Entry level workstation for GIS and scientific computing (1000$)

[femtomatic]

Hi everyone,

I'm a PhD student working in GIS and remote sensing and want to build a new entry level home workstation. This will be my first build and would appreciate some advice. So here we go:

Approximate Purchase Date: hopefully before christmas, but there is no rush.

Budget Range: 1000-1200$ CAD

System Usage from Most to Least Important: Satellite image processing, GIS, physics simulation, raw image processing (from DSLR camera), office work/word processing (publish or die!), maybe some gaming.

Are you buying a monitor: No (I already have two)

Parts to Upgrade: Everything except peripherals (my current build is 6+ years old...). I also have a 1TB external eSATA HDD that I will keep for backups and extra storage.

Do you need to buy OS: No, I will be using Linux exclusively

Preferred Website(s) for Parts: NCIX.ca or Newegg.ca (I'm open to suggestions)

Location: Quebec, Canada

Parts Preferences: I'm looking at a build based on the AMD FX-8350 for two reasons: price and the ability to use ECC ram. I'm looking for speed, stability and reliability, but my budget is a bit limited.

Overclocking: No

SLI or Crossfire: No

Your Monitor Resolution: 1920x1080 and 1280x1024

Additional Comments: I will be doing heavy multitasking, simultaneaously running a GIS software, a remote sensing software, an office suite and and a web browser. Some of the software used are multithreaded, but not all of them, at least one satellite image processing software uses OpenCL and I want to experiment with OpenCL programming myself for simulations.

And Most Importantly, Why Are You Upgrading: My current home station is an old Athlon 64 based system with 1Gb of Ram and an old Radeon X700 video card. The case is 10 years old and makes some anoying vibration noises. It's still quite usable for some basic tasks, Archlinux with XFCE kept it alive and surpisingly snappy, but it's clearly showing its age. The lab at my university has some workstations available, but the time sharing can get complicated and I need a new home machine anyways.

So here is what I had in mind for the build:
CPU: AMD FX-8350 8 Core Processor
Motherboard: ASUS M5A99FX Pro R2.0
RAM: Kingston KVR16E11K4/16 16GB ECC unbuffered
SSD: Samsung 840 Series 250GB SSD
Video Card: MSI Radeon HD 7770 OC 1020MHZ 1GB
PSU: Seasonic G Series 550W ATX 80PLUS Gold Modular
Case: Cooler Master CM 690 II Advanced ATX Mid Tower
Optical Drive: ASUS DRW-24B1ST 24X SATA DVD Writer
Card reader: Nmedia C98 5.25IN Internal Media Bay With Flash Card Reader
Wireless Card: ASUS PCE-N15 300Mbps 802.11B/G/N Wireless PCI-E Network Adapter

Admittedly the wireless card, flash card reader and optical drives are a bit of a coin toss, but I put a lot of thought (maybe too much) in the other components.

Maybe the ECC RAM is a bit overkill, but for the price difference I believe it's worth it, especially if I wan't to have reliable results for my PhD work. It's one source of uncertainty that I won't have to care about. The Asus motherboard I chose supports Unbuffered ECC, but if anyone knows of other MB that supports ECC and has similar features I'm open to suggestions.

I chose the AMD video card over Nvidia because they seem to have better performances for GPGPU. From what I read, the Nvidia gaming drivers are a bit crippled in this area.

Finally, I'm a bit hesitant with the new Samsung 840 SSD, seeing that they introduced their new homemade controller with this drive. Does anybody have any experience with it? I saw a few interesting benchmarks, and the 830 seemed to be really reliable, but I'm a bit unsure about the long term reliability of this one (and I guess only time will tell).

So, any comment/advice/suggestions will be greatly appreciated.

 

[Draven35]

what applications are you going to be using for GPGPU?

 

[FinneousPJ]

Solid build.

I believe Samsung offers a 3 year warranty on their SSDs so I'd say you're pretty well off buying it.

You could probably make do with a cheaper mobo.

I'd recommend an FX8320 and a CPU cooler for overclocking, is there any specific reason you don't want to OC?

 

[femtomatic]

Thanks for your replies,

@Draven35: The first application will be satellite image processing, I will be using Orfeo Toolbox which is basically a set of C++ libraries specialized in remote sensing applications. I also want to experiment a bit with physics simulations, essentially simulating interactions of electromagnetic waves with varous targets for satellite radar imaging applications. I have no idea if this last application is well adapted for GPGPU, but the first one had some interesting results.

@FinneousPJ: Like I said, this is my first build and I'm not too familiar with OC. My first impression is that it's more adapted for gaming and that I might loose some stability. The simulations I currently run sometimes takes more than 30 hours to complete (but that's on a slower machine), so I don't know how an overclocked processor would react on such a load.

Also, I just had a quick look at the CPU cooler prices and it seems to me like it could end up being more expensive to buy the 8320 with a good CPU cooler than a stock 8350. The price difference between the 8320 and the 8350 is only 30$ at NCIX. But, like I said I don't know much about OC so I will look into it, if I can reach higher performances with an OC'ed 8320 without sacrificing too much stability I will consider that option.

 

[Draven35]

Femto:

Ok, I just wanted to make sure that you weren't going to be trying to use CUDA stuff on a Radeon.

 

[femtomatic]

@Draven35: No I was planning on trying OpenCL, that's what Orfeo Toolbox already uses.

@FinneousPJ: Well, I looked into overclocking and found some contradicting opinions. Some say it's a big no no to OC for scientific computations, but some say that it can be worth it if done correctly.

Personally I think I'll pass, I'm not convinced that it's safe enough and I don't want to risk introducing errors from an improperly OC'd CPU. Plus, knowing how cautious and perfectionist I am, I would probably waste so much time trying to attain the "perfect" setup that it would totally counterbalance the speed gains, at least for my immediate PhD work.

As for the mobo, the other option that I considered is the ASUS M5A99X Evo, but right now the M5A 990FX pro is only 10$ more at NCIX until the end of the month, so I thought it was a no brainer. However, I'm not sure if the boards based on the 970 north bridge would be good enough for this setup. Would the slower FSB cause a bottleneck with these components? I admit I don't know much about these types of details. Would the ASUS M5A97 R2.0 AMD970/SB950 be good enough?