This is an old question, but attempting to apply a new field (either NEWID() or ORDER BY rand()) to a table with a large number of rows would be prohibitively expensive. If you have incremental, unique IDs (and do not have any holes) it will be more efficient to calculate the X # of IDs to be selected instead of applying a GUID or similar to every single row and then taking the top X # of.
DECLARE @minValue int;
DECLARE @maxValue int;
SELECT @minValue = min(id), @maxValue = max(id) from [TABLE];
DECLARE @randomId1 int, @randomId2 int, @randomId3 int, @randomId4 int, @randomId5 int
SET @randomId1 = ((@maxValue + 1) - @minValue) * Rand() + @minValue
SET @randomId2 = ((@maxValue + 1) - @minValue) * Rand() + @minValue
SET @randomId3 = ((@maxValue + 1) - @minValue) * Rand() + @minValue
SET @randomId4 = ((@maxValue + 1) - @minValue) * Rand() + @minValue
SET @randomId5 = ((@maxValue + 1) - @minValue) * Rand() + @minValue
--select @maxValue as MaxValue, @minValue as MinValue
-- , @randomId1 as SelectedId1
-- , @randomId2 as SelectedId2
-- , @randomId3 as SelectedId3
-- , @randomId4 as SelectedId4
-- , @randomId5 as SelectedId5
select * from [TABLE] el
where el.id in (@randomId1, @randomId2, @randomId3, @randomId4, @randomId5)
If you wanted to select many more rows I would look into populating a #tempTable with an ID and a bunch of rand() values then using each rand() value to scale to the min-max values. That way you do not have to define all of the @randomId1...n parameters. I've included an example below using a CTE to populate the initial table.
DECLARE @NumItems int = 100;
DECLARE @minValue int;
DECLARE @maxValue int;
SELECT @minValue = min(id), @maxValue = max(id) from [TABLE];
DECLARE @range int = @maxValue+1 - @minValue;
with cte (n) as (
select 1 union all
select n+1 from cte
where n < @NumItems
)
select cast( @range * rand(cast(newid() as varbinary(100))) + @minValue as int) tp
into #Nt
from cte;
select * from #Nt ntt
inner join [TABLE] i on i.id = ntt.tp;
drop table #Nt;