550
votes

Which method provides the best performance when removing the time portion from a datetime field in SQL Server?

a) select DATEADD(dd, DATEDIFF(dd, 0, getdate()), 0)

or

b) select cast(convert(char(11), getdate(), 113) as datetime)

The second method does send a few more bytes either way but that might not be as important as the speed of the conversion.

Both also appear to be very fast, but there might be a difference in speed when dealing with hundreds-of-thousands or more rows?

Also, is it possible that there are even better methods to get rid of the time portion of a datetime in SQL?

23
I've tried this out on one million records in one of my production tables and I couldn't get an accurate reading on performance either way. Both methods returned the exact same amount of data though.Stephen Perelson
On 18,000,000 rows this is what I've found (SQL Server 2008): Method b is about 24% slower than method a. CAST(FLOOR(CAST(getdate() AS FLOAT)) AS DATETIME) is 3.5% slower than method a. Method a seems to be a winner with regards to performance. Thanks all for the great answers.Stephen Perelson
Why the heck doesn't SQL have a built-in function to do this anyway?!!Gary McGill
SQL 2008's new DATE datatype will handle this.Philip Kelley

23 Answers

582
votes

Strictly, method a is the least resource intensive:

a) select DATEADD(dd, DATEDIFF(dd, 0, getdate()), 0)

Proven less CPU intensive for the same total duration a million rows by someone with way too much time on their hands: Most efficient way in SQL Server to get a date from date+time?

I saw a similar test elsewhere with similar results too.

I prefer the DATEADD/DATEDIFF because:

Edit, Oct 2011

For SQL Server 2008+, you can CAST to date i.e. CAST(getdate() AS date). Or just use date datatype so no time to remove.

Edit, Jan 2012

A worked example of how flexible this is: Need to calculate by rounded time or date figure in sql server

Edit, May 2012

Do not use this in WHERE clauses and the like without thinking: adding a function or CAST to a column invalidates index usage. See number 2 here Common SQL Programming Mistakes

Now, this does have an example of later SQL Server optimiser versions managing CAST to date correctly, but generally it will be a bad idea ...

Edit, Sep 2018, for datetime2

DECLARE @datetime2value datetime2 = '02180912 11:45' --this is deliberately within datetime2, year 0218
DECLARE @datetime2epoch datetime2 = '19000101'

select DATEADD(dd, DATEDIFF(dd, @datetime2epoch, @datetime2value), @datetime2epoch)
74
votes

In SQL Server 2008, you can use:

CONVERT(DATE, getdate(), 101)
60
votes

Of-course this is an old thread but to make it complete.

From SQL 2008 you can use DATE datatype so you can simply do:

SELECT CONVERT(DATE,GETDATE())
20
votes

In SQL Server 2008, there is a DATE datetype (also a TIME datatype).

CAST(GetDate() as DATE)

or

declare @Dt as DATE = GetDate()
20
votes
SELECT CAST(FLOOR(CAST(getdate() AS FLOAT)) AS DATETIME)

...is not a good solution, per the comments below.

I would delete this answer, but I'll leave it here as a counter-example since I think the commenters' explanation of why it's not a good idea is still useful.

8
votes

Here's yet another answer, from another duplicate question:

SELECT CAST(CAST(getutcdate() - 0.50000004 AS int) AS datetime) 

This magic number method performs slightly faster than the DATEADD method. (It looks like ~10%)

The CPU Time on several rounds of a million records:

DATEADD   MAGIC FLOAT
500       453
453       360
375       375
406       360

But note that these numbers are possibly irrelevant because they are already VERY fast. Unless I had record sets of 100,000 or more, I couldn't even get the CPU Time to read above zero.

Considering the fact that DateAdd is meant for this purpose and is more robust, I'd say use DateAdd.

6
votes
SELECT CAST(CAST(GETDATE() AS DATE) AS DATETIME)
6
votes

I really like:

[date] = CONVERT(VARCHAR(10), GETDATE(), 120)

The 120 format code will coerce the date into the ISO 8601 standard:

'YYYY-MM-DD' or '2017-01-09'

Super easy to use in dplyr (R) and pandas (Python)!

3
votes

BEWARE!

Method a) and b) does NOT always have the same output!

select DATEADD(dd, DATEDIFF(dd, 0, '2013-12-31 23:59:59.999'), 0)

Output: 2014-01-01 00:00:00.000

select cast(convert(char(11), '2013-12-31 23:59:59.999', 113) as datetime)

Output: 2013-12-31 00:00:00.000

(Tested on MS SQL Server 2005 and 2008 R2)

EDIT: According to Adam's comment, this cannot happen if you read the date value from the table, but it can happen if you provide your date value as a literal (example: as a parameter of a stored procedure called via ADO.NET).

2
votes

Strip time on inserts/updates in the first place. As for on-the-fly conversion, nothing can beat a user-defined function maintanability-wise:

select date_only(dd)

The implementation of date_only can be anything you like - now it's abstracted away and calling code is much much cleaner.

2
votes

See this question:
How can I truncate a datetime in SQL Server?

Whatever you do, don't use the string method. That's about the worst way you could do it.

