Visualizing 144 TeraByte of CFD data for fun

FluidX3D source code: github.com/ProjectPhysX/FluidX3D
The 20 seconds of video show the car driving at 100km/h for 1 second.
Mattia Binotto's Ferrari SF71H in #CFD. In this 10 billion voxel #FluidX3D simulation you see the wild aerodynamic optimization for a very successful F1 car. #OpenCL compute (2152×4304×1076 resolution grid, 217k time steps) plus rendering 3x 20s 4K60 video took 14 hours. Shown is velocity-colored Q-criterion isosurfaces with marching-cubes. Reynolds number is 3.75 Million with Smagorinsky-Lilly subgrid model.
How is it possible to squeeze 10 billion grid points in only 512GB VRAM?
I'm using two techniques here, which together form the holy grail of lattice Boltzmann, cutting memory demand down to only 55 Bytes/node for D3Q19 LBM, or 1/3 of conventional codes:
1. In-place streaming with Esoteric-Pull. This almost cuts memory demand in half and slightly increases performance due to implicit bounce-back boundaries.
Paper: doi.org/10.3390/computation10...
2. Decoupled arithmetic precision (FP32) and memory precision (FP16): all arithmetic is done in FP32, but LBM density distribution functions in memory are compressed to FP16. This almost cuts memory demand in half and almost doubles performance, without impacting overall accuracy for most setups.
Paper: www.researchgate.net/publicat...
Graphics are done directly in FluidX3D with OpenCL, with the raw simulation data already residing in ultra-fast video memory. 1 frame of the velocity field is 120GB, 1201 frames are generated, which would be 144TB. No volumetric data ever has to be copied to the CPU or hard drive, but only rendered 4K frames (33MB) instead. Once on the CPU side, a copy of the frame is made in memory and a thread is detached to handle the slow .png compression, all while the simulation is already continuing. At any time, about 16 frames are compressed in parallel on 16 CPU cores, while the simulation is running on GPU.
Paper: www.researchgate.net/publicat...
Timestamps:
0:00 side view
0:20 front view
0:40 top view
Thanks to the people at Jülich Supercomputing Centre for letting me test their hardware!
The 3D model is from Thingiverse: www.thingiverse.com/thing:299...
#CFD #GPU #FluidX3D #OpenCL

[@ProjectPhysX]

There was a debate about how the airflow is affected if the wheels rotate (here they are not rotating). I did the simulation again, with rotating wheels: kzbin.info/www/bejne/q3i7pKymgc2pf5Y Commercial CFD software would take months for a simulation this detailed. I did it in 14 hours on 8 GPUs, including rendering. I have put the source code of FluidX3D on GitHub, for free: github.com/ProjectPhysX/FluidX3D

[@Blackoutfor10days]

I want to see this with all brands of cars

[@SHAURYA181]

I guess it would have taken weeks for this kind of simulation on Ansys.

[@ProjectPhysX]

@@SHAURYA181 more like 3-6 months :)

[@rodionstepanov2034]

on 8 GPUs ... cool! Does it mean that FluidX3D can be run on multi-GPU?

[@ProjectPhysX]

@@rodionstepanov2034 not yet, but soon. I'm still working on multi-GPU support, here it's the current prototype of it. I will release the update on GitHub once it's ready!

[@nonoaidnono]

I can only imagine how much better this looks originally, rather than KZbin struggling to compress it

[@fakestiv]

As a bonus you have an excellent youtube stress-test

[@jwalster9412]

What compression? I see no KZbin compression.

[@jorgereyes8071]

@@jwalster9412 wdym, it looks like shit

[@jwalster9412]

@@jorgereyes8071 nooo. Definitely not.

[@jorgereyes8071]

@@jwalster9412 maybe you have premium idk

[@jwalster9412]

@@jorgereyes8071 _(sarcasm)_

[@erhanakin]

Although Im not a CFD guy, I have to say that the code you shared is so much valuable to people who deal with fluid simulation. Im speechless. Congrats man for writing the code and sharing it free publicly.

