How much more can FSX handle? (be warned, lots of opinion)

[BaldyB 20]

 

 

As long as it's open to third party developers there is no danger of dumbed down physics taking a back seat to visual appearance.

You hit the nail on the head Kevin IMHO. MS default aircraft IMHO never even came close to what PMDG and Level-D's 757. Years ago I flew in the real world too. I lucked out and had a fellow student practice with/on me. The fact is he was going for Inst., instructor and I private pilot. He already had private, commercial, Multi Engine, Etc. He had me under the hood many, many times. It's amazing how you can get so in tune with your airplane you almost don't need to get nervous you just handle the situation. If your upside down you believe whats in front of your face and  NOT the seat of your pants. Put the hood on do a few stalls or whatever to put the aircraft in an abnormal state, then tell me to unfold my arm's and recover the plane attitude assume straight and level flight and take him to an airfield of his choice and shoot some instrument landings. After a few weeks he said your no fun any more. To me it was like going to buy a pack of butts. His was Steve Chapdelain, he wont to work for the FAA and was stationed in Denver, last I knew. I know it may sound stupid but when I ran excavators and a rock got stuck in between the buckets teeth, my teeth began to bother me. When I raced stock cars I felt like I was almost part of the car.  A man can get that way with a machine!  A class 4 flight simulator machine is probably the closest you will ever get. Perhaps some form of and the right equipment using VR will be closer?

Their is, sort of a lot to think about.

 

Baldy 

[sascha 314]

 

 

In contrast X-Plane is far too lively and skittish, especially when simulating large aircraft.

 

It is said that X-Plane is very accurate IF you use very high-quality control devices.

[kevinh 409]

It is said that X-Plane is very accurate IF you use very high-quality control devices.

Yes a lot of things are said about how accurate X-Plane is, but they still have to keep on improving it. PMDG have described the problems they had making a large multi engined piston powered aircraft work well with it. Like anything else, an X-Plane aircraft can only fly as accurately as the designer makes it. Garbage in, garbage out.

 

I don't think it's reacting to controller noise. If it is it's far too sensitive to very small inputs. Don't get me wrong, I like both FSX and XP. Both have good and bad points.

[MorsAbAlto 19]

 

 

A GPU can calculate a graphical effect for display. It could possibly use the shape of the model, plus additional data relating to airflow, to compute aerodynamic forces. But how does it feed that back to the sim? It might become possible in the future but for a simulation it's better to keep the core of it in the CPU.

 

It would feed back to the sim like any GPU does to a program. GPUs are used for way more than just graphics these days. NVIDIA and AMD are both building cards for Deep Learning and other applications.Calculating fluid dynamics is done with GPUs. Look at shared computing like BOINC and how it uses GPUs for all kinds of calculations. Folding proteins, analyzing radio signals, etc..

 

Things like NVIDIA PhysX exist. I really don't see why this could not be done in a flight sim in real time with a degree of accuracy that would improve upon existing consumer market simulations. 

 

 

 

DTG's new flight sim could well do exactly that. As long as it's open to third party developers there is no danger of dumbed down physics taking a back seat to visual appearance.

 

That is a big IF right there. Why would they deviate from their current policies and offer such a sophisticated product? I sure wish they would and if they do I'll be the first to buy it and apologize. But until then I remain highly skeptical. That said, LockMart would be the more likely candidate to develop the flight model as you suggest - but again not for the private customer. 

 

The 3rd party market is full of talented people. I hope we will have a sim in the future that provides a more sophisticated basis for their offerings. We all seem to agree that FSX has certain limitations and that even things like Active Sky are limited in their fidelity due to this. 

[kevinh 409]

Things like NVIDIA PhysX exist. I really don't see why this could not be done in a flight sim in real time with a degree of accuracy that would improve upon existing consumer market simulations. 

 

It may be possible but I don't think it's necessarily desireable. It relies on there being a good CFD physics engine in the GPU. So the flight simulation is only as good as the GPU's CFD implementation. Apart from anything else, to get an accurate flight simulation you would need accurate 3D data for the aircraft. The equations used are complex. The advantage of the data table method is that you don't need this proprietary data and the calculations are much less intense.

