Alteryx Hardware - The Perfect "Alteryx" Build

Question

Lately my ADHD has been bringing me to the topic of computer performance. So I've been thinking more and more that there needs to be a high-end computer build JUST FOR ALTERYX! More powerful than the Alteryx 'High Performance' Recommendation here!

Why not? They have one for the Adobe's, Davinci's, Cinema, Unity, Redshift, AutoCAD, Redworks, Solidworks, (other works); they even have builds for Machine Learning (Tensorflow, Anaconda, etc...). What about Alteryx? Nothing. Sure, there are a few posts that scratch the surface (See patrick_mcauliffe's "Hardware Matters", or MarqueeCrew's answer here ) but nothing too deep on the details. What makes Alteryx so unique are the variety of tools that you can use. I propose that the variety of Alteryx tools each use hardware a little bit differently. From Sorting data to Rendering Reports, different components are used in different ways.

I've been trying to get Puget Systems to start testing Alteryx for a good build, but they lack the demand to justify the cost. So it brings me to this... I'm calling you Alteryx Programmers out! Please help me figure this out! For each of the tools individually below, what does that tool rely on to be powerful? Single-Core? Multi-Core? Could a graphics card be used to spread the love? Does it rely on a fast Hard Drive (Fast SSD)? What do we need more of to get the most out of:

* Render Reports* Report Map
* Interactive Chart,
* Overlay
* Table
* etc...

* Preparation* Sort
* Multi-Field Formula
* Multi-Row Formula
* Auto Field

* Transform* Count Records
* Summarize
* Crosstab
* Transpose
* Arrange

* Join* Regular Join
* Join Multiple
* Append
* Make Group
* Find Replace
* Fuzzy Match

* Spatial tools* Create Points
* Distance
* Find Nearest
* Trade Area
* etc...

* Macros* Regular
* Batch
* Iterative

* R-Based Tools* Plot of Means
* Decision Tree
* AB Trend
* etc...

* Python Tool
* Developer Tools* Dynamic Input
* Dynamic Rename
* Dynamic Select
* etc...

This is the beginning of something great. If this works out, I plan on going to Puget Systems and having them build the perfect "Alteryx" build.

...then I will name my PC @Ned

patrick_mcauliffe · Accepted Answer

I'm going to kick off this party.

First - the "perfect" Alteryx build has to be something that's constantly evolving, just like Alteryx and the available hardware.  That's always been my number one reason for not committing to exact parts.

So, let's make this a snapshot in time and agree we can update it later.

Here's what we don't need for a strictly Alteryx build: GPU.  It doesn't do anything to move your workflows right now.

Here's what we do need:

Multi-core CPU.  I'd say a minimum of 8 physical cores because I like to have more than one workflow running at a time.

High frequency CPU.  Mine is overclocked to all cores at 5 Ghz.  That's worked out well for me.  When Intel comes out with something that can be clocked higher, I'll probably get that.

That brings me to.... Intel CPU.  Xeon to be specific.  I have no hard facts to back this up, I've just always gotten better performance out of Intel.  If someone wants to donate an AMD system for me to supercharge and prove this wrong, I wouldn't turn down the opportunity.

FAST storage.

OS Storage: I would use nothing less than a PCI-e connection.  Ideally Gen 4.0 x8, per today's standards.  Also, it's gotta be Samsung for me (PM1735 for sure)

Other storage:   Probably a few SAS SSD's (PM1643 anyone?).

High bandwidth, FAST memory (and a lot of it).

Let's say 128 GB of ECC DDR4-3200.

Oh, and we'll assume we're connected to a solid LAN/WAN network, so it has to have a 10G SFP+ connection.

simonaubert_bd · Accepted Answer

Just a quick reply to complete @patrick_mcauliffe :

-the last AMD CPU are amazing in every benchmark, far beyond intel CPU in term of perfomance/$ (https://www.cpubenchmark.net/cpu.php?cpu=AMD+Ryzen+9+5950X&id=3862). However, I don't have a feedback on Alteryx with it.

-the more RAM you have, the less your workflow will have to use the disk : it will go on disk only if the RAM quantity is not enough. But sometimes you have to use the disk... so yes, PCIE-4 NVME disk or maybe even an intel optane or a micron x1000 ssd (both wth 3d xpoint)
-8 cores is good, not that expensive. But note that for a server, the price of license is by core with a minimal of 4 and if you want 8 cores, it's really more expensive. If you have big data and you ambition to go on a server later, I would recommand to use in-db workflow, so that you can put all the calculation power on the database infrastructure. Money matters 😉