Pending resolution of #46 we have to also find a way to make use of both GPUs on our system. Current single-GPU or no-GPU operation is viable up to a limit. However to make it actually workable across the board requires us to be able to dynamically distribute the calls. This means finding a good way to make the code recognize the presence of multiple GPUs; track the load per-GPU; and distribute calls as needed;.
All of this can be done but needs some better load management using standard engineering practice.
Pending resolution of #46 we have to also find a way to make use of both GPUs on our system. Current single-GPU or no-GPU operation is viable up to a limit. However to make it actually workable across the board requires us to be able to dynamically distribute the calls. This means finding a good way to make the code recognize the presence of multiple GPUs; track the load per-GPU; and distribute calls as needed;.
All of this can be done but needs some better load management using standard engineering practice.