 

 

That is a big IF right there. Why would they deviate from their current policies and offer such a sophisticated product?

 

Indeed it is. They are already attempting to do so, but we have to wait and see how good it ends up being.

[MorsAbAlto 19]

 

 

The advantage of the data table method is that you don't need this proprietary data and the calculations are much less intense.

 

Oh yes, that is something I have not considered before. Since we got very accurate visual models from PMDG in our sims it was my assumption that the exact shape of the airfoil and such things were available to the public, but it makes sense if they are not. 

 

The advantage with GPUs is the high amount of data they can process versus a CPU, so these intense calculations might be possible on them? 

 

I am a layman concerning CFDs but this is what NVIDIA offers on the matter, so I let you be the judge:

http://www.nvidia.com/object/computational_fluid_dynamics.html

[downscc 1,327]

 

 

The advantage with GPUs is the high amount of data they can process versus a CPU, so these intense calculations might be possible on them?

 

There's no need. If you were running a finite element analysis engineering study to predict airfoil coefficients then okay, but the coefficients are a given and this kind of parallel processing is not required. Not even close.

[crvm 0]

I see sims such as AeroFly 2 becoming the top dog in this arena. With such fluid graphics that run on practically any machine, it's hard to go back to the slide show MSFS plus serious addons deliver. 

 

 Sure we can argue 'realism' but I believe it's well on it's way to taking the next steps. Can you imagine PMDG aircraft in AeroFly 2? I can. 

[Paul_Smith 92]

 

 

...Apart from anything else, to get an accurate flight simulation you would need accurate 3D data for the aircraft...

Exactly how difficult do you think it is to get 3D data for an aircraft? Is there some arcane knowledge that is required about the shape of a plane that can not be determined (or at least estimated) by just looking at it? Sorry Kevin, I don't know if you are trolling or just mis-informed, but I am struggling hard to find anything you have said on this topic that I can agree with. And what was that BS about how a GPU "...could possibly use the shape of the model, plus additional data relating to airflow, to compute aerodynamic forces. But how does it feed that back to the sim? It might become possible in the future but for a simulation it's better to keep the core of it in the CPU"? Since you obviously have no idea about how a GPU or any other co-processor works, I will try to explain it for you. Feel free to ask questions about any bits you don't understand. 

 

Most complicated programs are built with a combination of coordinators, schedulers and workers. Coordinators are responsible for determining what resources are available and how they will be utilized by different jobs, and schedulers decide what jobs need to be done and in what order and are responsible for managing the input and outputs of those jobs. Workers do the jobs and for each distinct job there will be a worker that knows how to do it. The jobs can usually be descried as 'Apply this action to this data and update the results'. It could be a big job such as 'lower the landing gear' or a little job such as 'calculate the acceleration on a given object from applying a given force'. Big jobs are usually made of lots of little jobs. Some little jobs will be done a lot so it would be quite normal to have different versions of the job that are optimized for different environments. Now this is the bit you would be interested Kevin, if the coordinator has detected the presence of a physics co-processor that supports a mutually understood interface, and if the resource allocation budget allows, then it can assign the co-processor versions of scheduled jobs to that co-processor. If there is one or more GPU's with a supported interface and available resources, then it can assign the GPU version of pending jobs to the GPU. If there is a CPU available, it can assign the CPU version of the jobs to the CPU. In the worst case when resources are exhausted, it could decide to assign the low CPU version of the job that only approximates the results, you know, the one that does a table lookup. 

 

Table lookups provided a balance between processing cost and accuracy of the results that was a good solution when it was introduced, but was nearing the end of its acceptable life at the turn of the century. I remind you that back then, one GFLOP of processing power cost about $1000. Today it costs less than $0.08. What possible reason do you have to continue to defend its use?

