none
Information about performance of PQ on different Intel processors RRS feed

  • Question

  • I have tens of Queries (with tens of steps) in PQ pipeline. Whether it Excel or PBD, it works, but UI is very slow. I wonder - what is the correlation between processor power + memory size and PQ performance?

    1. In particular, what performance gain I get if I change i5-9500 on i9-9500 (8 cores)? I understand that it depends on the nature of queries, but can you provide some general idea regarding the issue? Does number or cores mean something for PQ?

    1.1 A subquestion - I see that PQ loads 3 of 6 cores of the i5-9500, while on i5-6500 it loads all the 4 cores. Why PQ doesn't load all the cores on i5-9500? Sometime it happens, but very rare.

    And regarding memory:

    2. Why (the hell :-)) PQ uses only 256 MB per Mashup engine?? Isn't that too little nowadays?

    2.1. So does clause 2. mean that adding memory doesn't mean anything for PQ performance for a comp with 16mb (current overall utilization is 50% of memory)?






    Sunday, June 16, 2019 4:58 PM

Answers

  • I think this have a pretty good take on the issue you are asking about:: https://blog.crossjoin.co.uk/2019/04/21/power-bi-dataflow-container-size/

    My take is that RAM helps a little up to a certain point, but the 256MB RAM usage limit per query in the Mashup engine forces immediate offloading to "page filing" for any need beyond 256MB.  Here's a quote: "a query which uses more than 256 MB is now forced to page RAM to/from disk. Enough paging, and the performance cost can be quite dramatic."  My best performance improvement came from replacing my hard drive with an SSD.

    Tuesday, June 25, 2019 7:57 PM

All replies

  • I think this have a pretty good take on the issue you are asking about:: https://blog.crossjoin.co.uk/2019/04/21/power-bi-dataflow-container-size/

    My take is that RAM helps a little up to a certain point, but the 256MB RAM usage limit per query in the Mashup engine forces immediate offloading to "page filing" for any need beyond 256MB.  Here's a quote: "a query which uses more than 256 MB is now forced to page RAM to/from disk. Enough paging, and the performance cost can be quite dramatic."  My best performance improvement came from replacing my hard drive with an SSD.

    Tuesday, June 25, 2019 7:57 PM
  • Hi Andrey. If Estaben's reply is not sufficient, can you explain what's lacking instead of merely unmarking his reply as an answer?

    Ehren

    Tuesday, September 3, 2019 10:13 PM
    Owner