Varying Precision


The above graph has logarithmic QUIPS on y-axis. It shows HINT run by varying precision (data for the computation) i.e. long (32 bits), double (53 bits) and longlong (64 bits) on a single computer.

There is a tradeoff between answer quality versus calculation speed. The 32-bit graph has the highest QUIPS, but exhausts its precision at about 0.1 second. The 53-bit precision of an IEEE double allows answer refinement to about 3 seconds. The 64-bit graph, though lowest in QUIPS, run the longest time before running out of precision.

If the NetQUIPS for integer precision is higher than that of floating point, it may indicate a computer designed for business and text-processing work. Scientific computers typically achieve higher NetQUIPS when using 64-bit floating point data types. (HINT can be run using any data type that support +-*/ on whole numbers).

next up previous

Next: Varying Clock Speed Up: Understanding HINT graph Previous: Memory Regime Graph