[scandinavian13 5,192]

 

Sorry Kevin, I don't know if you are trolling or just mis-informed, but I am struggling hard to find anything you have said on this topic that I can agree with. And what was that BS about how a GPU "...could possibly use the shape of the model, plus additional data relating to airflow, to compute aerodynamic forces. But how does it feed that back to the sim? It might become possible in the future but for a simulation it's better to keep the core of it in the CPU"? Since you obviously have no idea about how a GPU or any other co-processor works, I will try to explain it for you. Feel free to ask questions about any bits you don't understand. 

 

There's truth to what he said. X-Plane uses the 3D model itself to calculate forces (this is where the oft celebrated blade element theory comes in). This could be made a lot more efficient by using the GPU's processor as it's literally made for raw throughput in this number-crunching and physics-based environment. Unfortunately, none of the major sims seem to be taking advantage of this, and CPUs are simply not as good at it when compared to the GPUs of today (particularly the higher-end ones) at this task.

 

It may be time to evaluate your emotional connection to the argument versus the actual arguments. Debate using facts. Keep the ad hom and colorful language out of it.

[MorsAbAlto 19]

There's truth to what he said. X-Plane uses the 3D model itself to calculate forces (this is where the oft celebrated blade element theory comes in). This could be made a lot more efficient by using the GPU's processor as it's literally made for raw throughput in this number-crunching and physics-based environment. Unfortunately, none of the major sims seem to be taking advantage of this, and CPUs are simply not as good at it when compared to the GPUs of today (particularly the higher-end ones) at this task.

 

It may be time to evaluate your emotional connection to the argument versus the actual arguments. Debate using facts. Keep the ad hom and colorful language out of it.

 

Ah! Someone in the know chimes in to support my point. Thank you very much! It's such a mystery why GPU power is not being used for this. You would think that LockMart would be the first to implement something like that, but maybe it's not possible with the FSX/ESP code base? 

 

This is starting to go off the rails a bit, but has lead to an interesting discussion so far. 

 

As a customer I wish there was some kind of PMDG roadmap to estimate when it makes sense to switch platforms and to which platform in particular. The market seems so fractured and convoluted for someone who has been away from simming for some time. Three big sims and developers cross-release their products. 

 

So a direct question at this point: Do you anticipate functional differences between the FSX and P3D version of the QotSII or other products in the near-term?

[kevinh 409]

Exactly how difficult do you think it is to get 3D data for an aircraft? Is there some arcane knowledge that is required about the shape of a plane that can not be determined (or at least estimated) by just looking at it?

 

If you want an accurate simulation you would need accurate 3D data. The exact 3D shape is pretty much proprietary data. Boeing wouldn't want Airbus to easily access such information and vice versa. Yes you could get approximate data, but then you would be faced with the task of fine tuning the 3D model to adjust and calibrate sim performance. Far easier to adjust tabulated data.

 

Sorry Kevin, I don't know if you are trolling or just mis-informed, but I am struggling hard to find anything you have said on this topic that I can agree with. And what was that BS about how a GPU "...could possibly use the shape of the model, plus additional data relating to airflow, to compute aerodynamic forces. But how does it feed that back to the sim? It might become possible in the future but for a simulation it's better to keep the core of it in the CPU"? Since you obviously have no idea about how a GPU or any other co-processor works, I will try to explain it for you. Feel free to ask questions about any bits you don't understand. 

 

Suggesting I'm trolling is trolling in itself. Being inaccurate is not trolling. I've always regarded a graphics card as an output device, not a coprocessor. So I'm happy to admit I was wrong in that regard. Are you happy to admit you were wrong to suggest FSX was not physics based? Being patronising doesn't help you either.

 

Table lookups provided a balance between processing cost and accuracy of the results that was a good solution when it was introduced, but was nearing the end of its acceptable life at the turn of the century. I remind you that back then, one GFLOP of processing power cost about $1000. Today it costs less than $0.08. What possible reason do you have to continue to defend its use?

 

Interesting then that all Full Flight Simulators use table lookups. Last century technology? I don't think so. CFD based sims are good for aerodynamic development, getting things right before an aircraft flies. They are nowhere near so useful for flight simulation in pilot training, let alone hobby use..

