Hi All,
I've got a desktop license and its currently running a windows machine (AMD Ryzen 5 2600 Six-Core Processor 3.40 GHz,all fast SSDs,16gb RAM). I do lots of predictive stuff (training & scoring forest models) and lots of spatial joins.
Really struggling performance wise, and I've been asked to do a bit of digging on improving performance but not upgrading to server.
So, some questions for you all!
-Is there a physical limit of resource that the designer can consume?
-Does it utilise GPU if I throw that at it?
-For predictive work, do I need to up mainly the RAM/CPU or both?
-What should I NOT chuck cash at upgrading?
And any time gain anecdotes from your own upgrades appreciated.
Thanks!
Steve
I'm sorry to hear that you are struggling (performance wise) and hope to give you a little pill to get your pep back into your step. That being said, each workflow has it's own performance characteristics and profile. So what I'm going to provide you with are tips rather than recommendations.
Tip1: Memory You'll always benefit from more RAM. You might consider asking for an upgrade to 32GB RAM. 16 is adequate, but you're doing predictive and R is a poor manager of RAM. Alteryx runs entirely in memory and will run fastest given plenty of RAM.
Tip2: AMP The AMP (e2) engine will speed your data management processes (not your predictive processes). The AMP engine multi-threads (specific) tools as listed here. I haven't tested model building or scoring with AMP turned on. It might run faster, but I'm concerned that if it might cause memory limits with R code.
Tip3: CReW Spatial Match You might want to test the standard spatial match with the same match (AMP turned on) and with this macro. The macro runs faster than the standard tool. When you're performing spatial joins, remember to drop the UNIVERSE spatial object (heck, drop as many spatial fields as possible) to speed up your workflow. If you don't need it, drop it.
Here is the Alteryx AMP Engine help documentation.
You don't need server. You should be able to tweak and performance tune your workflows. Here's a bonus video talking about the merits of tuning the workflow: @NicoleJohnson agrees! VIDEO POST
Cheers,
Mark