Hardware recommendations for heavy ArcGIS Pro usage

15879
14
05-30-2022 02:55 PM
Labels (1)
burque
by
New Contributor II

Hello Everyone,

We are going to start managing the IT for a local environmental consultant. One of the projects they want us to work on is a new PC upgrade for the users. They use ArcGIS Pro, which is new to us (the MSP). From my research, and from talking to our new client, it looks like they are considered to be heavy users of ArcGIS Pro. We know that they tend to use spatial analysis often, which looks like can be accelerated using a decent GPU. If we knew more about ArcGIS Pro, we would include more examples. We think it's safe to assume that they need hardware that can handle all of the different types of workloads ArcGIS Pro is capable of producing.

So here are some questions.

1.  Does the CPU benefit from:

1A.  A faster clock speed? (single threaded performance)
1B.  More cores? (multi threaded performance)
1C.  More on-die CPU cache? (for example: AMD 5800X3D since it has AMD 3D V-Cache)

For question 1, we believe the answer is 1A (a faster clock speed as ArcGIS Pro is mostly single threaded). However, we don't know if there are certain use cases where 1B or 1C would be better answers. If so, which use cases?

2.  Consumer or Workstation GPU?

2A.  If we stick with NVIDIA GPUs, would additional CUDA cores be the best way of gaining performance?
2B.  Would double precision floating point performance make a perceivable difference in speed?
2C.  Would there be a point of diminishing returns? How about for CUDA cores and RAM capacity? Is a 3060 (non-Ti) be enough for heavy ArcGIS Pro usage?

For around the same price of a RTX 3080 Ti 12 GB card (~$1230-$1300), would a RTX A4000 16 GB (~$1130) offer more performance? The RTX 3080 Ti generally has better performance (40% more CUDA cores), but the RTX A4000 has better double precision FP performance, and 16 GB (4 GB more, but also slower) RAM.

We know that spatial analysis is able to use GPU acceleration. Though I'm not sure which type of GPU (consumer/workstation) is best for performance per $.

3. Would we be better off sticking with 64 GB / 128 GB of DDR4? Or with 64 GB / 128 GB of DDR5? Beyond looking at the steep price of DDR5...

3A. At the same capacity, would the speed or latency matter more? DDR4 can easily have 14/16 CL, and is typically around 3600 MT/s. While current high end DDR5 is 30 CL, but is typically 5600+ MT/s.
3B. How rare would it be for ArcGIS Pro to consume more than 64 GB of RAM? Should we aim for 128 GB of RAM instead?

For the NVMe drive. We were planning on using a 1TB Samsung 980 Pro for the OS, and a 1 TB Sabrent Rocket 4 Plus that will only store the ArcGIS Pro files. We're planning on putting the Sabrent drive (ArcGIS files) in the M.2 slot directly connected to the CPU, and the Samsung drive (OS) in the M.2 slot connected to the PCH. We figure this would be the best way of gaining local disk performance.

Here are the two builds we was looking into. The mid range build is ~$2300, while the dream machine build is ~$4,300. We have a feeling that the mid range build wouldn't be bottlenecked by anything ArcGIS Pro is possibly capable to do. But since we (the MSP) know basically nothing about ArcGIS Pro, that might be a poor assumption. How much more of an improvement should we expect with the dream machine build? We are starting to doubt that it will be worth it, since it is nearly 2x the budget. And should just stick with recommending the mid range build.

Mid range build: https://pcpartpicker.com/user/pcpartpicker33/saved/3kBFZL
Dream machine build: https://pcpartpicker.com/user/pcpartpicker33/saved/wsNspg

Thanks for your help!

0 Kudos
14 Replies
RobertBorchert
Frequent Contributor III

does not instill a warm fuzzy when an IT professional is asking advice on computer hardware.

Most of your options are way overkill for the hardware requirements of GIS Software

i9 Quad Core

32 GB RAM

RTX 2000 series or higher.

All of these are overkill as well but will seriously maximize the performance of a GIS

0 Kudos
RobertBorchert
Frequent Contributor III

Define Heavy Use

0 Kudos
burque
by
New Contributor II

We're not asking for advice on computer hardware, as much as we're asking how ArcGIS Pro utilizes the resources and how to best optimize the hardware for this GIS software... And to figure out at which point does diminishing returns start to happen?

And I know that will depend on usage. However, we have never used ArcGIS Pro before to know what to look for specifically. Any guidance would be helpful in that realm.

The end user doesn't want to wait for anything. They are expecting things to take only a few seconds, max. So we need to make sure the hardware is fully optimized.

If ArcGIS Pro isn't resource hungry under any circumstances. Then I guess this post would seem like a funny question. However, it does look like there are certain circumstances when having the right hardware for ArcGIS Pro matters.

RobertBorchert
Frequent Contributor III

I get it now.   I don't know if you will find a lot of people that can provide nitty gritty statistics.

The hardware recommendations are pretty low.  But that being said there is going to be poor performance with those recommendation.

Pro will utilize multiple cores at a time depending on what the user is doing. 

Think of it as very similar usage to a high end video game.

My company low balls us with GIS Computers and it is frustrating waiting for my computer to catch up.

I only have a i7 Gen 8 with 4 cores, 16 GB RAM, and worst of all integrated grapics.

For most daily basic operations this works fine.  But when I try calculating thousands of entries at a time it is bogged down, or the further I zoom out and the more that is on my screen it really goes down hill.  Especially if I have large Rasters visible.  

I have a wall map project I have set up. On my work computer it may take a minute to several to generate the full view.  And then exporting that map can take a long time.

As an experiment I loaded Pro onto my personal computer

i9 10th Gen with 8 cores, 32 GB of ram and a RTX 3080 GPU.

When it takes me 10 minutes to process that wall map for export, my personal computer does it in 10 seconds.

Suffice it to say Pro can stress processor, ram, and graphics capability.

 

0 Kudos
burque
by
New Contributor II

Bummer. I was hoping going on the Esri forums would be the best shot of getting attention from someone at Esri that has some knowledge in this field.

Can you re-run that test and see if Task Manager reports either the CPU, SSD/HDD, RAM, or GPU as using close to 100% utilization? From what you are telling me, I'm guessing it probably doesn't go beyond 20% utilization. If not, please let me know which component was utilized the most.

Thanks!

0 Kudos
BillFox
MVP Frequent Contributor

Hello burque

Have you had a chance to review this resource yet?

http://www.wiki.gis.com/wiki/index.php/System_Design_Strategies

Reference: https://support.esri.com/en/white-paper/1022

 

0 Kudos
BillFox
MVP Frequent Contributor

and this one?

https://enterprise.arcgis.com/en/documentation/

Not knowing your client's specific needs for GIS, maybe start with one of each machine including a couple LARGE high resolution monitors for each workstation.

GIS will take you on a journey that can begin as a solitary desktop endeavor and through a universe (multi-verse) of possibilities including single layers of data on a local machine to terabytes/petabytes layers and 3D imagery to cloud resources, enterprise geodatabases, and back.

0 Kudos
burque
by
New Contributor II

Thanks for the wiki link. I actually have skimmed through it already. It has some good pointers, especially in a VDI environment. However, I don't see the answers to questions 1, 2, or 3 in my original post.

My gut is telling me that the mid range build wouldn't be fully utilized, so sticking with that hardware (or even less powerful hardware) is going to be pretty safe. Even under a heavy load.

Or am I missing something?

0 Kudos
RobertBorchert
Frequent Contributor III

Here is  some moderate usage with my work computer

RobertBorchert_0-1654111342413.png

This is with 4 cores and 8 logical processors

 

0 Kudos