CPU and GPU computing speed
For the calculation of a small model (about 10,000 particles, some of which are used to represent spherical particles, using the Chrono library), will the following situation occur: For the same model file, the GPU calculation is slower than the CPU calculation?Is this entirely possible? Mainly look at the model (performance) parameters of CPU and GPU?
In addition, I still have a problem, I don't know how to solve it, who can give me some suggestions or solutions?
Comments
It is difficult to answer your questions when you end up asking the same questions in multiple threads. Please try to stick to one thread and bump it with new information.
To answer your actual question in this post, then when you have such a low number of particles the speed of CPU vs GPU is not really relevant. Can a CPU be faster? Maybe, but it is not really relevant, since both will be very fast at low particle numbers.
The moment you start getting a bit higher, 100k-1000k you would see GPU starting to take off and everything above 1 million GPU would win easily. This is my personal experience feel free to test it on a dam break case.
You can have a case in which the fluid flow is very lenient and requires no effort to calculate, i.e. no sharp interfaces, no violent flow etc - and in this kind of model you might have some Chrono capability included which is very cumbersome, imagine friction between a lot of difficult shapes. In such cases a better CPU will help increase performance since Chrono is only CPU bound.
In general SPH is much more performant on GPU. If you use Chrono in difficult cases, your bound on speed can be CPU. In such cases stronger CPU makes more sense than a strong GPU. In ideal world one would have very strong GPU and very strong CPU - not always possible.
Kind regards
Thanks very much. In my case with a small number of particles, the simulation is 0.55 seconds, the minimum time step is -7 power, and the calculation takes about 24 hours, which is very surprising.