Question: evalhf, is it using 32 bit or 64 bit? Using Maple 17.02, windows 7, 64 bit

I was answering this question on another platform, and wanted to compare with Maple. On Matlab, this is the result

format long
916.536 + 3.3

      9.198359999999999e+02

When using Maple I get

evalhf(916.536 + 3.3);
     919.836000000000013

I am on a 64 bit intel PC, and the OS is windows 7, 64 bit, and I my Maple is 64 bit version also.

Using Mathematica on my PC, which is 64 bit also, I get same as Matlab:

FullForm[916.536 + 3.3]
    919.8359999999999`

and help on evalhf says:
"A call to evalhf evaluates an expression to a numerical value using the floating-point 
hardware of the underlying system. The evaluation is done in double precision."

So, the question is, why I am not gettting the same result as those others shown above?
May be it has to do with this:

"The evalhf function converts all its arguments to hardware floats, computes the answer 
and converts the answer to a Maple float result."

?