You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Parallelize over multiple accelerators at the same time (i.e. use both CPU and GPU or multiple GPUs)
Investigate the use of a property to skip suffix calculation ~50% of the time.
Note: Property is untested, just mathed out in a document. If it works it can tell us whether continuing with the calculations might equal the final hash or definitely won't equal the final hash.
The text was updated successfully, but these errors were encountered:
I considered storing name batches as 3D arrays in the hope of better performance but I did not follow up on this idea because my 3-days dirty code rush was over 😅.
It would require to rewrite the HashStringsBatchOptimized kernel and I'm not sure it would enhance performance.
I also think time is lost in the batch generation process itself so I had plans to build name batches with a kernel.
I guess having the batches generated asynchronously would be another option.
Just communicating my current plans.
Note: Property is untested, just mathed out in a document. If it works it can tell us whether continuing with the calculations might equal the final hash or definitely won't equal the final hash.
The text was updated successfully, but these errors were encountered: