[1.3.1] Ferram Aerospace Research: v0.15.9.1 "Liepmann" 4/2/18

[etheoma]

On a pleasant note Unity 5 IS thread safe so the move over when it comes will make things a little more enjoyable for FAR with regard to multithreading. I do wish Unity had gone through with the switch to Bullet that they hinted at before the U5 announcement though since it would have meant a platform agnostic acceleratable physics base...

KSP is already multithreaded the problem is that the phyic's is not multi threaded which is like the biggest bottleneck EVER for this game considering KSP is a physics based game. and ferrem in his last post said that FAR runs on its own thread as is, I don't know if unity5 would allow him to use more than 1 thread for the hole of FAR though.

And it hasn't been stated the the phyic's engine for Unity will become multi threaded in U5, which is the bottleneck and any other threading is kinda superfluous to KSP, you could probably free up another maybe 10% of the physics thread by multi threading some of the other stuff, further more even just a 64 bit client probably wont come with U5 because its a problem with getting at deeper functionality in Unity that has prevented a stable 64 bit client which probably wont change in U5, so I'm not all that excited.

You can prove that KSP is multithreaded because if you only give it access to 1 thread the music and some other things become Skippy which I wanted to limit it to 1 core because I could turn you the turbo boost on my CPU when only 1 core is being utilized to like 4.5Ghz 4.3 - 4.2 when 2 cores are active which is most likely because there are still background operation's.

Edit; Sorry my bad yes the physics engine has been slated to be made multi threaded, so yay and given optimization made in the physics which is more what I heard it would be we might even get 4x performance on a average PC of today, U5 will not necessary mean a 64 bit stable client on windows though.

And yeah being able to delegate some of the physics to GPGPU would have been nice, but ehhh.

Edited April 29, 2015 by etheoma

[marach]

And it hasn't been stated the the phyic's engine for Unity will become multi threaded in U5

Umm you know U5 released in like January right? U5's physics handling is in fact as I said threadsafe now.

The main problem for 64bit in U4 was that the debugger could only handle 32bit software (yeah it's f'd up but that's what happened) the new debugger is 64bit friendly and you know works which is a big plus!

[etheoma]

Umm you know U5 released in like January right? U5's physics handling is in fact as I said threadsafe now.

The main problem for 64bit in U4 was that the debugger could only handle 32bit software (yeah it's f'd up but that's what happened) the new debugger is 64bit friendly and you know works which is a big plus!

Well considering Squad them selfs have said that U5 does not necessarily mean a stable 64 bit client and far from it in fact... Ehhhh no offence but trust squad one that one.

And I did correct my self on the physics, the last time I looked over the specs of U5 was in mid 2014 so hands up my bad. But I corrected my self before I saw your comment so I would like that to be noted.

Edited April 29, 2015 by etheoma

[ferram4]

@etheoma: JSYK, building the voxels is already multithreaded, and on my little potato it yields ~2x speedup above singlethreaded. The main problem is that there is always the chance that the threads try to work on the same section of the voxel, and so some performance needs to be given up to prevent them from clashing. In addition, the actual solidification pass can only manage ~2x speedup, ever. The method simply doesn't lend itself to multithreading given the amount of data that would need to be transferred between threads and the cost of keeping them synced to keep problems from showing up there. Seriously, I tried multithreading that even more and I got worse performance because of the sync costs.

I think the voxel generation is already fast enough for our purposes, and its memory usage isn't that bad for the data that it holds, and considering that it's optimized for speed rather than memory usage.

[marach]

Well considering Squad them selfs have said that U5 does not necessarily mean a stable 64 bit client... Ehhhh no offence but trust squad one that one.

And I did correct my self on the physics, the last time I looked over the specs of U5 was in mid 2014 so hands up my bad. But I corrected my self before I saw your comment so I would like that to be noted.

Yeah 64bit will still take time but it /should/ be possible though clunky at first. The physics change is perhaps the thing I like best in U5 threadsafe physics makes many things much nicer to use. The other reason I would have liked bullet is it's hardbody interactions are quite nice and would likely have meant opencl being integrated with U5 which would as you said earlier have been great for things like the voxel code here. Tbh though Ferrams maths code is some of the nicest I've seen in years kinda wish squad had taken him up on his offers to fully integrate

[Gfurst]

By the way, just wanted to point out something:

The new stock dynamics somewhat seems better than older version of FAR, Just tested though with a small plane craft that I was capable of building with early career mode parts.

The really small craft seems pretty stable and is quite agile, now descent and losing speed seems much more accurate, as I've mentioned before with previously well build craft, loosing speed for landing was a big hassle and I practically couldn't stall the craft.

Now loosing speed with going for a descend and/or nearly stall seems much better, the landing got a lot easier, even considering the bumpingness of the runway at level 0.

