Looking for system specifications for a modern GIS workstation [closed]

Question

My current employer has given me the opportunity to design my own GIS workstation. The problem is I have no idea what is important when it comes to hardware! I have been given a datasheet from Fujitsu which includes the latest CELSIUS M720 series workstations but the configuration includes about 30 different processors, graphics cards, and hard drives.

I will be working a lot with the ESRI desktop suite (spatial analyst and 3D analyst) and doing a fair amount of raster processing (DEMs, visibility analysis, etc.), but I also use the GDAL library, SAGA-GIS, Quantum GIS and others. Does ArcGIS 10 actually use multiple cores?

I have the choice from 4 cores with 8 threads up to 8 cores with 8 threads - 16 threads?

Is SSD SATA III better than HDD SATA III?

The Celsius R920 supports up to 512 GB RAM. But what do I need?

It will probably be the last time I get a new workstation for the next 5 years, so I am curious what others have ordered or feel will be necessary for the coming years. Obviously the cloud isn't taking over as quick as they say!

Answer 1

I'm composing a build for a workstation to support consulting. At work for the past 12 years I've had daily use of various Xeon flavors of engineering workstations. The "corporate" standard build for CAD designers is typically what I draw from---and sometimes help IT staff tune the specs for.

The newer ArcGIS Server (10.1 and up) seem to run much faster for map rendering, but use much more memory; where 2 GB/thread used to be fine, now that need is edging up toward 3.5GB/thread.

On workstations, I've been rather pleased with SSD system drives, although I'd say that after a few months their difference versus a 10k-spin hard disk is not overwhelming. I'd spend the extra $ (or local specie) for an SSD system drive and make sure to have plenty of USB 3.0, or their successors for fast external commodity disks. Maybe don't even have a spinning hard disk inside the case. If you are doing a lot of raster work, it might be worth considering the throughput advantage of a pair of 10k spin drives for separate read and write bandwidth.

For reliability, consider assembling your own external drive of Network Storage class (like Western Digital Red), and only use the Costco-type drives (usually WD Green or frail Seagate models) for backup and exchange.

For ArcGIS Desktop use, sure we're limited to a thread or two per process. But if you have a big load of processing to do, consider this trick to use up to 15 threads: copy separate File Geodatabases for each process, and then launch as many ArcCatalog instances as you need. If each one is working on a separate tile, you can load up a 16-thread workstation to whatever level is tolerable. Before long, you'll need that pair of 10k drives to handle the storage bottleneck.

So in general the clock speed and generous memory is the first best choice, but if your workflows involve bulky geoprocessing, be that raster, or map cache rendering, hydrologic modeling or contouring, it can be very reasonable to consider the advantages of a 12-thread workstation over an 8-thread.

The Xeon chips in a unitary processor system don't seem to have performance advantage for ArcGIS; they don't overclock but should be very reliable at whatever speed they're rated for. For a GIS workstation these days it seems unnecessary to have multiple CPU sockets. Overclock an i7-3930K and it might be more stable and longer-lasting than an i7-4930K; overclocked with extra cooling, either would likely outperform the Xeon E5-1650v2 found at comparable price point. Those LGA 2011-board chips are options for 12 threads. At 8 threads you can use LGA 1150-board chips like the newer i7-4770K, have a bit less L3 cache memory on the CPU itself, and burn about 2/3 the energy (84W vs. 130W design power).

The LGA 2011 chips can handle quad-channel DDR3 memory, so for those it might be worth getting 4x8GB or 32GB of DDR3-1866 (or even faster memory if your ambitions run toward overclocking). For a 12-thread system right now, 16GB is probably enough. If your build is reliable it might last long enough to want more memory in the next three or four years. Making use of quad-channel options you'd be looking at replacing all four banks of memory for an upgrade; consider just getting it now.

The capabilities of graphics cards have become so far beyond the needs of most GIS apps that I'd consider any current or potential gaming desires to inform that decision. Right now, ArcGIS is not tapping into the floating-point power of graphics cards; geophysical software probably is.

Just a single 27-inch to 30-inch display (at 2560x1600) might have plenty of resolution for interactive GIS work, and lessen the need for graphics cards that support two monitors.

The opinions above are entirely my own and do not represent those of my employer.

Answer 2

Disk I/O has usually been the bottleneck when it comes to GIS for most uses. A reliable (keyword) Solid State Drive will be your best bet assuming you have at least a Sandy bridge processor (i'd wait a few weeks for Ivy Bridge if you don't have a processor yet) and a decent amount of ram (8gb minimum for today's Ram prices).

Unfortunately, Esri's ArcMap is a quite inefficient program clock-for-clock compared to other major software (adobe suite, microsoft office, SQL db's oracle db's, etc). There's nothing we can do about their inefficient programming but vote with our money and use alternative software.

Answer 3

Does ArcGIS 10 actually use multiple cores?

ArcGIS 10 can leverage multiple cores by e.g. launching geoprocessing tools as background processes. Unless you're running a massive amount of parallel geoprocessing tools, I wouldn't go for maximum number of cores. It's better to get fewer, but with more horsepower.

