Post reply

Handle NULLs

  • July 11, 2011 at 11:59 am

    #1351256

    opc.three (7/11/2011)

    . . . Anyone have thoughts on The Dixie Flatline's fine solution? Hopefully conclusions won't depend on a video GPU or sound card or anything like that :hehe:

    . . .

    Are you saying that SELECT...INTO is 10% faster on Sandy Bridge than CREATE TABLE + INSERT INTO on Nehalem?

    Or is it that SELECT...INTO on Sandy Bridge 10% faster than SELECT...INTO on Nehalem? . . .

    Yes, parallelizable queries are faster on Sandy Bridge.

    Re results varying depending on GPU, you may expect that in the next version of SQLS: the new version of Visual Studio has heavy focus on async and parallelization, and the newest generation of NVIDIA cards, such as Tesla C2050, double as vector processors. I think that SQLS guys will take advantage of that, if it is available.

  • July 11, 2011 at 12:04 pm

    #1351260

    SQLkiwi (7/11/2011)

    opc.three (7/11/2011)

    Craig's testing showed it for a single-column narrow PK (INT). I wondered whether wider clustering keys would have SELECT...INTO start outperforming the CREATE/INSERT method, and while I did some basic tests using 3-column INT keys showing CREATE/INSERT was still faster, I did not wonder enough to do any exhaustive testing including character-based columns etc. so I go with Craig's conclusion 😀

    From the post http://www.sqlservercentral.com/Forums/FindPost1130873.aspx

    I added my opinion to that thread.

    Geez, you could have just said "It Depends" Paul, but then again what fun would that ever be 🙂

    Seriously though, it's good to know about the default preferences from experienced folks like yourself. Thanks for sharing that write-up, awesome stuff!

    There are no special teachers of virtue, because virtue is taught by the whole community.
    --Plato

  • July 11, 2011 at 12:13 pm

    #1351264

    Revenant (7/11/2011)

    Re results varying depending on GPU, you may expect that in the next version of SQLS: the new version of Visual Studio has heavy focus on async and parallelization, and the newest generation of NVIDIA cards, such as Tesla C2050, double as vector processors. I think that SQLS guys will take advantage of that, if it is available.

    Yeah, maybe someday it'll be fully realized...I've been hearing about the possibilities attached to leveraging GPUs for floating point math and other intense calcs for years now.

    There are no special teachers of virtue, because virtue is taught by the whole community.
    --Plato

Viewing 3 posts - 91 through 92 (of 92 total)

You must be logged in to reply to this topic. Login to reply