing I wanted or needed to do.
Moving on to the testing of Vulkan settings, I reloaded the sim with all graphics settings full-left to establish a baseline performance level. Full-left settings gave me 89 FPS clear skies, 70 FPS scattered clouds. This 20-frame hit from clouds under full-left settings was the largest cloud hit I had in testing, interestingly. With the baseline established, I began adjusting one setting at a time to see its impact on frames. Here’s what I found:
Visual Effects: Full left 89 FPS, Med 88 FPS, HDR 80 FPS, Max 80 FPS. (I forgot to test clouds here, but more on that later). Conclusion here is that medium is as good as full-left, and moving to HDR costs about 10 FPS. One would think that max right has no impact from this, but with clouds and objects in a later test we will see this isn’t the case.
Texture Quality: Full left 88 FPS, Max 87 FPS. At max it loaded about 4,100 MB of data, and with my GPU it had no real effect on performance. Others with less VRAM will have other results. Interestingly, when switching other settings (like anti-aliasing or visual effects) the sim would reload textures. There would be a brief frame hit while it did so, but once they were loaded frames would return to prior levels. Conclusion (for my card): Max out the texture quality.
Antialiasing w/o HDR: Full left 88 FPS clear, 60 FPS scattered. 4x 88 FPS clear, 60 scattered, 8x 80 FPS clear, 55 scattered. Conclusion: 8x costs about 10 frames, and while one would expect a 20-30 FPS hit for clouds on 4x or 8x, later with objects loaded we find this is not the case.
Antialiasing w/HDR: I only tested 4xAA and up here, as I really don’t like the visuals without it. 4xSSAA 78 clear, 55 scattered. 4xSSAA+FXAA 75 clear, 55 scattered. 8xSSAA+FXAA 53 clear, 50 scattered. Conclusion: with HDR 8x costs about 20 frames, but does not hurt cloud performance.
Antialiasing w/Max Visual Effects: 1xAA 80 clear, 65 scattered. FXAA 80 clear, 65 scattered. 2xSSAA+FXAA 73 clear, 65 scattered. 4xSSAA 42 clear, 34 scattered. Given that large drop, skipped 4xSSAA+FXAA and went to 8xSSAA+FXAA 16 clear, 14 scattered. Conclusion: Max visual effects has a huge hit at 4x AA and up. Avoid as I can’t run 4x AA at reasonable frames (remember I have no objects loaded or shadows in these tests).
Those are the settings that are GPU-limited. I then turned to the CPU-limited settings (and my system is definitely CPU-limited, given the age of the i7-6700K):
Shadows: Visual effects and AA full left, no objects, 80 FPS shadows off, 70 FPS shadows on. Max objects: 30 FPS shadows off, 15 FPS shadows on. Conclusion: Shadows have a minimum 10 FPS hit in Vulkan, at least for me.
Objects: Full left 80 FPS. Low, 60 FPS. Med 45 FPS. High 35 FPS. Max 30 FPS. Conclusion: Deciding to have any objects at all gives the biggest frame hit for me, at 20-25 FPS. After that each increase in objects appears to hit my CPU for a lower number of FPS (15, then 10, then only five from High to Max).
Reflections: On my rig I only get an FPS hit when reflections are set to the third setting or higher, and that hit is significant (10 or more frames). Conclusion: Might as well use them on the second notch, but never more.
Parked Aircraft: Did not test as I know they have a negligible effect on my system.
With this testing done I went into final trials of what could be optimized settings for my rig, at least in this Vulkan beta.
Option 1 high objects no shadows: Visual effects medium, textures max, AA 8x, shadows off, objects high, reflections low, parked aircraft: 47 FPS clear, 44 FPS scattered. (Note that with these settings clouds have a minimal hit on frames.)
Option 2 high objects with shadows: Visual effects medium, textures max, AA 8x, shadows on, objects high, reflections low, parked aircraft: 39 FPS clear, 27 FPS scattered. Note the big FPS hit from shadows, and that they made the clouds impact higher as well.
