In servers, laptops and desktops, FP has roughly the same performance as integer. In lower performance setups (phones, embedded platforms etc) FP is sometimes up to 200x slower. Sometimes you will see a setup where FP is actually faster than integer, but not by a large margin.

FP used to be almost universally slow 15 years ago, but that time is over. It ended when Intel made Pentium FP multiplication to execute at 3 clocks, while integer multiplication stayed at 9 clocks. And just when people started converting integer code to FP to take advantage of it, Pentium II had both FP and integer multiplication at 1 clock. Nowadays there is even greater variety of CPUs around, so you cannot be sure of FP vs integer performance before you test.
Division is costly operation as compared to multiplication.

You can see speed comparision@
http://nicolas.limare.net/pro/notes/2014/12/12_arit_speed/

Sparc Architecture FPU performance
http://pages.cs.wisc.edu/~fischer/cs701/sparc.htm

GPU SP and DP Performance: DP are really slower on GPU
http://arrayfire.com/explaining-fp64-performance-on-gpus/

On modern workstation/laptop-range processors it isn't really slower at all; vector registers can pull off 4 or 8 flops per cycle if you stroke them right, and they're not half bad without making any particular motions either.

A rule of thumb would be that the memory footprint of your data has greater impact on speed than the arithmetic does nowadays. Comparing apples to apples, say 64-bit integers with 64-bit floats, there is no good reason to worry about slowing down your program just from calculating with decimals here and there.

It depends on what you mean by "slower". In terms of pure throughput, most processors -- even ones in smartphones -- are capable of pushing through single-precision (SP) FP calculations as fast as 32-bit integer calculations for addition/subtraction and multiplication (along with various other misc operations like absolute value). Divisions are more complicated. Intel's desktop/server class processors do FP divisions much faster due to AVX (being capable of 8 32-bit FP divisions in parallel) compared to integer divisions. Their Atom processors aren't as good, being able to do 4 32-bit FP divisions in parallel but with a long latency but still faster than integer divisions.

1. Data points for Nehalem:


• IDIV does 1 (up to 64-bit) integer division at a time. It can have a throughput of 1 every 5-14 cycles (depending on the value of the divisor). 32-bit divisors have lower latency than 64-bit divisors but it doesn't say how much exactly.
• VDIVPS does 8 32-bit SP FP divisions every 14 cycles.


2. Data points for Atom:


• IDIV does 1 (up to 32-bit) integer division at a time. It can have a throughput of 1 every 56 cycles.
• DIVPS does 4 32-bit SP FP divisions every 69 cycles.