dicehub (@dicehub) header
dicehub (@dicehub) profile picture

Streamlined engineering simulations for all

Mainz, Germany
Joined September 2014
Photo by dicehub (@dicehub)Photo by dicehub (@dicehub)Photo by dicehub (@dicehub)Photo by dicehub (@dicehub)Photo by dicehub (@dicehub)Photo by dicehub (@dicehub)Photo by dicehub (@dicehub)Photo by dicehub (@dicehub)Photo by dicehub (@dicehub)Photo by dicehub (@dicehub)
dicehub retweeted
reran turbine case in @dicehub (4 procs, 440k cells, 3000 iterations) #openfoam "fluid" solver, with cooling flow inside. The hardest part was creating the #paraview orbit, convergence below...
1
4
17
1,159
dicehub retweeted
#blockMesh GUI allows fully structured mesh (890k cells) with continuous boundary layers on the whole vehicle (including wheels) which passes all the quality metrics in #checkMesh and ran in @dicehub 4 procs in 16 minutes - available here github.com/garcfd/blockMesh-…
1
3
34
1,774
dicehub retweeted
#blockMesh GUI now allows structured meshing with o-grid (layermesh) around the whole vehicle. solved in @dicehub #simpleFoam in 16min 4procs. The wheels also have layermesh but the cell quality (wheel/ground interface) needs improvement. github.com/garcfd/blockMesh-…
1
4
17
898
dicehub retweeted
#blockMesh GUI was used to create a fully structured car mesh and solved on @dicehub (4 procs) in 15 mins, good cell quality and good convergence. Available here github.com/garcfd/blockMesh-…
6
46
2,252
dicehub retweeted
interesting to see the new AI developments in @dicehub
1
2
276
That gust that almost steals your scarf next to a skyscraper? It's not bad luck. It's fluid dynamics. The tower pulls fast wind from up high down to the street, and squeezes it around the corners
1
1
1
120
We will be at the @EXPOREAL_Munich next week.
44
Accelerating Multiphysics Simulations in DICEHub with Agentic AI We still believe a clear, reproducible workflow cannot be replaced by an AI chatbot. However, AI assistants are valuable for answering questions, making suggestions, and handling specific coding tasks.
1
3
10
698
dicehub retweeted
Okay, Opus 5.5 is amazing at motion design. This is my first attempt .. not bad.
1
3
71
dicehub retweeted
I was successfully pressured into posting more and showing some of the projects and experiments I am working on. So here we go ... Highly experimental, first iteration: Generative AI studio in @dicehub, a template that generates customized applications based on user input and workflow requirements. The idea: you know what you need for your simulation and you already have your own custom workflow, you just talk to the chat and it generates everything based on already available components (3D visualization, storage, logs, HPC, etc.), attaches a domain-specific model to your problem and gives you a reproducible result that you can share. How is this different from just asking Claude to do it? You have an already prepared library of components and blocks that just work and you do not need to do everything from scratch. Writing the app was never the hard part anyway, the hard part is the solver coupling, the result readers, the visualization, finding the optimal settings, validating the result, etc. So the chat only has to work out your workflow instead of reinventing all of it every time. You run your custom but reproducible workflows, check the results with the domain-specific model and iterate. The goal is that the model and the workflow get better with every run. In the video you can see it generating simple apps, running workflows, etc. but there are still a lot of improvements to make. In my opinion this is probably where we are headed for specialized apps that still require a bit of detailed setup and a straightforward workflow setup, you do not want to prompt yourself to death when working on engineering problems. For most other applications the AI Chat will probably eat everything up, we can already see this happening.
1
1
1
79
dicehub retweeted
Bring high-performance 3D visualization directly to the browser with #VTK-wasm. Join us September 3 to explore standalone web applications, client-side rendering for #trame, and browser-based volume rendering. Register: ow.ly/8B4q50ZCmT0
1
4
254
dicehub retweeted
#UFOCFD (cartesian) 15M cells, ran on @dicehub T4 GPU in 3.5 minutes, Mach 0.7 case, cell size 2cm. Testing out the Roe upwind scheme implementation.
1
15
950
Two design options, same site. One ruins the street corner in wind. Which one? Import both as STL in dicehub, same easy template setup for each, parallel cloud runs, a comfort map for each. Answer while the design can still change. #CFD #PedestrianWindComfort #UrbanDesign
79
STL in. Pedestrian wind comfort map out. The PWC template in dicehub builds the simulation case: mesh, wind data, comfort criteria. The cloud runs it. You review where people can sit, stand, and walk. #CFD #PedestrianWindComfort
72
dicehub retweeted
Typhoon geometry from thingiverse.com/thing:478598… at Mach 2 in #OpenFoam ran on @dicehub 8 procs, 2000 iterations, runtime 27 mins, mesh only 800k cells including layer mesh. Here showing isentropic Mach contours calculated from surface temperature. Ran in #OF12 "fluid" module.
2
3
24
1,104
dicehub retweeted
ran generic supersonic missile at Mach 5 in #openfoam (#rhoSimpleFoam) in OF12 on @dicehub for 1000 its. Proper mesh with 3 layers, convergence bit wobbly after 600 its, maybe because of the freestream outer boundary.
1
4
327
dicehub retweeted
#VAWT ran in #OpenFoam using @dicehub 'case run' feature - took 31 hours to run, then downloaded the results and did all post-processing on laptop using #Paraview.
3
21
1,405
dicehub retweeted
Ran this #VAWT case on @dicehub recently - it took 31 hours, on just 4procs. More of an experimental run to scope out some of the issues/constraints. Model from thingiverse.com/thing:673454…. The rotation is 60rpm, wind speed 5m/sec. Ran as #incompressibleFluid in OF12 using #NCC
1
6
37
3,180
Done with @dicehub, beautifully enhanced wit @GoogleLabs
🤖 Made with AI
3
18
973
dicehub retweeted
#incompressibleVoF catamaran ran on @dicehub this time with CFL reduced to 1 (instead of 10) for better resultion of the timesteps, and more accurate solution.
3
5
309