Given that going from high to max objects only costs me about five to 10 frames with these settings, if I’m not in a highly-populated area I may load with max objects, but these will be the defaults. From a plug-in standpoint, the only graphic plugin I will run on top of these is the FlyAgi Tweak Utility, which is my preferred visual-modification and frame-limiting plugin. I use it with the level of detail knocked down three clicks, which sets the LOD to about 82% (and which is where I like it).
So how do these settings work? To test them yesterday afternoon I did a PilotEdge flight from Palm Springs to San Diego using ortho scenery and the “Daytime” settings above. Visually it was the best flight I have had in four years with the home cockpit. I have a video of the flight below, and I also believe that visually it is the best video I’ve streamed to date (the 1080p60fps stream works very well, and the Vulkan performance is nearly devoid of stutters). Beginning on the ramp at Palm Springs I had 50 FPS. They dropped to about 35-40 during takeoff, climbed back up to 50 en-route, and were at 30 coming over San Diego and at landing. Weather was clear but I know from the testing that clouds would have cost me only about five frames. (Finally, if you watch the video, note that the audio was off for the first 4:20.)
Visual performance in flight simulators is very rig-specific. Your testing may vary widely from what I learned and have posted above. But it also might be helpful to you, even if it you simply follow a similar process. Regardless, it’s clear that Vulkan is going to be a game-changer for X-Plane.
I’ve been running beta 3 of X-Plane 11, deciding to wait on updating based on reports about betas 4 and 5. Beta 6 came out last night, and based on initial reports I decided to go ahead with the update. Initial impressions are that things only get better under Vulkan. Performance that yesterday was 27-30 FPS in the home cockpit with HDR, 4xAA+FXAA, and one below max objects today was 40 FPS under the same conditions, or 30 FPS with max objects. Adding clouds changes little to nothing in terms of performance, at least in my rig. I seem to see just a few micro-stutters compared to beta 3, but the performance is certainly better and I know the stutters (which are very minor) will disappear with later optimizations. I find this whole beta test hugely encouraging, and as I have said before, X-Plane 11.5 looks to me like it will be transformative in terms of performance.
One other benefit, at least to me, is that with reduced load on the CPU everything is working better: all my add-ons, the Saitek FIPs etc. I have far fewer performance problems on the rest of the gear in the sim now that X-Plane is more efficient. Overall it’s produced a far more solid and trouble-free experience. I have to say — knock on wood — that the simulator is performing extremely well and it’s been a pleasure to use.
I took a pass on beta two after reading the initial reports, but today held my breath and updated to beta three hoping for the best. I had downgraded my NVIDIA drivers to one that folks had reported as having better performance, and that caused me problems, but as soon as I updated to the most recent drivers everything worked great and was smooth as silk. Notably, upping the AA (HDR off) to 4x and even 8x isn’t having a significant hit on performance on my system, even sitting in SoCal, with ortho and full overcast (using VMI twick as a frame limiter, but with it set to 83% objects and 63% clouds). This gave me 24-25 FPS, running streaming software and a bunch of other things on the PC. Better yet, not a single stutter than I could notice. It’s very, very smooth.
Vulkan makes it a whole new day for X-Plane, even in beta. It’s a great, great improvement in the performance of the sim.
Today I had a window in my schedule and it was a prime time to do some real-world-quality practice in the sim. I haven’t been in the actual Cirrus SR20 I rent for two weeks thanks to Covid-19, so it’s essential that I use the sim to keep the aviation brain fresh, especially for checklists and IFR procedures.
Fortunately, VR in the vFlyteAir SR 20 is an excellent way to do this, given its very realistic flight and operational model. So today I decided to log on to PilotEdge and simulate a flight under real-world weather conditions from Seattle to Spokane (KSEA to KGEG), filing ZADON ZELAK ZIRAN ZOOMR.ZOOMR1 GANGS. PilotEdge ATC gave me the Seattle 7 departure, so I was able to practice a SID as well, and when arriving in Spokane I shot the RNAV (GPS) Y RNW 8 approach. The vFlyteAir simulation of the Cirrus Garmin Perspective system handles this sort of IFR work reasonably well, with some modifications and concessions compared to the real-world system.