[mtrainer 136]

Hello,

 

With regard to letting the massive parallel pipeline afforded by today's GPUs do all the flight dynamics, I know from experience how incredibly difficult parallel programming can be.  It's a bear.  Consider the Sony PS3 system- a marvelous machine for its day but incredibly difficult to code for, which is why they abandoned their Cell Architecture and went back to an Intel based (albeit AMD) solution.

 

And many problems don't lend well to a parallel solution - for instance raising an integer to a very large power (previous output needed before continuing to next the step).   Consider that the ULTIMATE FLIGHT SIMULATOR would be a simulation of the planet earth, with the plane simply being the focal point.  It's a darn tall order....with many parts well suited to parallelism, others not at all. 

 

Just the same, the immediate future looks to bring us continued advances in parallelism and what may actually be a reduction in clock speed due to thermal issues- so....here we are and someone is going to have to master these difficulties if we are to take home based flight simulators to their next logical level.  Which is something we all would appreciate.

 

All the easy coding is behind us- to take it to the next level is going to require PhD level folks who are skilled across multiple disciplines - mathematics, programming, aeronautics, and physics just for starters.  And as I stated earlier in the thread, the kind of capital to pull something like this off, from scratch, boggles the mind.

 

I'm mostly a forum lurker but this topic is one that has been on my mind for many years, which is why most of my posts go long.  Here is a tiny bit of background-

 

In the early to mid 80s I was majoring in computer science and taking one of many calculus classes when I started talking to a fellow about my age, early 20s, who was taking the same class.  Unlike most of the other students who were simply doing the minimum to get by, this fellow was front and center with me and was taking the class extremely seriously and asking great questions.  When I asked him what is major was, he indicated he wasn't there to get a degree- he was a software engineer on a full motion KC-10 Extender Singer-Link System at Barksdale AFB, Bossier City Louisiana (where my dad had just retired).  He was tasked with programing the "bow wave" effect experienced by a KC-10 that was refueling from a KC-135 and he had to pull it off using assembly language and keep CPU resources to a bare minimum....the machines were already at 90% CPU capacity (yes physically large machines...that took up the basement of the complex).

 

From that moment on, I knew I wanted to be a software engineer for industrial flight simulators.  Writing programs that mimic real world events is, to me, the most rewarding type of development out there.  While I never realized that particular dream, I did get to fly that simulator on more than one occasion and it remains one of the biggest thrills of my life.

 

The future of MSFS will be baby steps forward I'm afraid, and perhaps not entirely a bad thing.  Table based lookups will be replaced only when it makes sense to do so, from a simulation/physics perspective as well as a cost/benefit perspective.

 

Like all things, if you don't understand something a company is or is not doing, simply follow the money.

 

Mark Trainer

[Paul_Smith 92]

If you want an accurate simulation you would need accurate 3D data. The exact 3D shape is pretty much proprietary data. Boeing wouldn't want Airbus to easily access such information and vice versa. Yes you could get approximate data, but then you would be faced with the task of fine tuning the 3D model to adjust and calibrate sim performance. Far easier to adjust tabulated data.

 

Suggesting I'm trolling is trolling in itself. Being inaccurate is not trolling. I've always regarded a graphics card as an output device, not a coprocessor. So I'm happy to admit I was wrong in that regard. Are you happy to admit you were wrong to suggest FSX was not physics based? Being patronising doesn't help you either.

 

Interesting then that all Full Flight Simulators use table lookups. Last century technology? I don't think so. CFD based sims are good for aerodynamic development, getting things right before an aircraft flies. They are nowhere near so useful for flight simulation in pilot training, let alone hobby use..

I am trying to take Kyles points on board so I will stick to the facts. To get an accurate simulation you need accurate data, and the more accurate your data, the more accurate your simulation. If you look at a fuzzy photo of a plane, you will get a fuzzy impression of its 3D shape. The better the photos and the more of them you have, the more accurate your model of the shape becomes and the more accurate your simulation of its performance. Since the results of a simulation can be compared against the published information for well known aircraft in different regimes, the accuracy of the models can be tested and improved. There is nothing special about the 3D shape of any object, so there is no reason to think it would be in any way difficult to get, after all, the aircraft is parked on the tarmac right in front of of you.

