@@Bubbly_Dragon Once chicago lifted half a city block of brick buildings and it weighted at 35000 tons, so assuming a whole city block would be 70000 tons, you could probably lift around 10 blocks. hmmmm. Also, this came from 3 minutes of googling.
@@JonaB03 rimworld has full modding capabilities though, with the right tools and knowledge you can do literally anything. But in KSP, there are hard coded limitations that would be difficult to bypass.
Remember when this used to be a voiceless channel about challenge ideas in KSP. Now it's Challenge Ideas while disassembling KSP on a technical level. Love it even more
@@Stratzenblitz75 the way you methodically disassemble KSP just fascinates me. Your landspeed record exploring how terrain works, mass cannon exploring how acceleration works. Megaton Lifter exploring how lag works.
@@Stratzenblitz75 side note. Crossfeed causing exponentiall lag is such an obvious answer if you start thinking about it. Odd that it didn't come to mind until pointed out.
@@GrandProtectorDark Yea it's quite a "Oh... duh!" moment. Like giving birth to sextuplets and realizing you only have 2 breasts. Life comes to a grinding halt during feeding time! Well, ok, bad analogy come to think of it, as you could just offset feeding them to groups of 2, but still... Babies haven't yet grasped the concept of "logic" yet, and neither has the Unity Engine on which KSP operates! :P
@@sparty9938 If not, it burns up in the atmosphere creating a very high firework explosion, or possibly heating up the atmosphere to oven like temperatures.
my man with the NASA supercomputer fears Stratzenblitz. Legend has it that his absence was spent solely on recording the footage for this mission alone.
@@Stratzenblitz75 I want to do this.... but... 2 things, don't have making history and if your computer struggled... well... roses are red my screen is blue I think I just flew a 1 megaton boom that is what would happen to my computer
I think this goes to show how important procedurally scalable parts are, being able to use a single large tank of arbitrary size is a lot more efficient than many smaller ones
That would help, certainly, but I don’t think it’s necessary to fix the cross feed issue. What ksp should do is approximate each crossfeed region into a single big fuel tank. Then, each engine only has to consult one virtual tank instead of fifty real ones. Then, once all the engines have checked in, the virtual tank updates the levels in the real fuel tanks. With some clever math, you can make the system behave exactly the same as the current one from a user and craft design perspective.
There are lots of ways to make an optimized system for fuel flow, I'm sure the devs could figure it out. But scalable parts are also really important for physics engine performance. You could make many fuel tanks transfer fuel as efficiently as one big tank, but the big tank will always be easier to simulate in terms of physics
@@TheArtikae Hmm. I don't think that would be too difficult to implement either, as you could just subtract from the virtual fuel tank the combined amount of fuel used by all connected engines. (only has to be calculated when the craft is loaded in and when an engine is destroyed) This is probably riddled with practical issues I'm not seeing, but this would theoretically work. Oh, and this makes it so you don't have to update a lot of parts. Just display it as an individual fuel amount, when in reality it's only 1 virtual tank, and you just divide the virtual fuel tank level by the number of fuel tanks.
@@blakeforland8238 I should clarify, I did not finish grad school. I decided it wasn't the best route for me so I moved on to other things. But it was in mechanical engineering
The issue with the grass isn't that it isn't perfectly flat, it's that the rocket is so long the planet was literally curving generating a bulge to destroy the outer edges of the rocket.
@@Bitgedon This is why planes roll forward on the KSP runway: it *is* perfectly flat, so gravity pulls them slightly forward until they reach the center.
I would love to see what happens when you turn it 90 degrees and the bottom part in orbit slightly touches the atmosphere while the top part is still in space. It should start to flip very slowly. while also the rotation would be towards the direction of the orbit, which means it could be just right, that the rotation euqlas out with the rotation around Kerbin...but I dont know.
Nah it would just not interact with the atmosphere, physics with atmospheric simulation is only done if the rockets centre of mass is under 70 KM of Kerbins surface.
My computer would do what his did if I attempted this........ well sort of. I do have a 5900x with dual 360MM radiators, but I also have 64GB ram and I'd be surprised if he does. The ram probably doesn't affect it, but who knows. Maybe it does. I've seen some bizarre behavior that I didn't expect with this much ram. I've seen processes on my old 16gb i7-2600 that didn't seem to care about a 16gb limitation go past 30gb. Also I doubt the geforce 780 I currently have installed really affects it much either as we all know KSP doesn't really care about graphics unless it's intel. Either way I do have a 3070 (also on the same water setup) in hand that I'll install this weekend. So next KSP time for me will be on the new card.
I only started getting into it last year or so, but I'm enjoying it a ton! Building PCs and tuning parts is a lot of fun (especially RAM, that was fun to try). Its awesome how you can really make your computer personal.