At the same time though, with the new revamp of engines thrust and etc, I can't even get higher than 13k altitude, which also seems accurate and harder for some contracts.

[Cinocal]

Was thinking the 1.0 Aero would be good, but after much experimentation. It is faking the funk that is FAR.

The stock drag model is a joke.

[ferram4]

Considering that oldFAR overestimated subsonic drag, and that your response is that you're happy because it has higher subsonic drag, are you sure that your measure of "better" is really all that helpful to me? I'm going for the most realistic aerodynamics I can manage, not personal opinion of what "feels" right. As I believe I've mentioned before, if you can point to papers or data that include models and correlations that get the performance that you'd like to see that prove that it's realistic, I'll gladly implement it. Now, show me the data on how it compares to reality vs. oldFAR.

[mariohm1311]

Well said, ferram. Before we had to put up with noobs saying FAR is too difficult, now we have to put up with noobs saying stock is more realistic...

Edited April 29, 2015 by mariohm1311

[CDRW]

I've been wondering, are aerodynamics affected at all by the scale differences between KSP planets and real planets?

[Hodo]

By the way, just wanted to point out something:

The new stock dynamics somewhat seems better than older version of FAR, Just tested though with a small plane craft that I was capable of building with early career mode parts.

The really small craft seems pretty stable and is quite agile, now descent and losing speed seems much more accurate, as I've mentioned before with previously well build craft, loosing speed for landing was a big hassle and I practically couldn't stall the craft.

Now loosing speed with going for a descend and/or nearly stall seems much better, the landing got a lot easier, even considering the bumpingness of the runway at level 0.

At the same time though, with the new revamp of engines thrust and etc, I can't even get higher than 13k altitude, which also seems accurate and harder for some contracts.

Considering that in old FAR I was able to make a to scale replica of a Mig-21 and it handled pretty much like a Mig-21 is stated to handle, it even topped out at near the same speed for the mass of the craft and TWR. The only difference was the Mig-21 I made could climb much higher then the real one. But that was an atmosphere scaling issue more than a FAR issue.

The NuStock is a good model, WAY better than the old but it is not FAR or even NEAR. It is more or less something between.

And Ferram4 I would just ignore anyone who says FAR is easier than stock. I have heard it before the 1.0 came out and I laughed at them.

[tetryds]

People have different opinions and different biases about how aerodynamic effects work.

I understand the fact that some people will have a wrong bias, and that is due to being exposed to inaccurate models their whole life, but that does not mean they are inferior.

Please, keep the posting here civil, feedback on FAR is aimed at realism, because it's meant to be a realistic model, how it feels is not a thing here, that is true.

So rather than point fingers at them, simply tell them that.

Stock is not as accurate as nuFAR or FAR 0.14, but it's way better than the previous stock model, which is very good.

The stock game is not meant to be as accurate as possible, it's meant to be good enough, and forgiving, which it is, and in my opinion it fits the stock game perfectly.

So, stop talking about people who rather stock as it is now than FAR, you are not better than them, and they are not better than you.

About people who think that stock is more accurate than FAR (better is too subjective), either explain them what the two are about or leave them.

[mariohm1311]

I'm tired of explaining that on reddit and getting downvoted to hell because:

1) Nobody can complain about Squad's work.

2) You're supposed to have fun playing a game (while it has been proofed that most of the people like realism).

[toril]

Considering that in old FAR I was able to make a to scale replica of a Mig-21 and it handled pretty much like a Mig-21 is stated to handle, it even topped out at near the same speed for the mass of the craft and TWR. The only difference was the Mig-21 I made could climb much higher then the real one. But that was an atmosphere scaling issue more than a FAR issue.

The NuStock is a good model, WAY better than the old but it is not FAR or even NEAR. It is more or less something between.

And Ferram4 I would just ignore anyone who says FAR is easier than stock. I have heard it before the 1.0 came out and I laughed at them.

FAR is easier then stock always has been, if your a trained pilot. I could never fly or design anything that flew anywhere remotely like a plane in the old placeholder aero. People always said the old placeholder was like flying in soup, but I disagree it was like flying in some state of matter not currently known to man. Closest I could ever come to describing it, was like flying through silly-string. FAR was much, much better but still off a bit. What ferram has accomplished with FAR is nothing short of amazing to me, when you realize he did it all in his free time, and the handicaps he had to work with. Without FAR I would never have been able to play with anything that looked remotely like a plane in KSP.

[riles]

Im not a good player at all, but I have really enjoyed Far and Near, Both seem to be more intuitive to me. I dont know the math or modeling required to make this stuff works but man I really enjoy that I have something closer to home to play around with

[Juncoph]

nuFAR hype train, anyone?