[@HyperMario64]

At first I thought it was some mechanical engineer posting some render from a proprietary software running on a mega render farm. Damn it's all your own thing. And not even for a thesis or so. Hats off.

[@ProjectPhysX]

The software is sort of a byproduct of my Masters and PhD. At the beginning of my PhD I was faced with the impossible task to simulate thousands of raindrop impacts for statistical analysis. With any existing commercial software that would be >3 years compute, and the PhD is intended for only 3 years. So writing my own, much faster software was the only way, and it worked, the raindrop simulations took just a week. I then thought, since I had created such a powerful tool for CFD analysis, why not publish it so others can use it for free! And here we are today :)

[@jwonz2054]

@@ProjectPhysX I love you.

[@rishabhlakhara9583]

@@ProjectPhysX damn man you're awesome

[@avithedev]

@@ProjectPhysX King 👑

[@spectralvoodoo5233]

@@ProjectPhysX How did you manage to make something better than all of those commercial products made by million dollar companies?

[@catalandesdepolonia]

Wonderful job, thanks for sharing! It would be super interesting to see Mercedes’ no sidepod design to compare

[@user-hw4st7br9a]

Finally someone actually uploaded a aerodynamic simulation in F1 cars. Thank you so much🙏

[@Razzbow]

The flow conditioning on the front wing has so little overspill its amazing, the brake ducts are also so heavily loaded so i guess thats how they are so efficient!

[@TheMilitaryONE]

Oh man, another classic afternoon having fun with visualizing 144 TeraByte of CFD data. Always having the time of my life!

[@vasve2071]

Piece of art ! I'm still in shock how good this is ...

[@ScreamingElectron]

You'd think the flow would be much more laminar. Very interesting.

[@te0pol159]

Yeah, air is surprisingly messy.

[@sirsanti8408]

@@te0pol159 well, air flowing over stuff at least

[@jacobkeltz3584]

Interestingly enough, you'd think that they'd try to avoid turbulence with f1 type cars, but that's basically impossible with the open wheel design. Rather, they don't look so much at reducing the drag (most f1 cars have high drag coefficients greater than 1 even) but instead choose to maximize the downforce they get for massive traction.

[@laty9f1]

This is because this is a previous car, not 2022… 2022 cars direct the flow (most of it) up and over the car behind

[@joshuapeligrino]

​​@@laty9f1 it's actually a 2018 car because of the front wing i think?

[@ian.backstrom]

So rad, thanks for posting this work!

[@tillenleskolesnikpikl3114]

you have all my respect, this is crazy 🙌

[@davidhagemann7037]

This was pretty impressive to watch!

[@xonio2881]

keep up the good work man! hope you get your hands on the solid bodies plastic flow one day

[@CreamAle]

This is beautiful. Simply beautiful.

[@bigiron4018]

I have been getting familiar with your software, working up in model refinement and resolution, end-goal focused on simulation of my own racecar. and although i only have a single A4000, the performance is mind boggling. what's the catch? How were these gigantic optimization leaps made? I used solidworks flow simulation prior (i know, not terribly great compute wise) but still, I just dont get where all this performance is coming from!

[@ProjectPhysX]

Have fun with my software! :) I wrote the it for GPU parallelization from the very beginning. Various performance optimizations came along (faster swap algorithm, better coalesced memory access, array of structures memory layout, no PCIe transfer etc. Every added extension comes with a preprocessor switch, these #define statements, that turn off the entire block of code at compile time if the extension is not in use. So the core solver always stayed at peak memory efficiency. Frankly I have no idea why commercial CFD software is so damn slow despite being so expensive :D

[@w花b]

@@ProjectPhysX I don't understand anything you've said except the fact that maybe you're doing It in C/C++ but it seems like fun to program.

[@ProjectPhysX]

@@w花b the core part of the software that runs on the GPUs is written in OpenCL C. It's a lot of fun, I love that language!