Should be fairly easy to send a massive payload to every single body this way. Make an orbiter, lander and (if possible) rover for each planet and moon
Multicore support is difficult to implement ... and in some cases it would not be a miracle solution ... You can only parallelize calculations in a "good" way if they are NOT dependent on each other ... Example ... You have 2 Rockets and you would tell you CPU Core 1 calculate rocket A, Core 2 calculate rocket B ... If for some reasons Core 1 is faster with calculating its stuff but Core 2 ist not ... Core 1 NEEDS to wait for Core 2 ... If the execution of stuff is strongly linear, parallelizing stuff is almost impossible .... In KSP paralleling stuff would be great for stuff like Core 2 please calculate the Space Trash in the background while i fly this Rocket and the Rocket is calculated completly on Core 1 ... I know, this is EXTREMLY simplifyed but works as example ...
@@buddycat2757 the cpu in your computer had multiple cores in it. Each core can execute 1 instruction at a time e.g. add 2 numbers together. If your computer has 4 cores then it can execute 4 instructions at once. Ksp only uses one of these cores, no matter how many you have, so is giving up a lot of potential speed
@Stratzenblitz75 I think the performance drop is due to fact that the fuel cross feed system requires more calculations since fuel is drawn from all tanks at the same time, it requires more processing power to calculate the fuel informations such as how much fuel is being used, How much fuel is left etc as compared to a non fuel cross feed rocket
@@orbitalengineer2800 With enough development time, it's probably possible to isolate the parts of the fuel flow calculation that are constant between one physics frame and the next and cache those results.
It makes sense that cross-feeding the fuel would decrease performance with x^2 parabolic curve rather than a linear one. If each engine is accessing each fuel tank then the effective interactions per part are engines * fuel tanks, instead of just engines + fuel tanks. Though I am surprised that the direct-feed was so linear, as presumably, each section interacts with every other surrounding section and through them the next section, increasing the total number of interactions in a sum(n) from 0 to n function, which is also a parabola and not linear.
This is a basic graph problem. Given the graph doesn't change outside of staging, O(n^2) is a worst possible performance solution. At worst, this could be optimizable to O(n*log(n)), and I wouldn't be surprised if there is an O(n) linear solution for crossfeed. The bigger concern I have is with 10 years of development, did they never perf their game and figure out where they were bottlenecking?
as I started watching this video, I was thinking "hmm, wonder how much better KSP would be with large crafts, going from my old intel 3700k to my new amd 5950x" and then you actually go ahead and do this for me. thanks for saving me the trouble bro!
3930k master race here, DDR3 is the superior RAM & everyone else is lying. Lol, but I really am stoked on the 10 year run this CPU has given me more or less while still keeping up in most things.
Personally the most interesting part of this video is the discovery that fuel crossflow causes so much of the lag associated with larger ships. I wonder how those calculations are being handled to cause that much lag and if there's an easy fix to it, or a fix at all
There are definite fixes for it. Fuel drain should be of constant rate until either the user adjusts the throttle (or otherwise alters feed rate) or some subset of the tanks empty. This is something that has an easy analytical solution, so, when crossfeed has been constant for a couple of frames, one's algorithm could save the current rate and project/copy-paste it forward (just subtracting a precomputed fuel fraction at each step) until the point in that function at which one or more of the tanks completely drains, or until user input triggers it to revert to the old, "manual" calculation for a bit. It might still get laggy for a moment when a tank fully drains, RCS kicks on/off or the user adjusts the throttle--but it should be *loads* faster the rest of the time, when nothing in particular is happening.
Stratz you madman. I'm now expecting a "toolbox throw to Duna" as per our conversation on Discord. Edit: Also, my respect for Knight of St John has increased substantially, which I didn't think was even possible after that river run.
I hope KSP2 devs are watching. Building big is fun, but low frame rate detracts from that fun. Anything that can improve performance, like fixing fuel flow logic, is a huge win for everyone who dreams big.
I love seeing the lower debris encounter the atmosphere and get dragged back, actually a very good visualisation of just how rapidly air pressure/drag decreases with height.
@@chrisgaming9567 1 Million tons is actually ahead of my normal ball park - Saturn V to Saturn is 218,000t, so this is 5x bigger! Mind you, I do have larger vehicles, just not one I've made a video on.
Actually, I am curious what performance increases could be had by having multiple "smaller" rockets (say *Only* 500-1000 parts or so), and just having A LOT of those. If I remember correctly, KSP doesn't really have a system for splitting a single craft across multiple processing threads, effectively throttling the performance on a multicore/thread CPU. Combine this with the fact the CPU you used has 24 of these threads, and it might be possible to have significantly more performance available, even if only exploitable across a fleet of craft in physics range.
I don't think KSP calculates crafts in multiple threads. That would be really annoying to make work, and would lead to very odd behavior in some situations. Which you see none of.
