locked
CUDA for Power Query and Power Pivot RRS feed

  • Question

  • Hello!

    Just recently I stuck with some performance issues in Power Query and Power Pivot...

    So my question is - is there any plans to implement CUDA support into these applications?

    Friday, December 19, 2014 7:08 PM

Answers

  • You would be better served in the short term to ask whether there are optimizations you can make to your existing code, than whether the software will be updated at some time in the future to utilize new architecture that may not even address your problem.

    In Power Pivot, there are a number of single-threaded operations that no amount of parallel processing would address, only max clock and IPC would affect these operations. If you are using many iterator functions, particularly nested ones, you would see no performance increase from CUDA integration since these operations are implemented as single-threaded always.

    There are also some operations that DAX is not optimized to deal with which can be trivially solved with other tools.

    Why not examine your implementation of logic and determine where it can be improved.

    There's a good graph in this article illustrating the fallacy of throwing hardware at a poorly thought out implementation.

    Friday, December 19, 2014 7:27 PM

All replies

  • You would be better served in the short term to ask whether there are optimizations you can make to your existing code, than whether the software will be updated at some time in the future to utilize new architecture that may not even address your problem.

    In Power Pivot, there are a number of single-threaded operations that no amount of parallel processing would address, only max clock and IPC would affect these operations. If you are using many iterator functions, particularly nested ones, you would see no performance increase from CUDA integration since these operations are implemented as single-threaded always.

    There are also some operations that DAX is not optimized to deal with which can be trivially solved with other tools.

    Why not examine your implementation of logic and determine where it can be improved.

    There's a good graph in this article illustrating the fallacy of throwing hardware at a poorly thought out implementation.

    Friday, December 19, 2014 7:27 PM
  • Thank you for the clear answer.

    As for providing the code I'm not sure yet that it will help because the main problem in my opinion is that I just need to transform large amount of data from simple excel files using Power Query... I suppose the main problem is that all this data requires a lot of transformation but stored in a simple Excel files... That's why Power Query can't use a more progressive query language like SQL.

    Is there any way to convert all these excel files to any database format? Would it help?

    Saturday, December 20, 2014 1:44 PM
  • Power Query is not my forte, so I can't help out too much there. If you're more comfortable working in a SQL RDBMS, Excel can save files to .csv format, and pretty much every RDBMS I know of has a simple facility for consuming .csv files.

    Saturday, December 20, 2014 4:29 PM