Difference between revisions of "Type/float16/en"
Jump to navigation
Jump to search
| Line 5: | Line 5: | ||
The UAVCAN library has functions to convert float32 and float16. | The UAVCAN library has functions to convert float32 and float16. | ||
After searching, it seems that NVIDIA GPU directly supports it as hardware. | |||
Please see the next page: | Please see the next page: | ||
https://en.wikipedia.org/wiki/Half-precision_floating-point_format | https://en.wikipedia.org/wiki/Half-precision_floating-point_format | ||
Latest revision as of 02:10, 18 September 2021
float16 is a 16-bit float type that conforms to the IEE 754 standard.
Common computer languages are not supported. So, you need to find a SW library or build your own. The UAVCAN library has functions to convert float32 and float16.
After searching, it seems that NVIDIA GPU directly supports it as hardware.
Please see the next page: https://en.wikipedia.org/wiki/Half-precision_floating-point_format