Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

It seems like the TF32 format is similar to BF16 but with 3 more precision bits (in other words, it is FP32 with 13 low-order bits dropped instead of 16).

Full adders in FP units scale with the square of the mantissa bits, so if the number of mantissa bits stay the same, they can reuse the existing units.

Since Nvidia already has existing FP16 units on die, using those units for TF32 calculations probably doesn't cost too much additional die area.



Consider applying for YC's Summer 2026 batch! Applications are open till May 4

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: