Interview With Orbx’ John Venema

[Murmur 3,941]

I think (as HiFlyer) that the rendering engine of Aerofly FS2 remains the most efficient one. Probably not by the large margin it has today (because of the missing visual features I previously mentioned), but still capable of more performance for a given visual complexity.

I am glad that IPACS entered the flight sim market with a new paradigm, i.e. having max performance as their first priority, whereas for the last 20 years the conventional paradigm in the flight sim industry has been: "You get 10 FPS at max settings, 20 FPS are enough, and more than 30 FPS is too much" (an approach brought forward by the flight sim designers themselves). Fortunately things are changing, also thanks to the rise of VR requiring at least 45+ FPS and ideally 90+ FPS.

I'm not very convinced when people say that adding more (non visual) features in AFS2 will diminish performance. I think the performance on today flight sims is mostly bottlenecked by the rendering engine and the visual features, not by the other features. Compared to P3D, FS9 had basically the same capabilities in terms of AI, ATC and aircraft systems complexity, but on a top end PC you'll probably get hundreds of FPS in FS9 with maxed AI.

I also think it's not necessarily a matter of "no free lunch", what I mean is that we're still seeing limitations for example in P3D (blurries, terrain pop-up, etc.) probably because when the original code was written, the limited PC capabilities made the loading of terrain textures and mesh in chunks a necessity. So there could be reasons why new code could indeed be better and more efficient.

I like the approach of IPACS, and given their little staff, they've been very quick in converting AFS2 to Vulkan.

 

[HiFlyer 9,359]

5 hours ago, Paraffin said:

If the point you're making is that a "more recent" sim should be able to do more than a sim like P3D or XP using a legacy code base, then I'm not sure I buy that (with apologies if that's not the point you're making).

All flight sims share a common need to render 3D graphics at high enough frame rates to be flyable, while also running a bunch of other stuff under the hood like flight models and systems simulation. The "no free lunch" rule always applies when balancing complexity and frame rates.

I'll let Jet-pack, one of the developers, in a previous answer to this concern, say it all it more clearly and in a less long-winded way than I could manage:

Quote

 

- As I said multiple times in the past: just because the current sim doesn't implement everything a deep study level add on aircraft can offer - the performance will probably still be very good. Firstly, the frame rate and the complexity of an aircraft aren't directly linked in Aerofly: there is the physics core (CPU) which will see increased load but there is still a lot of headroom for it, and there is the graphics engine which runs on the GPU. If we increase complexity that doesn't affect the frame rate primarily, cause the complexity is run on the CPU and not the graphics card. And to get around the performance hit on the CPU we have built a very good foundation for the physics core. I've already implemented a lot of aircraft systems in the last two years and I didn't experience any performance issues. Dont underestimate the aerofly engine, or rather your CPU power, when used efficiently.

 

5 hours ago, Paraffin said:

For example, according to an older post in the dev blog, X-Plane doesn't support SLI because the frame rate would actually slow down when sharing rendering between two GPUs, due to the way graphics are handled in the sim. 

Even in the non-sim, gaming world, the use of multi-GPU setups has been fading in favor of single ultra-fast cards. That's partly due to developers not wanting the extra hassle of supporting it, and also the recent inflation of GPU pricing. For whatever reasons, it does seem to be a trend towards single cards now. I'm not sure Vulkan support for multi-GPUs is enough to reverse that trend. 

The "SLI" in the article I mentioned (I shouldn't really have even called it SLI) Is a bit different from what the dev blog you mentioned was talking about, and even that blog post was talking about X-plane 9, I believe.

As for the trend, you're right, I would previously have never been interested in an SLI setup. (I just drool over Robs toys)

The performance increase always seemed too negligible for a non-trivial outlay of money. On the other hand, there's the possibility of this new tech eventually allowing any GPU to be used with any other GPU.

What if that useless Nvidia card built into my motherboard could actually be enlisted to help out my main graphics card? Would I activate that feature? You bet I would; and I bet most of us would jump as well.

To me its an example of yet another feature never dreamed of when some of our current sims were on the drawing board. Something I suspect a newer/more agile, and yes, less complex sim (with less backwards compatibility to unravel) might more easily be able to use to its advantage as it finds its way in the next few years.

[Guest]

42 minutes ago, HiFlyer said:

As for the trend, you're right, I would previously have never been interested in an SLI setup. (I just drool over Robs toys)

I gave up on SLI a year or so ago, it was to driver dependent and I was getting tired of drivers working with SLI, then not working with SLI, then partially working with SLI ... each driver release was a roll of the dice.  

