Parallel GPU Processing

edited January 21 in DualSPHysics v4.2
I am trying to run a simulation of 15 million particles using Amazon GPU instances. There are system with multiple GPU's which costs high and it came to my attention that DualSPH only uses one GPU at a time. If that's true then renting system with multiple GPU would be a waste of money. Does multiple GPU (4* 16 Gb) will be faster than (1*16 Gb)?

Comments

  • Hi,

    Renting multiple GPU solution seems a waste of money in my opinion, since DSPH can only use one.
    When the code will be updated to multi-GPU, given that they are on the same machine (the code will not permit cloud GPU computing, only multi GPU on one local machine), 4 GPUs will be faster than than just one that is the whole point of multi GPU.
    But in the mean time, it won't make a difference.
    Nice day
  • Is that complicate to implement the multi-GPU function? If It is not complicate, may be we can start to implement it and share the workload with DualSPH team.
  • I don't know ! However, from my experience, it shouldn't be that trivial if you want it to be efficient (managing splitting of calculations, memory usage and sharing)...
    From what the DSPH team told me, they are not a private club ! So if you are willing and competent enough, you can join them and improve the code... I suggest you get in touch with them. Multi-GPU code is definitely part of their plan http://dual.sphysics.org/index.php/features/
Sign In or Register to comment.

Howdy, Stranger!

It looks like you're new here. If you want to get involved, click one of these buttons!