This is literally 200,000 tons heavier than the entire Golden Gate Bridge. It's as heavy as the Golden Gate AND the Sears Tower combined, with enough leftover for 30 Statues of Liberty and 30 Ford Tauruses.
Imagine being a kerbin resident, and its just after evening and your sitting on your porch, and then you see hundreds of rocket boosters burning up in the atmosphere
Did you do any tests of the fuel lines? It looked quadratic, which suggests that the problem is that every engine was querying every fuel tank it has access to. Fuel lines might behave differently. Makes you wonder if there is any smart abstraction or caching at all, or whether it queries the entire construction tree until it hits something it cannot pass fuel through every frame*. That would suggest to always put as few parts as possible on engine sections, and separate using parts that cannot transfer fuel. *Oh god, it gets worse. That stuttering looks like the garbage collector, suggesting that the engines are building a list of tanks to drain from, and discarding it, every frame. If you have 500 engines connected to 2000 fuel tanks that means 1 million allocations per frame from this alone. I used to think programming games was hard, because I would look at a problem like this and reject the first solution I came up with, based on it being moronic. Now I know better.
A mod could probably fix this issue quite easily o believe. I am not good at coding but I believe a good method would be: 1. Generate groups of connected fuel tanks and give them an overarching total fuel volume variable 2. Find all thrusters connected to each tank group and generate total fuel consumption variables for those groups 3. Now each consecutive fuel calculation only has to take the total fuel consumption from the total fuel volume and then update the total fuel volume for each group. I think this could massively increase the performance as the bulk of the calculations only have to be done once at the very beginning
I still remember finding you at 750 subs, and i have fi ally returned to the game after 5 years to find you are still pushing the limit of this game. bravo
This is super surprising! I would've never thought that fuel crossfeed would affect performance so much. It's much simpler computationally compared to physics. Devs definitely should take a look at their code, it could bring us a lot of frames.
It's interesting hearing the wide variety of opinions on this matter. Some people think this makes a lot of sense, whereas others think the crossfeed should be much less significant. It's hard to tell since we don't have access to the game's code, but its great to hear the ideas!
Holy shit, I was already in awe watching that behemoth fly up, but when the first stage of boosters peeled off like that and collided .. wow man. Amazing
patience IS a virtue, this video would not be without it! also the knowledge of "what" make the game lag can benefit a lot of us on how to optimize, like making our own tanks with more fuel to reduce crossfeed and not just part count!
The civillain kerbals during the last part gonna be looking at the boosters like 'Ahh, lovely meteor shower.' and ground control is just praying for their lives
I was looking for KSP benchmarks when I bought my 5900x back in December, coming from an 6700k and I was curious how much of a performance boost it would be. Cool video !
I am actually super impressed with the fact that the physics engine (minus fuel crossfeeding) scales linearly in processing time with part count. That is a major feat for KSP's engine!
I've always wanted to build the largest space station I could in ksp. Thank you for this knowledge, I definitely won't use this for anything nefarious.
My man got carpet bombing to a whole new level at the end... Really impressive, would love to see a full megastructure being brought to orbit next. Keep up the good work 👍
The scaling behavior you saw makes a lot of sense. The big trick KSP uses to simplify physics calculations is that the part connections form a tree (i.e. no loops) so you can propagate forces through all the parts in linear time O(N). The fuel consumption network is a general directed graph, so solving the fuel updates has super linear complexity (IIRC it's roughly O(N^3) ) in the number of network nodes. It's possible the devs realized this would be a bottleneck and decided not to do a full fuel consumption solve on every tick, leading to the stuttering you observed. The more you can disconnect the fuel network, the closer the time complexity will get to being linear. You shouldn't need to remove cross-feed entirely to get a big performance gain.
My god...it’s full of boosters. Lovely piece of work! I never got really deep into KSP because I’m not imaginative enough, so I enjoy watching your videos. Keep it up!
Computer scientist here, the fact that it's linear means the devs put a lot of effort into optimization. It's actually pretty impressive. anything sub n^2 is considered pretty good but we tend to aim for log(n) time complexity or nlog(n)
That staging explosion was probably one of the best KSP1 explosions I've ever seen. lol It's also nice to see a 5900x benchmarked like this since I have one.
Now KSP is not only getting bug-tested by the community, there's even people doing profiling ;D Very interesting and in-depth but still entertaining, as always, loved the professionalism at 5:28
I previous one I launched did not have the retro rockets and is still up there. You can see the debris pass over in a giant cloud over the ksc every orbit
In short, you have bought a new CPU and taught it to fear you.
fear will keep them in line
its better that sending a icbm at it
Gotta teach em young!
If i could I'd add 1 billion likes to this comment
Very much me
This is a very creative way for a whole city block to evade taxes
Don't need to pay taxes if a megaton rocket falls on top of you!
