Page 1 of 1

# How do you calculate unit roundoff in a binary floating point system?

Posted 07 December 2018 - 08:53 AM

We define unit roundoff (u) as: The smallest positive number where the computed value of 1 + u is different from 1.

Let's consider an 8-bit binary floating point format with:

- a sign bit
- a 3-bit exponent
- a 4-bit mantissa with a hidden bit x0 = 1

How could I calculate the value of unit roundoff for this representation?
Is This A Good Question/Topic? 0

## Replies To: How do you calculate unit roundoff in a binary floating point system?

### #2 modi123_1 • • Suitor #2
•    Reputation: 15222
• Posts: 60,932
• Joined: 12-June 08

## Re: How do you calculate unit roundoff in a binary floating point system?

Posted 07 December 2018 - 08:57 AM

What have you tried, thought about, or considered?

## Re: How do you calculate unit roundoff in a binary floating point system?

Posted 07 December 2018 - 09:11 AM

This is how far I got:

1.000 + 0.0001 = 1.0001 which should get rounded to 1.001

This would mean that 0.0001 is the unit roundoff in this case, in binary though. This would then be 0.0625 in a decimal representation. I don't really understand where the hidden bit comes into play with all this.

Page 1 of 1

 .related ul { list-style-type: circle; font-size: 12px; font-weight: bold; } .related li { margin-bottom: 5px; background-position: left 7px !important; margin-left: -35px; } .related h2 { font-size: 18px; font-weight: bold; } .related a { color: blue; }