Why does ArcGIS Pro have to be so slow???

91723
262
08-01-2017 11:31 AM
ericmeyers1
New Contributor III

Why is ArcGIS Pro so slow? To select assets, field calculate, display layers, change symbology... the easiest of tasks that are commonly utilized within ArcMap are a drag on the software.

When will ArcGIS Pro become faster than ArcMap? That will be the day it could replace it as the goto product for GIS professionals.

262 Replies
SeanHlousek
Occasional Contributor III

The CPU IS a marginal choice for Pro (I didn’t make that choice). As I say, I’m advocating for an upgrade that will perform better.

That is why I decided to test a project on my home machine. I can see the difference and it IS an improvement but there still seems to be some sort of , for lack of a better term, “Project Bloat”, that is occurring with projects with many elements.

I wanted to share my experience, (even though the CPU is making my experience worse than it would be), with the community in case others are experiencing similar or the same issues.

I remember in 2002 when I started working with ESRI people were struggling with the transition from ArcView to ArcMap. I don’t know if that transition was as challenging as this one but at some point, the majority of users were using ArcMap and presumably “satisfied” with it. Hopefully threads like this one will help all of us work a little bit more efficiently in the world of PRO in a few years.

AlexZhuk
Occasional Contributor III

I agree, perhaps we should not be using the old Commodore 64 computers to run ArcGIS Pro, but also, from my two decades of experience with ArcMap, I can hardly remember times when ESRI admonished us for using a "wrong" hardware. A large part of ESRI's user base is in the government-type agencies, notoriously slow in adopting the latest tech, so it can be a little counterproductive for the developers to be that stubborn in their insistence that "the client is wrong."

JohnBrockwell
Occasional Contributor III

Look at what I am running ArcGIS Pro on:

Correction: 12 core CPU 24 threads, 128GB RAM. I barely notice a difference running on my workstation or my laptop. SMH

AndrewQuee
Occasional Contributor III

Not being negative (genuinely interested) but why so much memory and so little CPU?  Even gamers with extreme rigs are not going to 128GB yet afaik.  Are you doing some serious caching in your workflow or holding entire raster datasets in memory?

As a comparison we've found 16GB more than ample for any kind of desktop GIS/raster work we do (ArcMap, Pro, Bentley/Hexagon, AGOL/Server publishing, FME geoprocessing/replication).  We do have 32GB in a ArcGIS Server workstation which is used for bulk geoprocessing and raster operations prior to publishing.

SeanHlousek
Occasional Contributor III

It's an example of why the people actually doing the work on Arc should build the machine and not IT.

I've learned that lesson.

Sean

JohnBrockwell
Occasional Contributor III

From what I've been told the workstation(s) were built with data science tools in mind, I.e. R, STATA, etc. The CPU has 12 cores. The problem is I can't get ArcGIS Pro to distribute to all the cores. The other item that is problematic for me is the GPU is an AMD product. I would never put an AMD GPU in a machine with ArcGIS. I have found that Nvidia works better with ArcGIS.

It would be nice if ESRI provided proper hardware recommendations. Providing minimum requirements is one thing. How about optimum requirements?

Thanks for the feedback.

SeanHlousek
Occasional Contributor III

John,

I honestly question whether they really KNOW ideal specs.

I think us early adopters if the tech might be be helping to figure that out.

(I'm partially serious, but hopefully joking??)

AndrewQuee
Occasional Contributor III

Thanks for the clarification and explanation of why it's specced the way it is, that makes sense.  I should say we went down the same route with our workstation.  ArcGIS with 64-bit extensions, then Pro (could not get it to reliably parallel process GP tasks even though this is one of the selling points) and finally we gave up fooling around with it and threw Server on there.  Very much overkill but if it looks stupid and works, it's not stupid, right?  (We have an ELA so no issue there)

Our next move is to look at 'sidegrading' these workstations and distributed hardware into simpler specifications and migrating the processing to a beefy dedicated server offering remote geoprocessing through web applications, GIS Server and hopefully FME Server (automations look very handy) with a view to expanding our ability to provide enhanced spatial products/processing via intra/internet.

Potentially this would also be cost effective by reducing hardware requirements/maintenance, maximising usage/scalable cloud solution and reducing the numbers of licenced extensions.  For example theoretically we could drop all but one 3D Analyst and get Server to manage geoprocessing requests of that kind for clients.

JohnBrockwell
Occasional Contributor III

Server side geoprocessing is still a cultural thing. I have not been able to overcome in my organization. Folks are very comfortable using desktop to geoprocess. It's been hard having colleagues migrate to some of the new enhancements in ArcGIS Enterprise. Great Feedback. Appreciate it.

0 Kudos
MarcoBoeringa
MVP Regular Contributor

As I also wrote for Sean Hlousek, your results regarding the desktop versus your laptop hardly seeing a difference in Pro's performance may not be an entire surprise...

Unfortunately, the laws of physics and practicality dictate that increasing single thread performance, which is largely dependent on increased clock speeds, is increasingly difficult. Read this nice article to understand why:

Why CPU Clock Speed Isn't Increasing - Make Tech Easier 

Writing software to take advantage of multiple cores is certainly possible, but there are many tasks that simply must run in single thread, also related to the software's user interface. Those won't take advantage of increased core count. As you can see from these stats from the Passmark website, your Intel Xeon E5-2687W v4 (Intel® Xeon® Processor E5-2687W v4 (30M Cache, 3.00 GHz) Product Specifications ) launched in 2016 has a single thread performance slightly less than the five years older desktop Core i7-2600 CPU from 2011, which itself is hardly beaten by a Core i7-7700HQ laptop processor as launched in 2017:

Intel Core i7-2600 - Multi-thread stat:8180 / Single-thread:1922, launch year 2011

Intel Core i7-7700HQ - Multi-thread stat:8761 / Single-thread:1991, launch year 2017

Intel Xeon E5-2687W v4 - Multi-thread stat:19863 / Single-thread:1894, launch year 2016

Your Intel Xeon E5-2687W v4 processor with its many cores, does have quite attractive multi-core performance though ... More than double than my own laptop with Intel Core i7-7700HQ. Any geoprocessing tool using the Parallel Processing Factor geoprocessing environment setting, should run really nicely on your machine!

Even more attractive seems the new AMD Ryzen ThreadRipper 3970X:

AMD Ryzen ThreadRipper 3970X - Multi-thread stat:48439 / Single-thread:2935, launch year 2019

But even this monster 32-core 7nm processor only has about 1.5 times the single thread performance of a nine year old Core i7 CPU based on 32nm process die...