The Celcius R920 supports up to 512 gb RAM !!! yes RAM!! but what do I need?

As for ArcGIS 10, the desktop version is still a 32-bit software which means that it cannot utilise more than ~2-3 GB of RAM per process (depending on your OS specs). 10.1 is supposed to introduce 64 bit version of the ArcGIS desktop in which case more RAM would make sense too. Other tools you mentioned (QGIS, GDAL) don't have this restriction if you're using the 64 bit versions of the software and the OS.

Answer 4

I would absolutely get an SSD over a mechanical hard drive as your system boot drive and application installation drive. If you deal with processing large data sets, you may want to use the SSD for that too (or get a second SSD to use as a scratch disk). You will probably still need a larger mechanical HDD for storage.

ArcGIS 10 cannot use multiple cores except by running geoprocessing in the background as jlehtoma states, but that still only uses two cores (one for the ArcMap window, one for geoprocessing). So you're better off with a quad-core CPU that has a higher maximum CPU speed than with an 8- or 12-core server CPU with lower speeds for each core. I have an Intel Core i5-2500K overclocked to a 4.2 Ghz turbo speed (some Intel Core CPUs will run one core faster and shut down others when needed, Intel calls that 'turbo boost'). I would NOT go with a Xeon processor if ArcGIS is your main software.

Answer 5

Some of those applications are threaded, some aren't, its a difficult question without knowing your exact usage patterns. As of v10, ArcGIS can use up to 2 cores simultaneously, one for the main application and one for a geoprocessing. Of course, license-depending, you can also run multiple copies of ArcGIS at once. ESRI's long winded answer to this question is here: http://support.esri.com/en/knowledgebase/techarticles/detail/31903

The only times you're ever going to max out cores is when you're doing heavy-duty (geo)processing, and that will only happen if the software itself can use multiple cores (most struggles, i.e. ArcGIS). I have access to a 16 core machine I can use for GIS processing and its incredibly rare for me to max out more than two or three of the cores at a time. I'd mostly concur with Cindy - more CPU power is probably more important than more Cores, but it really depends what you're going to be doing with the machines.

Semi-relatedly given the number and nature of applications you're going to be running, I'd strongly advise getting as much RAM as possible (8-12GB) and a 64bit OS to utilise it.

Answer 6

I have a Xeon at work, with 8GB RAM, which tends to stop responding when I work with raster images. My previous i7 with 6GB RAM handled it fine, although the Xeon appears faster in all other areas. I do think though, since you have been given free range, that you should get as much RAM as possible. I'm also looking to get a new workstation for GIS at home, so it would be great to hear what your final setup is.

Answer 7

Additional cores would be wasted on ArcGIS, so rather go for fewer cores at a higher speed. Although if your budget allows it, you might as well try to future proof your machine as much as possible (at least for a few months anyway).

Answer 8

You might as well consider going with the new Core i7 as opposed to the Core i5 CPU, just to future proof the system and give you better performance in other areas. Keep in mind that some of the new i7 motherboards can handle up to 128 gig of RAM. Of course, you need a 64 bit operating system for that.

Answer 9

I don't know ArcGIS , but i have some experience using FME (x64), postgresql, QGIS and other opensource apps. Only time when i have got all cores in my i7 to almost 100% was time when i processed 100+ laz files (reprojected and saved to local postgis) 8 files at same time from FME workspacerunner.

I would build desktop machine for processing like this:

OS: 64bit, if using windows be aware that not all windows licenses support same amount of RAM or cpus

i7 core speed is key here, but you want have several cores (minimum 8, would take more if possible). GIS on desktop doesn't need multiple cpu's (my opinion) If workloads need several CPU's is better to move those on servers

RAM so much you can get even if you use 32bit programs , why? because RAM is also used for caches and if data is in cache it will be 100 times faster to read it from there than disk also swapping kill performance. for my coding/database use and usually small amounts of gis data (i think record is 35 100Mt laz files , drape linestrings on surface using FME) i would take minimum 32G (16G is ok , but after one day with 16G i needed 24G+ to get one job done ) if money is not issue 128G. used memory should have good latency and througput

And last but now least, several hardisks, why? Because it is disk IO which kill performance when you handle one task with big dataset or when doing parallel processing. Again if money is not issue , would go 1 ssd for programs, 1 for tmp files, x times raid 1+0 spinning disk for those huge datasets which you use. The key here is share disk io between disks and use prober disk for data. SSD's have a lot better IOPS , but not so good Throughput, where raid 0 spinning disks have huge throughput but IOPS is bad. Databases usually like SSD's and big files like raid1 disks. Again, key here is share io load when reading and writing data.

Graphics, pretty much anything goes with enough RAM to handle two monitors. Note: If you have some special program that support OpenCL or nvidia CUDA then get one. Those GPU based systems can speed up processing 10x-100x ( i don't know if any GIS really supports CUDA (Manifold maybe ? ) )

This from guy who spends most of his days coding C# (servers) , developing web systems and now and then processing cad to gis and fixing topology errors (using code) in those mentioned cad files.