128 bit decimal floating point

[giangiquario]

Support for 128-bit Decimal Floating Point Numbers is expected for version 13.
Will the "machine epsilon" be lesser than 2*-53 for those numbers?

Thank you.

[Geoff|Dyalog]

The precision of 128 bit decimal floating point is 34 decimal digits. The range is from
¯9999999999999999999999999999999999E6011 to
+9999999999999999999999999999999999E6011
there should be 34 9's in those numbers.

[paulmansour]

Oh yeah!

[Morten|Dyalog]

Before you all get too excited, note that the performance of 128-bit decimal floating-point is more than an order of magnitude slower than 64-bit binary (because it uses software emulation), unless you are using recent IBM "Power" hardware, where it will be about 4x slower (due to having lots more bits to look at).

.. but the precision is impressive ...