2
votes

Already answered but ill throw this out there too... this suposedly also preforms well but it works by throwing away the decimal (which stores time) from the float and returning only whole part (which is date)

 CAST(
FLOOR( CAST( GETDATE() AS FLOAT ) )
AS DATETIME
)

second time I found this solution... i grabbed this code off

2
votes
CAST(round(cast(getdate()as real),0,1) AS datetime)

This method does not use string function. Date is basically a real datatype with digits before decimal are fraction of a day.

this I guess will be faster than a lot.

2
votes

For me the code below is always a winner:

SELECT CONVERT(DATETIME, FLOOR(CONVERT(FLOAT,GETDATE())));
2
votes

select CONVERT(char(10), GetDate(),126)

1
votes

I think you mean cast(floor(cast(getdate()as float))as datetime)

real is only 32-bits, and could lose some information

This is fastest cast(cast(getdate()+x-0.5 as int)as datetime)

...though only about 10% faster(about 0.49 microseconds CPU vs. 0.58)

This was recommended, and takes the same time in my test just now: DATEADD(dd, DATEDIFF(dd, 0, getdate()), 0)

In SQL 2008, the SQL CLR function is about 5 times faster than using a SQL function would be, at 1.35 microseconds versus 6.5 microsections, indicating much lower function-call overhead for a SQL CLR function versus a simple SQL UDF.

In SQL 2005, the SQL CLR function is 16 times faster, per my testing, versus this slow function:

create function dateonly (  @dt datetime )
returns datetime
as
begin
return cast(floor(cast(@dt as float))as int)
end
1
votes

How about select cast(cast my_datetime_field as date) as datetime)? This results in the same date, with the time set to 00:00, but avoids any conversion to text and also avoids any explicit numeric rounding.

1
votes

I think that if you stick strictly with TSQL that this is the fastest way to truncate the time:

 select convert(datetime,convert(int,convert(float,[Modified])))

I found this truncation method to be about 5% faster than the DateAdd method. And this can be easily modified to round to the nearest day like this:

select convert(datetime,ROUND(convert(float,[Modified]),0))
1
votes

Here I made a function to remove some parts of a datetime for SQL Server. Usage:

  • First param is the datetime to be stripped off.
  • Second param is a char:
    • s: rounds to seconds; removes milliseconds
    • m: rounds to minutes; removes seconds and milliseconds
    • h: rounds to hours; removes minutes, seconds and milliseconds.
    • d: rounds to days; removes hours, minutes, seconds and milliseconds.
  • Returns the new datetime

create function dbo.uf_RoundDateTime(@dt as datetime, @part as char) returns datetime as begin if CHARINDEX( @part, 'smhd',0) = 0 return @dt; return cast( Case @part when 's' then convert(varchar(19), @dt, 126) when 'm' then convert(varchar(17), @dt, 126) + '00' when 'h' then convert(varchar(14), @dt, 126) + '00:00' when 'd' then convert(varchar(14), @dt, 112) end as datetime ) end

1
votes

Just in case anyone is looking in here for a Sybase version since several of the versions above didn't work

CAST(CONVERT(DATE,GETDATE(),103) AS DATETIME)
  • Tested in I SQL v11 running on Adaptive Server 15.7
0
votes

If possible, for special things like this, I like to use CLR functions.

In this case:

[Microsoft.SqlServer.Server.SqlFunction]
    public static SqlDateTime DateOnly(SqlDateTime input)
    {
        if (!input.IsNull)
        {
            SqlDateTime dt = new SqlDateTime(input.Value.Year, input.Value.Month, input.Value.Day, 0, 0, 0);

            return dt;
        }
        else
            return SqlDateTime.Null;
    }
0
votes

I, personally, almost always use User Defined functions for this if dealing with SQL Server 2005 (or lower version), however, it should be noted that there are specific drawbacks to using UDF's, especially if applying them to WHERE clauses (see below and the comments on this answer for further details). If using SQL Server 2008 (or higher) - see below.

In fact, for most databases that I create, I add these UDF's in right near the start since I know there's a 99% chance I'm going to need them sooner or later.

I create one for "date only" & "time only" (although the "date only" one is by far the most used of the two).

Here's some links to a variety of date-related UDF's:

Essential SQL Server Date, Time and DateTime Functions
Get Date Only Function

That last link shows no less than 3 different ways to getting the date only part of a datetime field and mentions some pros and cons of each approach.

If using a UDF, it should be noted that you should try to avoid using the UDF as part of a WHERE clause in a query as this will greatly hinder performance of the query. The main reason for this is that using a UDF in a WHERE clause renders that clause as non-sargable, which means that SQL Server can no longer use an index with that clause in order to improve the speed of query execution. With reference to my own usage of UDF's, I'll frequently use the "raw" date column within the WHERE clause, but apply the UDF to the SELECTed column. In this way, the UDF is only applied to the filtered result-set and not every row of the table as part of the filter.

Of course, the absolute best approach for this is to use SQL Server 2008 (or higher) and separate out your dates and times, as the SQL Server database engine is then natively providing the individual date and time components, and can efficiently query these independently without the need for a UDF or other mechanism to extract either the date or time part from a composite datetime type.

-5
votes

I would use:

CAST
(
CAST(YEAR(DATEFIELD) as varchar(4)) + '/' CAST(MM(DATEFIELD) as varchar(2)) + '/' CAST(DD(DATEFIELD) as varchar(2)) as datetime
) 

Thus effectively creating a new field from the date field you already have.