Worst case is that you need to examine every column to make that decision. 

Unfortunately I am in this case.
And need to to get status for any change not only new inserts.
Additionally my source table can have duplicates rows eg same set of compared cols can occur many times and I have to log every single rows changes.
A good news is that the distinct count of all sets of cops to be compared is small say for the 85 millions in the source their distinct is about 3000

Any best solution?

How about a full outer joins - something like this.

demo@ORA12C> create table t1 as
  2  select d.owner, d.object_name, d.object_type, d.created
  3  from  dba_objects d ;

Table created.

demo@ORA12C> create table t2 as
  2  select *
  3  from t1;

Table created.

demo@ORA12C> insert into t1(owner,object_name,object_type,created)
  2     values('NEW','CONNOR','TABLE',sysdate);

1 row created.

demo@ORA12C> insert into t2(owner,object_name,object_type,created)
  2     values('NEW','CONNOR','INDEX',sysdate);

1 row created.

demo@ORA12C> commit;

Commit complete.

demo@ORA12C> select t1.rowid , t2.rowid,
  2     case when t1.rowid is null then 'Missing/changes in T1'
  3              when t2.rowid is null then 'Missing/changes in T2'
  4              end as results_output
  5  from t1 full outer join t2
  6  on ( t1.owner = t2.owner
  7  and t1.object_name = t2.object_name
  8  and t1.object_type = t2.object_type
  9  and t1.created = t2.created )
 10  where t1.rowid is null or
 11     t2.rowid is null ;

ROWID              ROWID              RESULTS_OUTPUT
------------------ ------------------ ---------------------
                   AAAa8bAAIAAAB3XAAA Missing/changes in T1
AAAa8aAAIAAABtXAAA                    Missing/changes in T2

demo@ORA12C>

I'm borrowing Connor's example, but with one million rows only. Since this solution uses analytic functions, it probably needs a fair amount of memory to run fast.

You want to synchronize the target table with the source table. When there is no primary or unique, non-null key, there is no way to do an UPDATE but we can synchronize with INSERTs and DELETEs. The MERGE statement can do both at once.

I start by comparing the two tables and creating a CDC-style result set: Z##OP contains a flag ('I' for insert, 'D' for delete), and z##RID contains the ROWID of rows to be deleted. The MERGE matches this result set to the target using ROWIDs to avoid an extra table scan of the target table.

SQL> define d_tab_size = 1000000;
SQL> create table t1 as
  2  select d.owner, d.object_name, d.object_type, d.created
  3  from dba_objects d,
  4  (
  5    select null from dual
  6    connect by level <= 1 + &d_tab_size / (select count(*) from dba_objects)
  7  )
  8  where rownum <= &d_tab_size;

Table T1 created.

SQL> create table t2 as
  2  select * from t1 where rownum <= &d_tab_size - 100;

Table T2 created.

SQL> insert into t2 select * from t2 where rownum <= 100;

100 rows inserted.

SQL> commit;

Commit complete.

SQL> merge /*+ qb_name(SYNC_PARTITION) USE_NL(O) */ into (
  2    select /*+ qb_name(target) */
  3      "OWNER", "OBJECT_NAME", "OBJECT_TYPE", "CREATED", rowid Z##RID
  4    from T1
  5  ) O
  6  using (
  7  select /*+ qb_name(CDC_PARTITION) */ * from (
  8    select /*+ qb_name(before_filter) */
  9      "OWNER", "OBJECT_NAME", "OBJECT_TYPE", "CREATED",
 10      case
 11        when Z##NEW = 1
 12          and sum(Z##NEW) over(partition by
 13            "OWNER", "OBJECT_NAME", "OBJECT_TYPE", "CREATED"
 14          order by null rows unbounded preceding) > sum(Z##OLD) over(partition by
 15            "OWNER", "OBJECT_NAME", "OBJECT_TYPE", "CREATED"
 16          )
 17          then 'I'
 18        when Z##OLD = 1
 19          and sum(Z##OLD) over(partition by
 20            "OWNER", "OBJECT_NAME", "OBJECT_TYPE", "CREATED"
 21          order by null rows unbounded preceding) > sum(Z##NEW) over(partition by
 22            "OWNER", "OBJECT_NAME", "OBJECT_TYPE", "CREATED"
 23          )
 24          then 'D'
 25      end Z##OP, Z##RID
 26    FROM (
 27      select /*+ qb_name(old) */
 28      "OWNER", "OBJECT_NAME", "OBJECT_TYPE", "CREATED",
 29      1 Z##OLD, 0 Z##NEW, rowid Z##RID
 30      from T1 O
 31      union all
 32      select /*+ qb_name(new) */
 33      "OWNER", "OBJECT_NAME", "OBJECT_TYPE", "CREATED",
 34      0, 1, null
 35      from t2 N
 36    )
 37  )
 38  where Z##OP is not null
 39  ) N
 40  on (
 41    O.Z##RID = n.Z##RID
 42  )
 43  when matched then update set
 44    "OWNER"=N."OWNER"
 45    delete where N.Z##OP = 'D'
 46  when not matched then insert (
 47    "OWNER", "OBJECT_NAME", "OBJECT_TYPE", "CREATED"
 48  ) values(
 49    N."OWNER", N."OBJECT_NAME", N."OBJECT_TYPE", N."CREATED"
 50  );

200 rows merged.

Elapsed: 00:00:08.392

If you run this statement a second time, you will see that no rows were merged, since the comparison result set is empty.

Best regards,
Stew Ashton

P.S. I generated this code using a package that I explain here:
https://stewashton.wordpress.com/2018/02/12/comp_sync-1-a-new-table-compare-sync-package/