Apple M1 Test Flight

[Timi]

A quick test flight on a MacBook Pro M1 with a native arm64 build of FlightGear 2020.4.0.

[Fahim Dalvi]

Thats quite nice and decent FPS given the lack of a discrete GPU.

P.S. The screen occupying only a quarter of the screen is a known issue in OSG for HiDPi screens: https://sourceforge.net/p/flightgear/codetickets/2321/. You can either use Zakalawe's forked OSG (linked in the ticket) or apply the fix mentioned here: https://github.com/openscenegraph/OpenS ... -600080664

[Timi]

Awesome, thanks for the tip! I'll try to fix it.

[Timi]

Rebuilding OSG with the patch fixed the problem.

[Timi]

Another test with fullscreen this time.

[Fahim Dalvi]

Its great that the fix worked!

If you get a chance, it might be nice to see framerates and performance with heavier aircraft/sceneries/settings just to have some datapoints for the M1 chip!

[vnts]

Interesting. Is the GPU capability of the Apple M1 not powerful as advertised? What type of settings can you get without photoscenery, is it too slow for trees.. Edit: you may be CPU bound(?), the GPU utilisation stats or reducing window size should give an idea. The resolution to test should be 1080p or lower (the video says 4k but that might just be upscaling).

There were some benchmarks [1] that indicated performance around GTX 1050, which is about 5 years old now (GTX 3000's are the current gen). I thought it would do a bit better for a current release.

Apple started with the lower end in rolling out their APU/SOCs. Thier plan is to switch all, or at least most, of the lineup to their own designs. Not sure what to think about this latest APU trend (the slowest laptops until now were non-3d based Intel integrated GPUs and the recent AMD APUs). And just when it looked like the reduction in power looked like laptops people generally would catch up with desktops or 3d laptops . If the 1st version is this slow it means there are probably going to be more people in the future with laptops that are under powered, than without the APUs.

These laptops are very good on battery life being SoCs, but Apple's primary motivation in switching to their own CPU designs appears to be cost saving and a larger profit margin (in the very long run anyway, as yields take time to improve and chip production is an operation that needs large scale) - up-to 40-60% according to google [2]. APU/SoCs don't have to be slow by nature - the new AMD consoles are also APU/SoCs but they sync with more recent PC hardware (GPU equivalent to a RTX 2080). So this APU trend will /likely/ mean the desktop quality laptop hardware, which is finally no longer battery hungry, gets pushed to higher priced 3d performance laptops - and more underpowered APU laptops get marketed as regularly priced laptops. This latest trend is not so good from the perspective of people having better hardware, and FG rendering being able to assume more people being able to enjoy higher settings. People finding FG tend to have bought regularly marketed hardware, so at the very least they'll have slower hardware - as they don't know about CPU/GPU bottlenecks, or look at lists of relative benchmarks [3] [4] when looking at options for hardware. (I don't mean Timi here - I know his main system is pretty fast going by other videos in his channel, and he likely just brought the M1 as a low-power/mobile secondary computer)

Thats quite nice and decent FPS given the lack of a discrete GPU.

In this case the approach is an APU and system-on-a-chip (SoC). The M1 is seemingly better than the non-3d integrated GPUs that Intel used as throwaways even on their i5s in-cases where people normally use discrete GPUs, although integrated GPUs could probably(?) run the graphics settings in the video. These days the APU/SoC approach isn't restrictive - the new AMD consoles have a RTX 2080 and an 8 core Ryzen processor at ~3.5 GHz (google: [5] [6]) - the problem with no competition with the hardware monopoly, is consoles can fall behind PCs horribly. AMD and Intel will likely respond with some APU/SoCs laptops of their own, but the trend will likely mean these get pushed in the regular laptop slot, so it could well be the case of lots of people having slower laptops all over again - instead of regular laptops with fast hardware.

Kind regards

[Timi]

Photoscenery has no fps effect that I can see but as soon as there are buildings
and other objects the fps comes down. And shadows can be forgotten as well.

And let's say I can run X-Plane 11 on the same hardware anywhere between 30
to 50 fps depending on the settings. So I'd say the hardware is pretty good.
I guess the Vulcan/Metal works wonders here.

[vnts]

Photoscenery has no fps effect that I can see but as soon as there are buildings
and other objects the fps comes down. And shadows can be forgotten as well.

The reviews talked about 1050, or even 1650. A 1050-ish (3000 performance rating) is the current high-ish/max-ish system requirements: enough to run C172P at 1080p with: max shaders, default lods, trees/tree shadows at very high, non-OSM2City scenery on, and overlays at with transparency AA set to multi-sample. That's a whole different class of performance.

Objects might be making you CPU bound due to OSG scene traversal (?). If performance is what was said in articles, you should be able to turn overlays with transparency AA, cloud shadows, and set ALS shaders to max without being CPU bound.

Kind regards

[Timi]

Here's another test with maxed out settings.

[vnts]

About 7-11 FPS @ 1080p, and max settings (I think the performance monitor may be heavy on the CPU). That's a bit more like it . I think having OSM2City buildings, roads, pylons etc would kill FPS by making it CPU bound. Maybe only using OSM2City buildings would leave FPS a bit better. Lower flights can use a shorter LoD rough to compensate. Lowering trees to very high should help. Random scenery objects also helps make FG CPU bound, so they could be turned off.

It seems the M1 doesn't like the airport keep grass overlay or something.

Not sure what is going on with CPU performance, looking at the initial article. As for CPU performance, the single-threaded performance the early reviews talked about is right at the top, better than a Ryzen 9 5950X or an i9-10900k (link) which has a ratings of approx 3500 and 3100 respectively (link). By comparison a max-ish settings and 20-30 FPS minumum CPU before OSM2City would be around a i5 2500 which has a rating of ~1700. The M1 multi-threaded ratings were also claimed to be pretty good (link). FG only uses about 2 cores worth across all threads, so performance should be fine(?). I'm not sure what's going on.

Kind regards

[palkoc]

Hello guys. I have MacBook Air M1 2020 and I have installed FGFS V2020.3.18 (https://downloads.sourceforge.net/fligh ... 0.3.18.dmg). Here's how it looks once I run it:



What should I exactly do in order to get a realistinc rendering as on screencasts above?

Thanks alot.

PalkoC

[Parnikkapore]

I can't see the image (try opening the image link in incognito mode), but my guess would be increasing the settings in View → Rendering Options?

[palkoc]

Picture link corrected