DX12 EMA mode SFR (frame pipelining article) is the operating construct that will leverage the best possible performance from one or more GPUs (any GPU, they don't have to be the same).  However, LM feel this would require considerable re-working of code and they aren't sure of the return on investment (they've heard good and bad), also have to consider that DX12 is not supported on Win7 (which still is 39.4% of the Windows family OS share with Win 10 at 47.2%).

Making performance compares and assumptions requires a detailed supporting feature list for whatever one wants to compare.  I believe iPACs are primarily focused on keeping it simple, visually nice, and ensuring high FPS.  If you're looking for AF2 as a replacement for P3D or XP11, I don't think that was ever iPACs agenda?  Or is it?  I "thought" they just wanted to bring a beautiful flight simulator that is smooth, enjoyable, not expensive, and works well on a variety of hardware from low to high end ... and they've succeeded in exactly that. 

With all that said, I'm pretty confident there are some performance optimization for P3D and XP11 in the near future, maybe not double FPS, but I could imagine maybe 25% to best case 50% improvement for both platforms.  Compatibility will always be a ball and chain, with P3D being the FS platform that has tried to retain as much compatibility as possible over the years ... a solution to this is two render paths (old compatible path vs. new less compatible path) ... this could be as simple as a checkbox on the graphics UI (simialar to FSX DX10 mode) or executing a different main EXE.

Cheers, Rob.

[HiFlyer 9,359]

31 minutes ago, Rob Ainscough said:

If you're looking for AF2 as a replacement for P3D or XP11, I don't think that was ever iPACs agenda?  Or is it?  I "thought" they just wanted to bring a beautiful flight simulator that is smooth, enjoyable, not expensive, and works well on a variety of hardware from low to high end ... and they've succeeded in exactly that. 

You would have to ask them.

For myself, I feel like they go back and forth, as in, I think at first they wanted to create the best sim evah'

Then, when it looked like the community was kinda "Meh, we luvs our current sims." they seemed to back off a bit and follow the money (and the love) from the VR community.

But.

I suspect in their hearts they would really love the chance to create the best thing evah' 

Don't most entrepreneurs (even if only secretly) want to be king? 😃

[jabloomf1230 2,445]

All of these gimmicks like SLI are end runs around the real solution which is faster hardware. Hardware is the main factor that limits performance in any flightsim. After that all the (cough, cough) legacy code in Windows, its video drivers and other APIs like OGL/DirectX are the main contributing factors. Lastly, the incessant piling on of    both bells and whistles features and complex add-ons is the remaining component. 

The actual code in each flightsim is a trivial factor relative to the ones previously listed.

 

[Paraffin 2,197]

4 hours ago, Murmur said:

I'm not very convinced when people say that adding more (non visual) features in AFS2 will diminish performance. I think the performance on today flight sims is mostly bottlenecked by the rendering engine and the visual features, not by the other features. Compared to P3D, FS9 had basically the same capabilities in terms of AI, ATC and aircraft systems complexity, but on a top end PC you'll probably get hundreds of FPS in FS9 with maxed AI.

I was curious about the "hidden" impact on FPS with systems modeling, including more cockpit instruments, so I tried some comparisons. This isn't a carefully controlled test, just one quick look at a few different aircraft in one sim platform, to see the effect on FPS with different aircraft. 

Setup is XP11 with my normal settings, main forward cockpit view, loading on 3nm approach to KSEA runway 34R (lots of Seattle outskirts buildings), clear skies:

Default Aerolite 103 --------- 44 fps

Default Cessna Skyhawk -- 42 fps

Default Boeing 747-800 ---- 34 fps

Carenado Pilatus PC-12 ---  32 fps

Default Boeing 747-400 ----  30 fps

So, something is dropping the frame rate besides just the eye candy outside. It might be the load on the CPU calculating flight models for the faster aircraft, although I've noticed that ramping up the number of flight models per frame doesn't really cause much of a hit. It could be the increased number of gauges and displays, although with the normal view on approach there isn't that much of the cockpit visible. I suspect it's more the plugins running the complex aircraft systems. Especially the PC-12, which has always had a fairly heavy hit on frames, and I suspect it might be a systems plugin that isn't well optimized. There's a reason why that ultralight is the frame rate champ -- there is almost nothing modeled for aircraft systems.

Anyway, I agree that visual effects are the main bottleneck, because I can easily bring my system to its knees just by turning on object shadows and throwing a bunch of heavy cloud layers in the sky. But if this test is any indication, there is another impact on performance that tracks with complexity of the aircraft being modeled. At least in this particular sim, and I suspect that's a universal no-free-lunch principle. If and when a sim like AFS2 starts including fully-modeled aircraft systems, there's no guarantee they'll be able to keep the same performance.