I read that as: "This is a very creative way for a whole city block to evade Texas"
@@ZeroneAngel same
@@ZeroneAngel ...Purely hypothetically, what would the weight of a city block reach?
@@Bubbly_Dragon Once chicago lifted half a city block of brick buildings and it weighted at 35000 tons, so assuming a whole city block would be 70000 tons, you could probably lift around 10 blocks. hmmmm. Also, this came from 3 minutes of googling.
When Stratenblitz starts drawing graphs of frame rate versus part count, you know he's about to show you powers that cannot be learned from a Jedi.
Basically every technical player will run into the way which real scientists does.
I'm glad that we are finally able to understand the science of ksp lag. I can think of a few plane builders who could really use this ;)
Hopefully some Modder will optimise KSP's fuel crossfeed process.
@@5000mahmud Modders can do a lot, but honestly I don't think that that is something easly fixable without messing with the actuall executable
Maybe they'll make new optimisation mods for lower end PC's or for giant crafts.
@@JonaB03 rimworld has full modding capabilities though, with the right tools and knowledge you can do literally anything. But in KSP, there are hard coded limitations that would be difficult to bypass.
I don't think it would be _impossible_ but you would need a lot of time and effort, probably a ritual sacrifice or two.
Remember when this used to be a voiceless channel about challenge ideas in KSP.
Now it's Challenge Ideas while disassembling KSP on a technical level.
Love it even more
Thank you! I'm enjoying the voiceovers a lot, so I'm glad you're liking it!
@@Stratzenblitz75 the way you methodically disassemble KSP just fascinates me. Your landspeed record exploring how terrain works, mass cannon exploring how acceleration works. Megaton Lifter exploring how lag works.
@@Stratzenblitz75 side note. Crossfeed causing exponentiall lag is such an obvious answer if you start thinking about it. Odd that it didn't come to mind until pointed out.
@@GrandProtectorDark Yea it's quite a "Oh... duh!" moment. Like giving birth to sextuplets and realizing you only have 2 breasts. Life comes to a grinding halt during feeding time!
Well, ok, bad analogy come to think of it, as you could just offset feeding them to groups of 2, but still... Babies haven't yet grasped the concept of "logic" yet, and neither has the Unity Engine on which KSP operates! :P
@@DUKE_of_RAMBLE That's the most beautiful analogy. I'm keeping it lol
Pretty sure that staging event at 18:00 constituted a war crime...
The debris that was left from the staging is is probably enough to decimate any city on earth
My man just straight up orbitally bombed an entire continent.
and he did it *twice*
The fuel at lift off has energy comparable to a decently sized thermonuclear bomb.
@@sparty9938 If not, it burns up in the atmosphere creating a very high firework explosion, or possibly heating up the atmosphere to oven like temperatures.
This video: Exists
My computer: *nervous sweating*
my man with the NASA supercomputer fears Stratzenblitz.
Legend has it that his absence was spent solely on recording the footage for this mission alone.
Don't be nervous, the download link is in the description :)
@@Stratzenblitz75 I want to do this.... but... 2 things, don't have making history and if your computer struggled... well...
roses are red
my screen is blue
I think I just flew
a 1 megaton boom
that is what would happen to my computer
@@monkeyanimationandgaming lmao
@@monkeyanimationandgaming just pirate the game abd woah! You have making history
I think this goes to show how important procedurally scalable parts are, being able to use a single large tank of arbitrary size is a lot more efficient than many smaller ones
That would help, certainly, but I don’t think it’s necessary to fix the cross feed issue. What ksp should do is approximate each crossfeed region into a single big fuel tank. Then, each engine only has to consult one virtual tank instead of fifty real ones. Then, once all the engines have checked in, the virtual tank updates the levels in the real fuel tanks. With some clever math, you can make the system behave exactly the same as the current one from a user and craft design perspective.
There are lots of ways to make an optimized system for fuel flow, I'm sure the devs could figure it out. But scalable parts are also really important for physics engine performance. You could make many fuel tanks transfer fuel as efficiently as one big tank, but the big tank will always be easier to simulate in terms of physics
@@TheArtikae Hmm. I don't think that would be too difficult to implement either, as you could just subtract from the virtual fuel tank the combined amount of fuel used by all connected engines. (only has to be calculated when the craft is loaded in and when an engine is destroyed) This is probably riddled with practical issues I'm not seeing, but this would theoretically work.
Oh, and this makes it so you don't have to update a lot of parts. Just display it as an individual fuel amount, when in reality it's only 1 virtual tank, and you just divide the virtual fuel tank level by the number of fuel tanks.
glad to see that $50k PhD is paying for itself already
@@Stratzenblitz75 glad you're happy, great vid btw
Is A PhD for 50k is worth the XP, greatest video in a while.
@@Stratzenblitz75 What's your PhD in ?
