sql server 2008 not generating random value - sql-server-2008

Question

I'm having a problem with random values being generated for each row in a result set in SQL Server 2008. I found a similar question here, but upon implementing the proposed answer, I saw the same problem as before. When running the query I have provided below, it seems that the same values will sometimes show up in consecutive rows, even though I'm calling for a new NEWID() with each row.
DECLARE #Id int = 0
DECLARE #Counter int = 1
DECLARE #Value int
CREATE TABLE #Table1
(
id int identity(1,1)
,Value int
)
WHILE #Counter < 100000
BEGIN
INSERT INTO #Table1 (Value)
SELECT CAST(RAND(CHECKSUM(NEWID())) * 100000 as INT)
SET #Counter += 1
END
SET #Counter = 0
WHILE #Counter < 5
BEGIN
SELECT
#Value = T.Value
,#Id = T.id
FROM #Table1 T
WHERE T.id = CAST(RAND(CHECKSUM(NEWID())) * 100000 as INT) + 1 + #Counter
IF #Id <> 0
SELECT #Value AS Value ,#Id as ID
SET #Counter += 1
END
DROP TABLE #Table1
If I change the INT to a BIGINT, as suggested in the link I provided, nothing is solved, so I don't believe that it's an "overflow" issue.

Answer 1

If I take the calculation out of the select, I don't get the doubled rows:
DECLARE #Id int = 0
DECLARE #Counter int = 1
DECLARE #Value int
-- new variable
DECLARE #DERIVED INT
CREATE TABLE #Table1
(
id int identity(1,1)
,Value int
)
WHILE #Counter < 100000
BEGIN
INSERT INTO #Table1 (Value)
SELECT CAST(RAND(CHECKSUM(NEWID())) * 100000 as INT)
SET #Counter += 1
END
SET #Counter = 0
WHILE #Counter < 5
BEGIN
--set here to remove calculation from the select
SET #DERIVED = CAST(RAND(CHECKSUM(NEWID())) * 100000 as INT) + 1 + #Counter;
SELECT
#Value = T.Value
,#Id = T.id
FROM #Table1 T
WHERE T.id = #DERIVED
IF #Id <> 0
SELECT #Value AS Value ,#Id as ID;
SET #Counter += 1
END
DROP TABLE #Table1
I'm seeing the duplicates every time with the pseudorandom generator inside the select. Oddly enough, I get about the same frequency of duplicates on the insert loop whether or not the calculation is inside the insert... select. It could be coincidence, since we are dealing with a randomly selected number. Note also, that since you're adding to the pseudorandom result, the results aren't technically duplicates. They're descending sequences:
11111 + 1 + 1 = 11113
11110 + 1 + 2 = 11113
Same overall result, different pseudorandom result. However, if I change
CAST(RAND(CHECKSUM(NEWID())) * 100000 as INT) + 1 + #Counter
to
CAST(RAND(CHECKSUM(NEWID())) * 100000 as INT) + #Counter + #Counter
I still consistently get duplicates. That implies that the optimizer may be caching/re-using values, at least on the select. I'd call that improper for a non-deterministic function call. I get similar results on 10.0.1600 and 10.50.1600 (2008 RTM and 2008R2 RTM).