[Hodo]

FAR is easier then stock always has been, if your a trained pilot. I could never fly or design anything that flew anywhere remotely like a plane in the old placeholder aero. People always said the old placeholder was like flying in soup, but I disagree it was like flying in some state of matter not currently known to man. Closest I could ever come to describing it, was like flying through silly-string. FAR was much, much better but still off a bit. What ferram has accomplished with FAR is nothing short of amazing to me, when you realize he did it all in his free time, and the handicaps he had to work with. Without FAR I would never have been able to play with anything that looked remotely like a plane in KSP.

Not going to disagree with you on most of that. I will disagree with you on the trained pilot part. You just have to have a basic understanding of how aircraft work, that doesn't require me to be a trained pilot just someone who paid attention in science class when we made paper airplanes in the 6th grade.

[Tarheel1999]

NuStock is an improvement over the old model but it is neither better nor more realistic than FAR or NEAR it is merely different. In fact there are several areas where it fails completely and does not even come close to modeling the forces that should be at play.

For example medium to heavy crafts do not pick up nearly enough speed while descending. I had an SSTO fail due to insufficient LFO. Rather than gain speed during a descent from 10.000m it was steadily losing speed

as I attempted to pilot it towards land.

Also with a small craft I am able to do tight loops at 40-50 m/s while flying through the still soupy lower atmosphere without stalling or losing control.

Edited April 29, 2015 by Tarheel1999

[toril]

Not going to disagree with you on most of that. I will disagree with you on the trained pilot part. You just have to have a basic understanding of how aircraft work, that doesn't require me to be a trained pilot just someone who paid attention in science class when we made paper airplanes in the 6th grade.

I should probably rephrase that to, "from a pilots point of view", non exclusive.

[Feradose]

nuFAR hype train, anyone?

hyped indeed. I wasnt able to fly without FAR and still cant fly without FAR in souptovacuumin15kmeterosphere. And then nuFAR saves us!

[Van Disaster]

I'm sure a lot of effort went into new stock, but even as someone who isn't a properly trained pilot but is just good with dynamics, something is off enough about it that I can't use it - I'm fairly sure it is actually still soup, so I wonder what it'd be like used for submarines.

I'd be happy to wait another couple of months for voxAir ( FARTHER? FLAPS? I'm sure someone can make a backronym out of the latter ) if it needed it. 0.90 didn't magically combust

[Dinokin]

nuFAR hype train, anyone?

Hype Plane*

[4plains]

I've been wondering, are aerodynamics affected at all by the scale differences between KSP planets and real planets?

Short answer:

No.

Long answer:

Not really. The scaled down planets have a faster atmosphere density scaling, meaning that it becomes less dense quicker than on Earth. But as long as the (simulated) composition of the atmosphere you're flying through isn't changed with the planet size aerodynamics don't change.

[StevenRS11]

Considering that oldFAR overestimated subsonic drag, and that your response is that you're happy because it has higher subsonic drag, are you sure that your measure of "better" is really all that helpful to me? I'm going for the most realistic aerodynamics I can manage, not personal opinion of what "feels" right. As I believe I've mentioned before, if you can point to papers or data that include models and correlations that get the performance that you'd like to see that prove that it's realistic, I'll gladly implement it. Now, show me the data on how it compares to reality vs. oldFAR.

I wonder if there is some super-accurate but slow aerodynamics simulator out there that can simulate arbitrary 3D models that's not some ultra-pricey piece of software. I could export KSP ships as a model (maybe .stl?), have it simulate them, and compare the results to what FAR gives.

The idea being that the separate piece of software doesn't have to work in real time, so it could afford to use far more computationally expensive but more accurate (and maybe simpler) methods. It might be useful for finding edge cases that FAR doesn't handle well, and if FAR does a really good job, then it would definitely be a source of bragging rights.

"Hey, I made a mod for KSP that simulates aerodynamics as accurately as this research tool. Oh, and mine does it in real time, too."

Or you may already do this, who knows. I have access to a university license of Solidworks Flow, so I may fiddle with that. Flow is geared more towards lower speed stuff like fluid in pipes, though. I know it does some aerodynamic stuff, but I'm not sure if it will do stuff like supersonic wave drag or compression lift, though. Depends on how general it's fluid dynamics solver is, I guess. One way to find out!

[4plains]

@ferram4:

All this discussion about nuFAR, oldFAR, nuStock and stuff has me ask myself a couple of questions:

-Will oldFAR get updates for KSP 1.0 and following versions? Because I REALLY like oldFAR.

-With the fact that nuFAR now does things based on a voxel based approximation of the aircrafts shape. That could approximate the leading edge of the wing to be rounded. Round airfoils have problems at supersonic speeds with shock waves and have very high drag.

So will there be a option that allows us to configure the simulated airfoil shape for supersonic flight.

Of course having the simulated wing optimised for supersonic airflow conditions will mean worse sub sonic handling than smooth wings designed for subsonic flying conditions.

Just a bunch of random questions. Thanks for answering in advance.