@@blakeforland8238 I should clarify, I did not finish grad school. I decided it wasn't the best route for me so I moved on to other things. But it was in mechanical engineering
The issue with the grass isn't that it isn't perfectly flat, it's that the rocket is so long the planet was literally curving generating a bulge to destroy the outer edges of the rocket.
I laughed at first but then I remembered ksp is 1/10th the size and it hit me that might actually be what’s happening
@@Bitgedon This is why planes roll forward on the KSP runway: it *is* perfectly flat, so gravity pulls them slightly forward until they reach the center.
@@waterlubber Huh. That actually explanes a lot
@@Bubbly_Dragon Bruh lol
B-but kerbin is flat bro.. this stupid kerbal space program is a lie lol
Stratenblitz then: Look, with 3 parts i went to eeloo and back
Stratenblitz now: sit down son, we're gonna have serious hardware discussion.
I would love to see what happens when you turn it 90 degrees and the bottom part in orbit slightly touches the atmosphere while the top part is still in space. It should start to flip very slowly. while also the rotation would be towards the direction of the orbit, which means it could be just right, that the rotation euqlas out with the rotation around Kerbin...but I dont know.
Nah it would just not interact with the atmosphere, physics with atmospheric simulation is only done if the rockets centre of mass is under 70 KM of Kerbins surface.
@@pokemonfanmario7694 okay, didnt know that. but then it effects all of the spacecraft?
@@ClemensAlive Anything within a 2.5 kilometer sphere of a vessel you are personally looking at is calculated in physics.
The atmosphere is not a sudden stop. It is more like a gradient.
@@ninjanicholas2459 in ksp its very much a sudden stop. You can feel the lag as it starts calculating aerodynamic forces
Child: Can we have Gamers Nexus?
Mom: No, we have Gamers Nexus at home.
Gamers Nexus at home:
BRO xDDD
a german gamers nexus is better than the original, huh
@@512TheWolf512 i mean, we have der8auer over here, so gernany might already have a small Bonus on this kind of Hardware UA-camr xD
@@p3chv0gel22 you kind of need der bauer and gamers nexus cause of the bromance. lol
The excitement I felt when I saw stratzenblitz
I just started the video, so no original comment
Thanks! I'm excited to have you here :)
Me too, I know it's not original but it's true. I think it's rather best to keep that to yourself because almost everyone here is a huge fan.
@@Stratzenblitz75 I finished the video and am wondering: How many debri strikes do you risk?
I think that Matt had a stroke when he saw all that debris. Either that or he drank a bottle of whiskey. 🥃
** hwisky
*hwiskey*
My computer would murder me if I attempted this...
My laptop would cook me alive
My computer would do what his did if I attempted this........ well sort of. I do have a 5900x with dual 360MM radiators, but I also have 64GB ram and I'd be surprised if he does. The ram probably doesn't affect it, but who knows. Maybe it does. I've seen some bizarre behavior that I didn't expect with this much ram. I've seen processes on my old 16gb i7-2600 that didn't seem to care about a 16gb limitation go past 30gb. Also I doubt the geforce 780 I currently have installed really affects it much either as we all know KSP doesn't really care about graphics unless it's intel. Either way I do have a 3070 (also on the same water setup) in hand that I'll install this weekend. So next KSP time for me will be on the new card.
Me to
my macbook air in cheapest config would probabaly just shut down forever xD
i can barely launch a minimalistic mun lander
my computer tried to murder me for even watching the video lmfao
19:23
How to induce kessler syndrome with this one simple trick!
i instantly thought of the same thing! :)
true!
The worlds heaviest harmonica
This is deeply satisfying in the same way turning a 1 hour long python script into a minute long one is.
You mean it's satisfying to directly use C? Yes
I'm always happy to see a fellow pc nerd.
I only started getting into it last year or so, but I'm enjoying it a ton! Building PCs and tuning parts is a lot of fun (especially RAM, that was fun to try). Its awesome how you can really make your computer personal.
@@Stratzenblitz75 nice to see
Now you gotta actually put 2 Komodo's into orbit
Make a 250kt rocket and put it in Orbit with this
Should be fairly easy to send a massive payload to every single body this way. Make an orbiter, lander and (if possible) rover for each planet and moon
Yo, it’s the ringy boi himself. Also, he could probably fit 5 Komodos on top
Starblitz
Me: I wish he would upload again
New
Me: *Happy noises*
The fact that KSP 2 WON'T have multicore suport is criminal
Multicore support is difficult to implement ... and in some cases it would not be a miracle solution ...
You can only parallelize calculations in a "good" way if they are NOT dependent on each other ...
Example ...
You have 2 Rockets and you would tell you CPU Core 1 calculate rocket A, Core 2 calculate rocket B ...
If for some reasons Core 1 is faster with calculating its stuff but Core 2 ist not ... Core 1 NEEDS to wait for Core 2 ...
If the execution of stuff is strongly linear, parallelizing stuff is almost impossible ....