Edited July 17, 2018 by Paraffin

[HiFlyer 9,359]

 

1 hour ago, Paraffin said:

So, something is dropping the frame rate besides just the eye candy outside.

Yup, you will find that in X-plane as well. I've done the experiment of turning everything down until there is essentially nothing onscreen, And something's still sucking frames. I've asked several times in various places over time about whut' the heck might be going on.

Mystery's of the sims (cue x-files music)

Its one of those things I probably shouldn't worry about, but still I keep going "Hmmmmmmmmm...."

[HiFlyer 9,359]

1 hour ago, Paraffin said:

I suspect that's a universal no-free-lunch principle. If and when a sim like AFS2 starts including fully-modeled aircraft systems, there's no guarantee they'll be able to keep the same performance.

You could be right. (we'll see) My (that's just me) instinct however, is to trust a programmer who is nose-to-nose with the code everyday, until he proves himself to be wrong, or at the very least, exaggerating.

[HiFlyer 9,359]

1 hour ago, Paraffin said:

I suspect it's more the plugins running the complex aircraft systems. Especially the PC-12, which has always had a fairly heavy hit on frames, and I suspect it might be a systems plugin that isn't well optimized.

If you read their forum, Ipacs is very alert to anyone from the outside getting into their code and possibly slowing it down. The idea catches their attention right away, and they always suggest would-be plugin people wait and let them implement whatever it is.

I've heard that's also why laminar implemented their own universal GPS: to keep outside developers from creating some big hog that slowed down the sim.

After the shutdown of Aces, the FSX addon/plugin world became kind of like the wild wild west, where anyone with the skill could pretty much throw anything they wanted into the pot, and X-plane certainly followed suit. That was absolutely a great thing for choice and diversity, but maybe not so hot for speed and stability......

[Murmur 3,941]

1 hour ago, Paraffin said:

There's a reason why that ultralight is the frame rate champ -- there is almost nothing modeled for aircraft systems.

But you can't say if it's mostly due to the systems modeled, or to the less complex virtual cockpit/external model.

I bet you will get a better performance on the default KC10 or C130 than on the Aerolite though. And those two have more complex systems than the Aerolite, although not using external plugins. But they don't have a virtual cockpit, which also explains their high performance.

Another example: Falcon BMS has probably the most detailed aircraft ever, down to its last system, with 2 MFDs and a HUD, multiple indipendent weapon sensors (Radar, FLIR, TGP, etc.), a fully modeled FBW computer, etc., and at min visual settings, on my old laptop I get 100+ FPS in VC.

With the headroom AFS2 currently has, modeling more complex systems could be doable without excessive drag.

[Murmur 3,941]

41 minutes ago, HiFlyer said:

If you read their forum, Ipacs is very alert to anyone from the outside getting into their code and possibly slowing it down. The idea catches their attention right away, and they always suggest would-be plugin people wait and let them implement whatever it is.

That is a good assurance that performance will be always optimized at the very best (that's probably the right thing to do), but this closed model becomes a limitation for the 3rd party ecosystem. Unless maybe they implement "building blocks" for aircraft systems (like you mentioned days ago, I think) allowing a compromise between optimized performance and flexibility in modeling more complex systems. But this should expand to avionics too. Customizable avionics/gauges is essential.

[Paraffin 2,197]

4 hours ago, Murmur said:

But you can't say if it's mostly due to the systems modeled, or to the less complex virtual cockpit/external model.

Why not? That ultralight model showing the best frame rate has almost no systems, and it also has the widest canopy (no canopy) view of the outside terrain. Something like 80% of the forward view is pure scenery. If it was just the outside scenery that was a hit on frame rate on this system, that model should have the lowest frame rate in the test. Especially on my fairly low-end GPU.
 

Quote

I bet you will get a better performance on the default KC10 or C130 than on the Aerolite though. And those two have more complex systems than the Aerolite, although not using external plugins. But they don't have a virtual cockpit, which also explains their high performance

Well, of course! Those are old XP10 or XP9 models that don't make full use of the current XP11 features like PBR rendering and windscreen/instrument reflections, or high res cockpit textures. They don't use plugins for more advanced systems modeling. It's not just the basic 2D cockpit that boosts the frame rate. 

It's when we load more advanced aircraft with plugins to extend the capabilities of the sim, that we start stressing the system.

[Guest]

I guess it is impossible to know right now if the performance of AFS2 will go down when more systems will be added. We will just have to wait and see. I do know though that with the latest update, which did bring more systems (specially in the Airbus bit also for instance a HUD on the Q400), fps mainly went up for most users instead of down. 😉

The upcoming TrueEarth Netherlands will be a nice test to see what complex scenery will do with performance! Flying the Airbus or Q400 above that scenery should give a nice indication of what's possible in AFS2.