[@williamvanderscheer4327]

@@ProjectPhysX I understood everything you said, and have had my fair share of OpenCL. Its epic porting something to the GPU, gives me UNLIMITED POWER vibes. As for commercial software being slow: They are typically not developed from the set go for performance, and companies tend rarely find optimization work a financial priority, over new features. Purpose build GPU implementations will kick anythings ass for sure, great work!

[@ProjectPhysX]

@@williamvanderscheer4327 I've heard it's a similar story with weather forecast models. They are old Fortran code, rigorously validated, but slow and stuck with CPU hardware. There is noone with the skills/time to translate them properly for modern GPU hardware. Porting always is troublesome and error-prone, to the point where a complete re-write makes more sense. But then again these codes are gigantic. Having OpenCL code from the ground up is a great advantage, as vectorized code is compatible with all hardware out-of-the-box.

[@houstonhelicoptertours1006]

Good stuff. We use similar methods to evaluate our airfoil and propeller designs.

[@DaedalusLegacy]

This is incredible, So incredible 4k renders the player inoperable.

[@Whitetigerclaw17]

i could watch it all day in a loop. :D

[@Weeaboo_AF]

Wow this is sick! I knew dirty air was a problem but jeez, is it any wonder Previous cars were stuck at 1.2 seconds behind

[@cvpuga]

one of the coolest things ive seen this week

[@XKontoXxxxx]

Das ist ja der absolute Wahnsinn 😍

[@ltitus8900]

This is AWESOME! Thank you very much!

[@chriswei6109]

Ich drücke die Daumen für dich, so...... geileeeeee Berechnung. Respekt!!!

[@rishabhlakhara9583]

Posting this to help with the algorithm good stuff

[@hayatojp1249]

it is so beautiful

[@mcinnes6175]

youtube bitrate killed this, looks pretty tho

[@sergiosanjose1131]

Nice work!!!🙃

[@KOENEO_]

this is the most realistic CGI fire I've ever seen and its not even CGI fire

[@altruistx]

Gorgeous

[@TranTran-ro8zb]

Oh god its resolution is higher than all of my game had been played.Well done !

[@imsleepy620]

bruh that's sick

[@danigarcia2294]

Finally, race car aerodynamic

[@user-yv8re2cz9o]

Bro’s PC enchanted with Unbreaking III

[@gaborvidovics6943]

Amazing!

[@fruity8562]

The sequel we all wanted but didn't ask for.

[@lightex174]

That's kinda fun!

[@Enchantaire]

The YT algorithm is sweating bullets haha Looks cool, and trippy

[@drifter4training]

that is meticulously trippy... 😯😮

[@StuntcatTV]

That was nice, actually. Thanks for doing that

[@aviksaha2746]

So much turbulence for such a enormous funded sport, None the less, your work is magnificent.

[@aviksaha2746]

@@yubnub3000 you are acting smart on the comment of an thermal engg who had worked with aerodynamic of F1 car design. The same thing but in Ansys. None the less, F1 cars do try to minimize drag (search drs), But if you beleive you know everything, I can't argue much.

[@Draxis32]

That is some "dirty air" you've got there!

[@andressamaniego]

Oh yes, I love me some neon data flames

[@chinogitarista7379]

I got here from the cow aerodynamic vid. I am still confused. I LOVE IT.

[@SuperMaDBrothers]

hey. that is pretty fun 😮

[@getsideways7257]

This simulation looks extremely detailed. Even though the amount of voxels is the same, this impresses me way more than the Shuttle. Guess it's all down to the overly complex aerodynamics of the modern F1 cars... By the way, it should be feasible to make a plane in those dimensions, right? Especially considering there is no need for such wheels... Only the propulsor(s) is going to "stick out".

[@ProjectPhysX]

The turbulence always looks super fancy in Q-criterion isosurface visualization, especially with a more sophisticated model geometry! Airplanes are also feasible!

[@getsideways7257]

@@ProjectPhysX Sure it does, but this particular video outdid everything else in my opinion - the amount of minuscule vortices is staggering, if not mindboggling. I simply cannot recall a single other video with this amount of eddy shedding complexity. As for airplanes I didn't mean in general, but rather - creating a plane by utilizing the same aerodynamics design process they use for F1. To simplify, we take an F1 car, drop the wheels, turn it upside down and strap a propeller onto it in order to make the end design about as narrow as that F1 car is. What do you think?

