Main siteMain site  ForumForum  ForumSearch  Private messageEmail contact  RegisterRegister  Log inLog in 
Topic: Does using float vs double float elements effect training
Reply to topic
Author Message
ltrigero



Joined: Sep 29, 2019
Posts: 16

PostPosted: Dec 7, 2019 6:08 PM    Post subject: Does using float vs double float elements effect training

What's the difference between the two and are there circumstances where it would effect training?
jarek



Joined: Oct 22, 2007
Posts: 750

PostPosted: Dec 7, 2019 7:00 PM    Post subject:

Double float is a higher precision, than a single float. But also potentially slower (only matters with large designs) and less tested (because rarely used).

Double float elements could be useful in some special situations, for example with very high precision ADC converters (more than 24 bits). But in most cases single float is perfectly enough.

I would stick with single float designs. And only use double float when really needed (they can be mixed).
ltrigero



Joined: Sep 29, 2019
Posts: 16

PostPosted: Dec 7, 2019 8:02 PM    Post subject:

Okay, and what about scalar and vector elements?
jarek



Joined: Oct 22, 2007
Posts: 750

PostPosted: Dec 7, 2019 8:09 PM    Post subject:

Those are unrelated. Scalar has one value. Vector has many.
Reply to topic