In KSP paralleling stuff would be great for stuff like
Core 2 please calculate the Space Trash in the background while i fly this Rocket and the Rocket is calculated completly on Core 1 ...
I know, this is EXTREMLY simplifyed but works as example ...
I mean the video shows that multiple tanks per engine is a problem. I see no reason that whatever they're doing there wouldnt be parralizable
what is multicore in KSP? i havent heard of it before
@@buddycat2757 the cpu in your computer had multiple cores in it. Each core can execute 1 instruction at a time e.g. add 2 numbers together. If your computer has 4 cores then it can execute 4 instructions at once. Ksp only uses one of these cores, no matter how many you have, so is giving up a lot of potential speed
ohh! thank you!
I wonder if the KSP devs can do something with this information, at least for the sequel.
I hope there's something they can do about it. The performance drop is simply insane. Even just having 2 fuel tanks per engine is noticeably worse.
@@Stratzenblitz75 Perhaps they aren't aware of it yet
@Stratzenblitz75 I think the performance drop is due to fact that the fuel cross feed system requires more calculations since fuel is drawn from all tanks at the same time, it requires more processing power to calculate the fuel informations such as how much fuel is being used, How much fuel is left etc as compared to a non fuel cross feed rocket
So perhaps there isnt much they can do about it
@@orbitalengineer2800 With enough development time, it's probably possible to isolate the parts of the fuel flow calculation that are constant between one physics frame and the next and cache those results.
Got any lag complaints? Write them on a piece of paper, and tuck it in between the wall of boosters.
The fact that you made this mod makes my respect for you even higher, which I didn't even know was possible
It makes sense that cross-feeding the fuel would decrease performance with x^2 parabolic curve rather than a linear one. If each engine is accessing each fuel tank then the effective interactions per part are engines * fuel tanks, instead of just engines + fuel tanks.
Though I am surprised that the direct-feed was so linear, as presumably, each section interacts with every other surrounding section and through them the next section, increasing the total number of interactions in a sum(n) from 0 to n function, which is also a parabola and not linear.
This is a basic graph problem. Given the graph doesn't change outside of staging, O(n^2) is a worst possible performance solution. At worst, this could be optimizable to O(n*log(n)), and I wouldn't be surprised if there is an O(n) linear solution for crossfeed.
The bigger concern I have is with 10 years of development, did they never perf their game and figure out where they were bottlenecking?
This rocket: "Hey friction, can ya just step out for a sec?"
Friction: "Aight, Imma head out"
Look at the size if this lad
*ABSOLUTE UNIT*
Indeed
Is this sponsored by AMD? 😂
Sadly not. These cpus were expensive!
@@Stratzenblitz75 of course they are a top class cpu means lots of money
maul clemens du bist unlustig
Stratzenblitz: Just quicksave and quickload!
Elon Musk: write that down, write that down!
Won't work in real life
@@benjaminbrown3939
I should hope not.
KSP+pc hardware=me super exited
as I started watching this video, I was thinking "hmm, wonder how much better KSP would be with large crafts, going from my old intel 3700k to my new amd 5950x" and then you actually go ahead and do this for me. thanks for saving me the trouble bro!
3930k master race here, DDR3 is the superior RAM & everyone else is lying. Lol, but I really am stoked on the 10 year run this CPU has given me more or less while still keeping up in most things.
@@von... I mean honestly bro, if it's working then it's working. May as well wait it out for KSP 2 in 2038
Personally the most interesting part of this video is the discovery that fuel crossflow causes so much of the lag associated with larger ships. I wonder how those calculations are being handled to cause that much lag and if there's an easy fix to it, or a fix at all
There are definite fixes for it. Fuel drain should be of constant rate until either the user adjusts the throttle (or otherwise alters feed rate) or some subset of the tanks empty. This is something that has an easy analytical solution, so, when crossfeed has been constant for a couple of frames, one's algorithm could save the current rate and project/copy-paste it forward (just subtracting a precomputed fuel fraction at each step) until the point in that function at which one or more of the tanks completely drains, or until user input triggers it to revert to the old, "manual" calculation for a bit. It might still get laggy for a moment when a tank fully drains, RCS kicks on/off or the user adjusts the throttle--but it should be *loads* faster the rest of the time, when nothing in particular is happening.
I can HEAR Steve Burke pulling out his hair in the background about "NOT ENOUGH GRAPHS"
Stratz you madman.
I'm now expecting a "toolbox throw to Duna" as per our conversation on Discord.
Edit: Also, my respect for Knight of St John has increased substantially, which I didn't think was even possible after that river run.
I hope KSP2 devs are watching. Building big is fun, but low frame rate detracts from that fun.
Anything that can improve performance, like fixing fuel flow logic, is a huge win for everyone who dreams big.
