It is notable for supporting 16-bit floating point values ("half-precision"), with a sign bit, five bits of exponent, and a ten-bit mantissa. This allows a dynamic range of over thirty stopss of exposure.
Lossless compression of high dynamic range data is also supported.