Currently, we send pack an entire table, and send metadata, and GPU data in two separate messages.
In the future, we might have GPU data residing in a set of buffers/ buffer views. These buffers could be,
- Fixed sized?
- Variable sized?
This might require considerable changes in the communication implementation.
Currently, we send pack an entire table, and send metadata, and GPU data in two separate messages.
In the future, we might have GPU data residing in a set of buffers/ buffer views. These buffers could be,
This might require considerable changes in the communication implementation.