This uses a binary mask to select the unique sets where one of each character are present. It runs in about 14 seconds.

declare @t varchar(10) = 'ABCDEFGH'

;with src(t,n,p) as (

select substring(@t,1,1),1,power(2,0)

union all

select substring(@t,n+1,1),n+1,power(2,n)

from src

where n < len(@t)

)

select s1.t+s2.t+s3.t+s4.t+s5.t+s6.t+s7.t+s8.t

from src s1, src s2, src s3, src s4, src s5, src s6, src s7, src s8

where s1.p+s2.p+s3.p+s4.p+s5.p+s6.p+s7.p+s8.p=power(2,len(@t))-1