demo, Sql Server - rextester

Language: Layout:

create table t  (ID int,      Name nvarchar(100),  Salary int, rownum int)
insert into t values
(1, 'Jon  ', 500     ,1)
,(1, 'Jim  ', 600     ,2) 
,(1, 'Jack ', 700     ,3)
,(1, 'Bob  ', 1000    ,4)
,(2, 'Adam ',  500    ,1)
,(2, 'Aron ',  600    ,2) 
,(2, 'James',  900    ,3)
,(2, 'Jay  ',  1000   ,4)
,(3, 'Adam ',  500    ,1)
,(3, 'Aron ',  602    ,2) 
,(3, 'James',  660    ,3)
,(3, 'Jay  ',  800   ,4)
,(3, 'Jay  ',  660   ,5)
,(3, 'Jay  ',  740  ,6)
,(3, 'Jay  ',  1000  ,7)
;

; with cte as 
(
    select *
    , toUpdate= case 
                    when 
                        salary - lag( salary) over( partition by Id order by rownum asc) between 0 and 100 
                    then 
                        1 
                    else 
                        0 
                end
from t
)

update t5 
set t5.salary=t4.newsalary
from t t5 
join
(
    select 
    t1.*, 
    rn=row_number () over( partition by t1.id, t1.rownum order by t3.rownum desc), 
    newsalary=t3.salary 
    from cte t1 
        outer apply
            (
                select *  
                    from cte t2 
                where t2.id=t1.id 
                    and t2.rownum<t1.rownum 
                    and t2.toUpdate=0
              ) t3
    )t4 
on t4.rn=1 and t4.toUpdate=1 and t5.id=t4.id and t5.rownum=t4.rownum

select * from t

View schema
Execution time: 0,03 sec, rows selected: 15, rows affected: 21, absolute service time: 0,19 sec