[@ProjectPhysX]

@@getsideways7257 there is similar lifting body aircraft designs, I've seen some as RC planes. Interesting concept, but probably not as good as traditional designs in flight characteristics.

[@getsideways7257]

@@ProjectPhysX That's interesting indeed. It's just that looking at the recent F1 cars was making me wondering whether the aerospace industry (at least when it comes to General Aviation) is falling way behind. Just compare the '80s F1 cars aerodynamic surfaces (that basically looked like the same old wings flipped upside-down) to the "organic" design of the recent ones, then look at, say, Cessna 150 and Cessna 350 - sure, the latter does look much slicker, but the overall difference is barely worth mentioning. So, I can't help but ponder if all this is because there is a lot of money and tech behind Formula 1 - also competition - or are F1 cars THAT fundamentally different from planes...

[@daylight31415]

@@getsideways7257 F1 cars are that fundamentally different. F1 cars are made to be the fastest around the track. Which means, aerodynamic-wise, that they need high downforce while keeping the drag to a minimum. And they are very restricted due to regulations. Planes on the other hand have different criteria. They only need high lift in two scenarios: The take-off and the landing. Flaps and slats are the elements that come into play in these cases, they increase the uplift and also make sure the plane does not stall. The aerodynamics of a plane are optimized for minimal drag and therefore fuel efficiency during cruise flight, not creating maximal up-or downforce. Therefore the lines are much sleeker.

[@Mart-E12]

That is insane

[@N8ternatenate]

So much fun

[@NickMaovich]

this is awesome

[@gabrielarcoverde9098]

masterpiece

[@Dynamitethedrummer]

ty for the data!!!!!!!!!!!!!!!

[@isaeljosafatmartinezclavel5283]

funny and terrifying at the same tiem

[@DiegoAndrade]

Bravo !

[@SirBapkins]

This is amazing! Care to try this with a model of the Aptera? I'm super curious!

[@FredoCorleone]

Cool! :D

[@fusta6208]

I finally get to see how Adrian Newey sees.

[@S-Freeze]

thats an omega fuck ton of data, and there's things so much more gargantuan

[@lau4893]

As a mechanical engineering student i would absolutely love to learn how to use the software! The fact that it can run on regular modern hardware makes this so impressive as well!

[@ProjectPhysX]

I wrote this software specifically make fast, physically accurate CFD accessible to everyone. To eliminate all (the high entry barriers that come with ;) commercial CFD software: No expensive Quadro GPUs required, any old gaming card from any vendor does the job. Not weeks/months of compute time, just a matter of minutes to hours. No super overpriced software license, free to use in public research/education/hobby. No secret blackbox, the entire source code is public. Because when I started my research, I didn't have any of these resources either. Not the money for hardware and license fees, not the time to wait for inefficient compute, not the will to debug a blackbox where I have no control over code quality standards. Have fun with FluidX3D!

[@humanbean3]

@@ProjectPhysX chad

[@neilmcmahon]

Always wondered how they create the curtain around the exposed face of the front wheel. In this sim, I can see the vanes trying to distribute around the wheel face but still seeing lots of 'wash' ?

[@varunahlawat9013]

No way! I've seen something like this for the first time in my life.

[@PplsChampion]

yeeeaaaa f1 car! have you rendered the "beach" example from setup.cpp on this channel yet?

[@ProjectPhysX]

Not yet! There is several setups I use for testing but still have not rendered to a video :D

[@BookOfMorman]

YT algo brought me here to blow my mind! Great video!

[@ProjectPhysX]

Welcome to my channel! I know sometimes the all-mighty algorithm™ goes completely rogue and starts recommending actually cool videos :D

[@BookOfMorman]

@@ProjectPhysX forreal! haha keep up the great work!