This flight was also over Orbx Washington TrueEarth HD scenery, which is always taxing on my CPU-limited system. For VR settings I selected HDR, FXAA, one notch below max objects, reflections at 1, shadows off. I used VMI-twick to regulate frame rates. To sharpen the image in VR, I used supersampling of 1.6. Weather was multiple levels of scattered and broken clouds. Overall, on the ramp in SEA I was getting 25 frames or so, and they were relatively smooth. At lower frame rates leaving Oculus AWS off provides a smoother experience for me. Once the frames get to 40 FPS, it’s smoother with AWS set to auto. The flight went with no problems at all — no crashes, no significant stutters, no hangs. The system preformed quite well. I find that in these types of scenarios — complex scenery, urban area, big airport, and weather — with Vulkan I just can’t push all the sliders to max in VR. My CPU just can’t take it (the 1080ti hums along just fine). But with them backed off just a bit, the performance is great. In more normal settings, I can up the anti-aliasing quite a bit (and turn off the supersampling). I also suspect the fastest performance I’ve had to date is to turn HDR off, and use supersampling with no anti-aliasing, but the shimmers are just hard to take. It’s clear to me, though, that Vulkan is providing sharper images in the VR headset than OpenGL did — I can read things just fine with HDR anti-aliasing on and no supersampling.
All this is to say that I’m quite pleased. Using VR to stay proficient is now really an excellent option as Vulkan makes the whole thing visually smooth and quite appealing on my system. It was good before, now it’s really good. And as new betas come online, I expect it to only improve.
Here’s a stream with yesterday’s VR testing. It runs very well for me, but if you keep fiddling with the settings it wants to give up and crash. But the performance is very smooth. In some later testing I found that ASW fixed 45 FPS worked best for me.
I could not resist, so today I backed up my 11.41 install and upgraded to the Vulkan beta. It took a bit to get things working — had to update the XP files a few times to overwrite some of the custom textures I was using as part of my Ortho installs — but after having just done a brief flight my initial report is:
Total game change. With my 1080ti and now-dated quad-core i7 I’m getting 20-25 FPS sitting on the ramp at Oxnard with Ortho, max objects, max reflections, HDR, 4x antialiasing+FXXA, clouds, max textures, draw airplanes, and shadows on. If I go down to 2xAA those numbers go up to 30 FPS. But as soon as I’m off the ground, they go up to 50 or so. Objects as far as the eye can see. Better, though, is the smoothness of the sim. As others on the .ORG have reported, even at 20 FPS the sim is extremely smooth with no stutters that I can see. In 11.41 I would expect to run with no shadows, 2xAA, one notch low on objects, and no reflections and be at right around 20-30 FPS.
Next step will be to try VR, and to turn on hyper threading not he CPU and see how the sim does with eight virtual cores rather than four physical. But so far, this is a remarkable improvement that I can only assume will improve. Well done Laminar team.
A buddy and I went flying the other day to build a little time, and I planned a route up to Daytona, then down to Cape Canaveral so we could request a low pass of the Shuttle Landing Facility. While all of the Cape sits within a restricted area, it’s only active around launches and it was quiet the day of our flight. And while you are not allowed to land at the Shuttle Landing Facility (KTTS), it’s a Class Delta airport much like any other, and you are able to request a low pass from the tower, which they nearly always approve. So we did, and they did.
It was something else to make a radio call that begins with “NASA tower,” and it was even better to set up on the same runway that welcomed so many Shuttle missions home and fly its length at 100 feet. The runway is 15,000 feet long, so even at 130 knots it took a while. One of the remaining Shuttles sits at the field, and the fly by provides great views of the launching pads and the VAB (Vehicle Assembling Building), the giant structure in which the United States assembled rockets prior to launch from Apollo to today. Here are some shots from the day. It was a great flight.