by Bitweasil » Wed May 18, 2011 6:44 pm
Current flow:
- Table is generated on GPU
- Table is sorted by final hash value
- Table is written to disk
- Next table is started.
So, yes, there is a bit of a delay from the disk IO.
I could create another thread to handle the writing to disk, and this would help throughput by a few percent. It adds complexity, though.
In general, I wouldn't worry about it. Unless you are concerned about raw peak performance, at which point I'm open to optimizing things a bit.