[@PunmasterSTP]

TeraByte? More like "Terrific; out of sight!" 👍

[@wanderingbufoon]

it's like lighting it on fire and driving it

[@ProjectPhysX]

Yeah the Q-criterion visualization for vorticity looks kinda similar to fire :D

[@shreyasashok1305]

Very cool stuff! Can you say a bit about the boundary conditions you used? Is the car using bounce-back BC or something more detailed? How about inlet and outlet? Also, is there any local grid refinement?

[@ProjectPhysX]

Standard bounce-back for the car, moving bounce-back for the road surface. Inlet/outlet is equilibrium boundaries. No local grid refinement, just brutal resolution.

[@eldon7182]

Amazing my eyes hurt

[@domenicozaza192]

Just.. wow

[@EinLauch]

nice

[@massimogalliano8337]

good video

[@Italiandogs]

I could hold almost 6 seconds of that 20 second clip before id run out of space

[@Hermis14]

Could you upload a video about the visualized surface pressure distribution?

[@ProjectPhysX]

Would have to redo the simulation then, as storing such a gigantic dataset is not feasible.

[@Enigma_Flow]

Thank you so much for this amazing work. Seriously hands down! As an aerodynamic design and CFD engineer (not developer though) who have been using FVM for some good time through OpenFoam and StarCCM+, I had some questions 1- Are lattice placed initially adaptive with respect to boundary condition types like walls or inlet or outlet? I mean very high density closer to the wall with some growth ratios. 2- I believe we resolve the boundary layer physically without any wall treatment modeling, correct? 3- Does your code support the motion of multi bodies like rotation or translation of several propellers? 4- Can it be integrated with Aeroaciustic models like FW-H ? 5- Can your code handle shock and super sonic flows? My apologies if these questions look rudimentary cause I am not very familiar with LBM !

[@ProjectPhysX]

1) The lattice is the same size everywhere, there is no grid refinement. 2) Yes, with large enough resolution this can resolve the boundary layer directly. 3) Yes, with force/torque summation and re-voxelization. It's more difficult to program such setups but already possible. The voxelization is rather slow however, and I'm still looking for a better solution. 4) I'm not familiar with FW-H in particular but aeroacoustics should be possible. 5) Shock waves yes, supersonic flow no.

[@beamboy07]

In the future video games will have this in real time in a personal computer

[@ProjectPhysX]

I hope so! With an RTX 4090 you can get more computing power on your desk today than all of humanity had back in the year 2000. I can't even imagine what gaming PCs will be like in another 20 years from now.

[@beamboy07]

@@ProjectPhysX yeah people think thr future of pc gaming would be cloud based but its definitely not

[@FelixFrankie]

This looks phenomenal! Excellent work. I'm familiar with the LBM, but I have always struggled with getting geometries into my lattice domain. I don't suppose you could point me in the right direction?

[@ProjectPhysX]

The algorithm you're looking for is called mesh voxelization. Once you know which grid points are solid geometry, the rest is standard LBM bounce-back boundaries. I have 2 variants of voxelization implemented currently: a) "3D-rasterize" the triangles of the mesh in the grid, or b) cast a ray from each voxel in a random direction and see how many times it intersects with the triangle mesh. Odd number of intersections means the grid point is inside the mesh. However a) only voxelizes the hull, and the sloshing in the geometry can cause instability, and b) is painfully slow, even when parallelized on the GPU. I'm still looking for a better solution myself.

[@FelixFrankie]

@@ProjectPhysX Thank you for the quick response! So, if I understand correctly, with a), the interior of the solid geometry is still fluid, and is therefore updated with the fluid collision and streaming steps?

[@ProjectPhysX]

@@FelixFrankie yes exactly. Since it's entirely enclosed it doesn't matter for the flow on the outside, but any hole in the geometry can let sloshing instabilities escape.

[@FelixFrankie]

Thanks again. Looking forward to seeing your future videos

[@ProjectPhysX]

@@shreyasashok1305 that looks crazy fast. I will have a look in the code and papers. Thank you for sharing!