 

I am sorry that this will sound patronizing, and despite Kyles hints to the contrary I will say it anyway. As I am a physicist by training and a software developer by career, I can say with some authority that the table lookup system that the Microsoft flight simulation franchise is based on is not physics based. It is a software solution that was designed to replace physics based calculations with crude approximations in an environment where a Cray super computer from 1985 has the same processing power as an landfill Android phone from today. In other words, an environment that no longer exists.

 

Still sticking to the facts, not all flight simulators use lookup tables. Not even most of them. Physics based flight simulators have been around since the original Star Wars arcade game using vector graphics on a processor less powerful then my watch. And while you are entitled to your opinions on the quality and suitability of CFD based sims, you will not be surprised to find that a rather large number of CFD based sim users would consider those opinions to be as accurate, fact based, evidence supported and up-to-date as your opinions on physics, look up tables and GPUs.

[kevinh 409]

I am trying to take Kyles points on board so I will stick to the facts. To get an accurate simulation you need accurate data, and the more accurate your data, the more accurate your simulation. If you look at a fuzzy photo of a plane, you will get a fuzzy impression of its 3D shape. The better the photos and the more of them you have, the more accurate your model of the shape becomes and the more accurate your simulation of its performance. Since the results of a simulation can be compared against the published information for well known aircraft in different regimes, the accuracy of the models can be tested and improved. There is nothing special about the 3D shape of any object, so there is no reason to think it would be in any way difficult to get, after all, the aircraft is parked on the tarmac right in front of of you.

 

Paul,

 

With respect it isn't that simple. Modern aircraft are full of very subtle multiple curvatures. Supercritical wings have all kinds of unexpected reflex curvature. You can get something that is a fair approximation from photographs, but if you want to fine tune it to be a closer approximation you will have to subjectively modify the 3D model by trial and error until it performs better. Which part do you adjust though, where do you start? It isn't that obvious. You could be at it for months. With a data table approach such "tweaking" is very much easier.

 

 

I am sorry that this will sound patronizing, and despite Kyles hints to the contrary I will say it anyway. As I am a physicist by training and a software developer by career, I can say with some authority that the table lookup system that the Microsoft flight simulation franchise is based on is not physics based. It is a software solution that was designed to replace physics based calculations with crude approximations in an environment where a Cray super computer from 1985 has the same processing power as an landfill Android phone from today. In other words, an environment that no longer exists.

 

Yes, it does sound patronising and you are also indulging in semantics. I'm a professional aeronautical engineer by training and career, so I understand what I'm talking about (though you clearly don't recognise that). Coefficients of lift, drag, pitch, etc aren't physical quantities, so you might not recogise calculating them as "physics based". All a data table does is calculate them from inputs such as angle of attack, sideslip and Mach number. Then you can calculate lift, drag, pitching moment, etc and use these forces and moments in the simulator's equations of motion. What isn't physics based about that?

 

 

Still sticking to the facts, not all flight simulators use lookup tables. Not even most of them. Physics based flight simulators have been around since the original Star Wars arcade game using vector graphics on a processor less powerful then my watch.

 

 

No Level D qualified full flight simulator uses CFD methods. Not a single one. The flight test data provided by aircraft manufacturers doesn't support that approach.

 

Which CFD based sims are you talking about, by the way? X-Plane certainly isn't one.

 

And while you are entitled to your opinions on the quality and suitability of CFD based sims, you will not be surprised to find that a rather large number of CFD based sim users would consider those opinions to be as accurate, fact based, evidence supported and up-to-date as your opinions on physics, look up tables and GPUs.

 

You said you weren't trying to be patronising earlier, but the BIB shows that is exactly what you were being. Perhaps you should have taken Kyle's advice after all?