Thanks for the question! Indeed, it should be possible to implement this on the GPU, as it uses an explicit stepping method with local stencils for the WENO spatial discretisation, so is very amenable to parallelisation at the "cell" level. If the video frames were also post-processed on the GPU as they were calculated, this would minimise the need to send very large data files back to the host (which might otherwise add significant I/O that would negate the parallelised speed-up). The implementation above is only single-threaded, there has been no addition of OpenMP or even MPI for distributed memory parallelism as of yet. An alternative to the GPU would be to use OpenMP, which would likely be a quick way to gain parallelism on the same node but it would require a CPU with significant memory bandwidth (e.g. AMD Epycs with many memory channels) in order to be able to utilise the multi-core nature of those processors.
@@EdgeOfInfinity42 makes sense - i would be very interested in seeing this simulation system at interactive frame rates. I am not sure if this is something you are comfortable with, but if you are willing to share the source code I would like to try and port it to shaders for execution on the GPU
Thanks for the question! This one actually uses a custom piece of software written in FORTRAN that solves the 2D Compressible Euler Equations using a numerical method known as WENO. There's a lot more detail on how it works in the video description, if you would like to know more.
can you implement this on the gpu for some faster render times? or is there cpu specific logic that would be difficult to port?
Thanks for the question! Indeed, it should be possible to implement this on the GPU, as it uses an explicit stepping method with local stencils for the WENO spatial discretisation, so is very amenable to parallelisation at the "cell" level. If the video frames were also post-processed on the GPU as they were calculated, this would minimise the need to send very large data files back to the host (which might otherwise add significant I/O that would negate the parallelised speed-up).
The implementation above is only single-threaded, there has been no addition of OpenMP or even MPI for distributed memory parallelism as of yet. An alternative to the GPU would be to use OpenMP, which would likely be a quick way to gain parallelism on the same node but it would require a CPU with significant memory bandwidth (e.g. AMD Epycs with many memory channels) in order to be able to utilise the multi-core nature of those processors.
@@EdgeOfInfinity42 makes sense - i would be very interested in seeing this simulation system at interactive frame rates. I am not sure if this is something you are comfortable with, but if you are willing to share the source code I would like to try and port it to shaders for execution on the GPU
Do you use Fractals or Imaginary Numbers for this?
Thanks for the question! This one actually uses a custom piece of software written in FORTRAN that solves the 2D Compressible Euler Equations using a numerical method known as WENO. There's a lot more detail on how it works in the video description, if you would like to know more.
@@EdgeOfInfinity42 🤗