[@JabbaKo89]

Das ist wirklich Geil! Ich kenne mich nur mit RANS/URANS Modellen aus. Gibt es bei LBM genauso die Möglichkeit rotierende Reifen. einzufügen und eine gleitende Wand (Boden)? Also rein Theoretisch, auch wenn es bisher nicht im Code implementiert ist?

[@ProjectPhysX]

Ja gibt es! Der Boden (lila) ist hier bereits als bewegte Wand simuliert; die anderen Wände in der Box (grün) sind nicht-reflektierende equilibrium boundaries mit vorgegebener Geschwindigkeit. Die Reifen gingen genauso als bewegte Oberflächen, sind aber der Einfachheit halber hier im Modell statisch. Das ist alles schon implementiert ;)

[@GamalKevin]

*_"...only 512 GB VRAM..."_* Me with 1GB Shared VRAM APU: _"Huh. Rookie numbers"_

[@TheLightningStalker]

It looks beautiful. I wonder what the accuracy of such simulations is.

[@ProjectPhysX]

Unfortunately I don't have experimental data for the car to compare with. But the software is as phsically accurate as it gets with the state-of-the-art LBM model.

[@potatoesarelyfe4706]

Brilliant job! My university club is looking at doing CFDs and we will probably end up using your solution if that’s okay with you.

[@ProjectPhysX]

I specifically targeted my code to be used in research/education. It's completely free of charge to use there :) Have fun with the code!

[@Mutrax4706]

the hardware needed for this

[@ProjectPhysX]

It's only 1 server node with 8x AMD Instinct MI200 GPUs with 64GB VRAM each.

[@Mutrax4706]

@@ProjectPhysX i mean it's only so expensive that nasa is jealous (jk idk how good theirs is)

[@maxdenbreejen9844]

This is what Adrian Newey sees every time he looks at a car

[@aidanf2610]

Wow very cool, and I can’t even begin to understand the models you used. What were the wheel BC’s modeled as? Can you assign them angular velocities with your LBM solver?

[@ProjectPhysX]

In this simulation the wheels are stationary and only the road surface is moving bounce-back boundary. I could give the wheels angular velocity as well, it's already implemented. But I didn't have the model with separate wheels on hand.

[@ProjectPhysX]

I just did the simulation again, with rotating wheels: kzbin.info/www/bejne/q3i7pKymgc2pf5Y

[@rheinferdous3821]

i am very new to CFD but this looks extremely awesome! besides how beautiful this looks, are there any industrial/practical benefits for computing aerodynamics in such high detail?

[@ProjectPhysX]

Yes, there is more such applications and demand than I can even imagine. Every modern airplane, every car, ship propellers, engine parts, even the fans of graphics card coolers. All is optimized in CFD nowadays, it's cheaper than building real world models/prototypes.

[@getsideways7257]

I believe it's not just eye-candy. Someone told me once that using ANSYS (for example) you are still having about 20% to 30% difference between the model and the real performance. That's a lot of difference, in my opinion.

[@ProjectPhysX]

@@getsideways7257 depends on the level of detail im simulation. Ansys cannot do nearly as high resolutions. Depending on the simulated system and quantitiy you compare, 20-30% difference can be both a good or bad prediction. Good prediction especially when no fitting parameters are used (1:1 experimental parameters).

[@getsideways7257]

@@ProjectPhysX Do you find OpenFOAM being more promising than the commercial suites? Personally, I believe that simulation and closed source is as "good" as non-disclosed algorithm cryptography - if not worse. Potentially life-threatening even, because if the code contains a flawed implementation of the algorithm or an outright wrong formula - in real world the crew and the passengers could pay a dear price for the mistakes in the code.

[@ProjectPhysX]

@@getsideways7257 yes. OpenFOAM is an excellent software, it's free and open source. I would prefer it over commercial blackboxes, because a) the licence is not $120k and b) you can look in the source code and check yourself that there are no errors. Keep in mind, they all just cook with water. At the beginning of my PhD (planned 3 year duration) I was faced with a computational problem (simulating thousands of raindrop impacts) that would take >3 years compute time with any existing commercial or open-source software like OpenFOAM. In the end I wrote FluidX3D and computed the raindrops in a week. This is also why I decided to make my source code available to the public now.