Oh god
His performance with his komodo is my usual frame rate
OH NO THIS IS WHAT HE WAS WORKING ON THESE 3 MONTHS
Those three months were spent flying the rocket.
3:15 dude's talking about in game parts' full names like he's got a sponsorship deal from kerbodyne.
This sounds like the best possible sponsorship
I love seeing the lower debris encounter the atmosphere and get dragged back, actually a very good visualisation of just how rapidly air pressure/drag decreases with height.
Now THAT, THAT is a thumbnail.
I came here to see the giant rocket and stayed for the interesting dive into KSP performance, very interesting!
Well, you launched something comparable in size in RSS, so, idk what's more impressive
@@chrisgaming9567 1 Million tons is actually ahead of my normal ball park - Saturn V to Saturn is 218,000t, so this is 5x bigger!
Mind you, I do have larger vehicles, just not one I've made a video on.
Have we gonna found a way to attack the kraken?
Nukes
Tryed and failed. As punishment I lost my rocket station.
Actually, I am curious what performance increases could be had by having multiple "smaller" rockets (say *Only* 500-1000 parts or so), and just having A LOT of those. If I remember correctly, KSP doesn't really have a system for splitting a single craft across multiple processing threads, effectively throttling the performance on a multicore/thread CPU. Combine this with the fact the CPU you used has 24 of these threads, and it might be possible to have significantly more performance available, even if only exploitable across a fleet of craft in physics range.
Thats an interesting idea: I think it should improve performance signifigantly
I don't think KSP calculates crafts in multiple threads. That would be really annoying to make work, and would lead to very odd behavior in some situations. Which you see none of.
@@nikkiofthevalley Your saying in ksp you don't see weird behavior in some simulations of Kerbal "glitch maker" Space Program, really?
Rip to all the Kerbals on Kerbins surface, they are going to suffer another mass extinction worst than that giant crater sea...
" ... and step three is rushing to the extinguisher when your PC catches on fire "
But there are no megastructures bigger than the love your patrons have for you.
(Maybe when I join them I'll be a part of that club)
This is literally 200,000 tons heavier than the entire Golden Gate Bridge. It's as heavy as the Golden Gate AND the Sears Tower combined, with enough leftover for 30 Statues of Liberty and 30 Ford Tauruses.
"apparently building megastructures in KSP can be fun" my last words
I loved how you wrote stuff on the screen instead of using text.
Rip computer you will be missed
Oh my God I am proud when I get a 3,000 pound rocket and you did 1 million
Launching the rockets at 15:58
This little maneuver is gonna cost us 51 years...
Just when you think he couldn’t do anything more in ksp he blows out minds with a massive rocket.
Nice you got NovaSilisko on board! OG dev team represent ;-)
I literally thought "Now that's a name I haven't heard in a long time... a long time"
Imagine being a kerbin resident, and its just after evening and your sitting on your porch, and then you see hundreds of rocket boosters burning up in the atmosphere
Next: 1 Megaton payload
Oh boy, the computer will love this.
In space where you don't need all rockets at full blast, you could fire certain patterns of the rockets to display an image like pixels on a screen xD
use kerbal os to play bad apple lol
Did you do any tests of the fuel lines? It looked quadratic, which suggests that the problem is that every engine was querying every fuel tank it has access to. Fuel lines might behave differently.
Makes you wonder if there is any smart abstraction or caching at all, or whether it queries the entire construction tree until it hits something it cannot pass fuel through every frame*. That would suggest to always put as few parts as possible on engine sections, and separate using parts that cannot transfer fuel.
*Oh god, it gets worse. That stuttering looks like the garbage collector, suggesting that the engines are building a list of tanks to drain from, and discarding it, every frame. If you have 500 engines connected to 2000 fuel tanks that means 1 million allocations per frame from this alone.
I used to think programming games was hard, because I would look at a problem like this and reject the first solution I came up with, based on it being moronic. Now I know better.
If this is true then that is some truly stupid code that could probably be improved by a insane amount through a smartly coded mod
A mod could probably fix this issue quite easily o believe.
I am not good at coding but I believe a good method would be:
1. Generate groups of connected fuel tanks and give them an overarching total fuel volume variable
2. Find all thrusters connected to each tank group and generate total fuel consumption variables for those groups
3. Now each consecutive fuel calculation only has to take the total fuel consumption from the total fuel volume and then update the total fuel volume for each group.
I think this could massively increase the performance as the bulk of the calculations only have to be done once at the very beginning
I still remember finding you at 750 subs, and i have fi ally returned to the game after 5 years to find you are still pushing the limit of this game. bravo
If NASA or SpaceX experience lag, we know what rocket we're going to use.
Don't they sprt of already experience that? In the form of input lag due to the limits of light speed.
My understanding is that modern rockets are controlled by on-board computers. They only needs to send commands in emergencies.
You are by far the greatest KSP engineer of all time. Your cinematography is the best out there too. My jaw never fails to drop watching your videos
hello stratz, what we doing today, chaos, okay then.
Hatsoff to you for being patient and not letting it be a requirement for the viewer.
Omg, he's unleashed ! ^^
This is super surprising! I would've never thought that fuel crossfeed would affect performance so much. It's much simpler computationally compared to physics. Devs definitely should take a look at their code, it could bring us a lot of frames.
It's interesting hearing the wide variety of opinions on this matter. Some people think this makes a lot of sense, whereas others think the crossfeed should be much less significant. It's hard to tell since we don't have access to the game's code, but its great to hear the ideas!
Hey Bro, you are awesome! I'm new at KSP and I tryed to play the training but it didn't helped me. Can you pls do some Tutorials?
Holy shit, I was already in awe watching that behemoth fly up, but when the first stage of boosters peeled off like that and collided .. wow man. Amazing
Do a flip!
Absolutely amazing your discovery with the KSP lag. Surely someone could mod the game somehow to optimise this.
i literally could not be more grateful that you showed stage 2 reentering
This puts everything Bradley Wistance's bill has done to shame. That's so much thrust
patience IS a virtue, this video would not be without it!
also the knowledge of "what" make the game lag can benefit a lot of us on how to optimize, like making our own tanks with more fuel to reduce crossfeed and not just part count!
The civillain kerbals during the last part gonna be looking at the boosters like
'Ahh, lovely meteor shower.'
and ground control is just praying for their lives
This rocket pumping enough greenhouse gases into the atmosphere to single handedly melt the ice caps.
I was looking for KSP benchmarks when I bought my 5900x back in December, coming from an 6700k and I was curious how much of a performance boost it would be. Cool video !
You had no idea how happy i was when i saw the youtube notification that you uploaded
That first stage segments colliding with each other, and spreading in the far back was a beautiful view. 17:46
The scene where it was raining rocket boosters at 14:37.... pure beautiful chaos
your computer must smell really nice after making this video
I am actually super impressed with the fact that the physics engine (minus fuel crossfeeding) scales linearly in processing time with part count. That is a major feat for KSP's engine!
Knowing Stratzenblitz it's entirely possible this entire video was just a setup for the Evergiven joke
I've always wanted to build the largest space station I could in ksp. Thank you for this knowledge, I definitely won't use this for anything nefarious.
My man got carpet bombing to a whole new level at the end... Really impressive, would love to see a full megastructure being brought to orbit next. Keep up the good work 👍
Honestly when he said he wished he was that good at blender i thought he was gonna say "which is why this video is sponsored by skillshare"
The scaling behavior you saw makes a lot of sense. The big trick KSP uses to simplify physics calculations is that the part connections form a tree (i.e. no loops) so you can propagate forces through all the parts in linear time O(N). The fuel consumption network is a general directed graph, so solving the fuel updates has super linear complexity (IIRC it's roughly O(N^3) ) in the number of network nodes. It's possible the devs realized this would be a bottleneck and decided not to do a full fuel consumption solve on every tick, leading to the stuttering you observed.
The more you can disconnect the fuel network, the closer the time complexity will get to being linear. You shouldn't need to remove cross-feed entirely to get a big performance gain.
CPU: "I fear no man, but that... thing."
*"It scares me"*
Now I'm intensely curious as to how other parts and functions affect performance--particularly flags, solar panels, and radiators.
Million ton rocket? Sounds like the most Kerbal rocket ever.
My god...it’s full of boosters.
Lovely piece of work! I never got really deep into KSP because I’m not imaginative enough, so I enjoy watching your videos. Keep it up!
20:15 when he said “like this one” while showing the ever given… laughter
Computer scientist here, the fact that it's linear means the devs put a lot of effort into optimization. It's actually pretty impressive. anything sub n^2 is considered pretty good but we tend to aim for log(n) time complexity or nlog(n)
20:09
Stratzenblitz launching the Ever Given into LEO:
_Our blockade is impenetrable!_
That staging explosion was probably one of the best KSP1 explosions I've ever seen. lol It's also nice to see a 5900x benchmarked like this since I have one.
Thank you! I was super happy with how that came out.
Also Ryzen gang :)
Now KSP is not only getting bug-tested by the community, there's even people doing profiling ;D
Very interesting and in-depth but still entertaining, as always, loved the professionalism at 5:28
I was just watching your 10k special when I realised you reached 100k. Congrats!
Your narration is excellent. I can hear the pure glee.
Is that rocket lifting off or is Kerbin being pushed away?
How much of an effect would exhaust occlusion from that tall rocket have on lag?
It's a good thing you added retro-rockets to the second stage because that is single launch Kessler Syndrome right there
I previous one I launched did not have the retro rockets and is still up there. You can see the debris pass over in a giant cloud over the ksc every orbit
@@Stratzenblitz75 on no 😂