Page 1 of 1

Cuda pathfinder algorithm on server GPU

Posted: Tue Feb 06, 2024 5:38 am
by Melanie Jansen
On multiplayer servers i assume the final motion is calculated there and then shared to the online players so the motion can take place on the client render side.
There and also local it would be great if parts of the code like the pathfinder algorithms that maybe could processed in parallel for every unit doing this on the GPU, with code like the Cuda engine embedded. Also these routines can be optimized because they are slow with large e/o complex maps e/o lots of units e/o players. I don't know (hint?) how the code on the server or local is calculating the pathfinder but with lots of units on large maps it gets quite slow. Like 'unit lag'. Its not a bandwidth problem.
Also if all the clients are in sync with each other and the server actually the only thing you need to know on the client side is who is taking what action at what time (tick). The client then only (...) have to figure out what the situation will be according to that click on that tick synchronized with the others. The only thing the server has to receive and send is who is clicking what and when.
Combining these two features could speed up large heavy games enormous. Like a bit of a GPU card can render 100.000 objects at 60 Hz if you like, that is not the problem. You don't render every tile and unit. You render the background and then the units.
The trick is then to keep as much memory plus the according calculations on the GPU and minimize the load between the CPU and GPU, leaving in the end the CPU just the driver behind the game.

Salut.

Re: Cuda pathfinder algorithm on server GPU

Posted: Wed Feb 07, 2024 11:55 am
by Punsho
OpenRA runs a lockstep simulation, it means that the server essentially exists just for passing orders between players, which is not really resource intensive