[@floodtheinbox]

KZbin compression go BBRRRRRRRR

[@stilllobster]

what GPU's did you use

[@ProjectPhysX]

4x AMD Instinct MI250, that's dual-GPUs with 2x MI200 64GB each. In total 512GB VRAM across 8 GPUs.

[@alexprost7505]

@@ProjectPhysX еба%$&. смоделируй ка столкновение галактик)

[@vaibhavG69]

potato lol

[@alanwatts8239]

@@ProjectPhysX Pffft, rookie numbers. You just have 508 more GB than me.

[@straeigel4525]

@@ProjectPhysX Will Minecraft start?

[@jorbedo]

Dr. Lehman, it is possible to share the final rendered frames to create a high quality MP4 clip on our side? YT compression sucks. Thanks

[@ProjectPhysX]

Not sure if there is a platform that supports such high-quality video. The 4K original is ~750 MBit/s, enough to crash my computer when playing from the local hard drive.

[@mojojoji5493]

Imagine racing games of the future using this technology under the hood

[@romaliop]

I don't think you'd gain much at all from that gameplay-wise compared to prebaked models, unless a part of the game is designing your own vehicle.

[@fedupadhesive4885]

What in the aerodynamics-

[@SonnyLikesScience]

Will you upload a tutorial on how to use your code? What you have been showing us lately is AMAZING

[@ProjectPhysX]

I will. Be patient, I have a lot going on right now, hope it will get better at the beginning of next year.

[@SonnyLikesScience]

@@ProjectPhysX YES ! thank you

[@Michallote]

@@ProjectPhysX don't underestimate the power of communities, we can start to make small progress with you only occasionally checking in and answering questions! Small written guides and so forth

[@ProjectPhysX]

@@Michallote I would absolutely love that!

[@Cyan1902]

I smell something burning over there...

[@elietheprof5678]

That's a lot of Navier Stokes

[@dar1e08]

Really interesting project. Have you done / or are planning on any validatory / correlation work with experimental test data?

[@ProjectPhysX]

I did rigorous validation on the software for my academic papers already, with analytic solutions, experiments and other codes. See here: www.researchgate.net/publication/362275548_Accuracy_and_performance_of_the_lattice_Boltzmann_method_with_64-bit_32-bit_and_customized_16-bit_number_formats doi.org/10.1186/s43591-021-00018-8 doi.org/10.15495/EPub_UBT_00005400

[@chrisbean4610]

The cow seemed more wind resistant

[@marcomajonchi9741]

Everytime the video change prospective is: from 0 to 100 in a blink

[@EveryLittleBitCounts]

This is exactly how it looks when I close my eyes and imagine something

[@w26895]

kewl

[@ankurage]

Even KZbin can barely manage to show all the eddies

[@Jeradactile]

Curious if they do these wind tunnel tests with the tires moving at speed or not. Would the magnus effect play a role or not as much since the tires are making contact with the ground. Cool video!!

[@Jeradactile]

Aaaand just read the pinned comment ha ha.

[@ProjectPhysX]

They actually do have a highspeed treadmill in the wind tunnel. It's quite mind-blowing, I wasn't expecting that either.

[@vinceking7878]

Is all the fuzzy green rough air vortexes? What would a GT car look like in this?

[@ProjectPhysX]

Yes. KZbin video compression unfortunately turns it all into fuss at 1080p. In 4K you can see the fuss is made up of myriads of tiny vortices. GT car is on the list!

[@dnomyarg32]

Nice work! Have you been able to validate your results with wind tunnel and/or telemetry data?

[@ProjectPhysX]

The software is validated on lots of test cases already, see here: www.researchgate.net/publication/362275548_Accuracy_and_performance_of_the_lattice_Boltzmann_method_with_64-bit_32-bit_and_customized_16-bit_number_formats But I'll have to do some windtunnel standard test cases still. Will do that next year when I have more time.