Interesting, there appears to be very little benefit to use #halfprecision #floats other than a #headache https://futhark-lang.org/blog/2021-08-05..
#float 16 coming the the #java #jdk #jre 20: https://download.java.net/java/early_acc.. Some people have already mentioned that this will likely be dead in the water, as it casts #floats to #shorts , which will mean they could be #misused , rather than creating a #dedicate #type for them.
Nice discussion on whether #halfprecision #floats are worth the effort or not, specifically using the #gpu https://futhark-lang.org/blog/2021-08-05.. Turns out the #speedup really isn't so much. I suspect it *may* be better when you have large vectors, you can literally keep twice as much in #cache and larger models in #ram ... But still, it's worth checking whether you actually get the expected speed-up or not.