Considering the Dark Side - Building a High-End PC for Work ... Advice?

[astrostu]

I'll start by saying I've been a Mac guy since I was under 5 years old. The obligatory homage to Apple done with now ...

Let's face it: Some stuff will ONLY work on Windows, Windows machines are much more customizable, upgradable, and generally cheaper due to more competition parts-wise. For work, I have to use ArcMap software which will ONLY work on Windows. I've been using Parallels for the last few years but as datasets grow, it's slower and slower on my 2008 Mac Pro, and comparable on my 2011 MacBook Pro. I also run code that eats up a ----load of RAM. This particular code has required 8.5 GB RAM and 9 hours on my laptop when analyzing a 39-Mpx image. I need to run it on ~1 Gpx images in as few pieces as possible, and the RAM requirements scale linearly.

I'd also like to run another piece of software but it is NOT compiled for Windows (it's USGS's ISIS software). It runs on Linux, so, is it at all possible to run Linux-type stuff in Windows? If not, would installing another HD, installing Linux on that, and running it in the same box work?

So I'm looking to build a machine with a near-top processor, maxed out RAM, and good graphics. This is NOT a gaming machine (I don't really game), but based on its nature I'll probably also use it for entertainment (blu-ray) since I don't have a TV, and I want this to be as upgradable as possible (don't think that'll be a problem, though). Since this is for work and I'm a contractor, I should be able to write this off on next year's taxes as a business expense, but I also don't want to break the bank. Under $2k base is what I'm shooting for. I also already have a 23" monitor I can use.

I've read some of the guides out there, and I spent a good hour or two on NewEgg this morning, and I was hoping to run this stuff by any of you knowledgable folkses out there to offer opinions.


Motherboard: Asus Maximus IV Extreme (supports Turbo Boost, 4 PCIe and 8 SATA, and the LGA 1155)
Processor: Intel i7-2700K (quad-core 3.5 GHz up to 3.9 with Turbo Boost) ... considering 2600 instead which is 100 MHz slower (3%) but $40 cheaper (13%)
RAM: Corsair Vengeance, 4x8GB
Graphics Card: MSI's version of NVIDIA's GeForce GTX 560 Ti (2 GB video RAM)
Hard Drives: 2x 1TB Western Digital (redundancy for data backup - maybe just get 1x and get an external for backup?), 1x Crucial 128GB SSD (boot drive)
Optical Drive: LG BD-R w/ 3D playback
Sound Card: (do I need this?) Creative Sound Blaster SE 7.1
Case: Cooler Master 932 (probably the red one, but maybe the blue one ... only diff I see is 4x USB 3 and 2x USB 2 versus those numbers reversed -- need this size case due to ATX-E motherboard)

Price from NewEgg -- $1780 without tax/shipping.

First, thoughts on those specs? Again, I'm not really sure what I'm doing here. So I won't take any critiques personally . Also, what else would I need other than software, OS, keyboard/mouse? Something tells me I will need a power supply, but I'm not sure how to spec that out other than I see on NVIDIA's website that the graphics card needs 500W and the i7 chip needs 95. so >600W supply is needed.

Thanks!

 

[dakwar]

...
Motherboard: Asus Maximus IV Extreme (supports Turbo Boost, 4 PCIe and 8 SATA, and the LGA 1155)
Processor: Intel i7-2700K (quad-core 3.5 GHz up to 3.9 with Turbo Boost) ... considering 2600 instead which is 100 MHz slower (3%) but $40 cheaper (13%)
RAM: Corsair Vengeance, 4x8GB
Graphics Card: MSI's version of NVIDIA's GeForce GTX 560 Ti (2 GB video RAM)
Hard Drives: 2x 1TB Western Digital (redundancy for data backup - maybe just get 1x and get an external for backup?), 1x Crucial 128GB SSD (boot drive)
Optical Drive: LG BD-R w/ 3D playback
Sound Card: (do I need this?) Creative Sound Blaster SE 7.1
Case: Cooler Master 932 (probably the red one, but maybe the blue one ... only diff I see is 4x USB 3 and 2x USB 2 versus those numbers reversed -- need this size case due to ATX-E motherboard)
...

PSU: Seasonic X760 Gold (760W) is a really good one. I'd recommend you absolutely stick with a gold-rated psu; they have the best available energy conversion efficiency so will save you a few bucks in electricity costs, save the environment just a touch, and, most noticably, will run cooler and quieter.

GPU: good enough. also consider one of the 7000 series ATI cards, or dual-card config.

HDD: sure. i'd suggest external hdd for backup as opposed to two 1tb internal drives.

CPU: the ivy bridge chips will be released quite soon (April 29th?). I'd wait and get one of them. They're faster and more energy efficient (so run cooler and quieter).

Sound: built into the mobo is good enough. no need for a dedicated sound card.

RAM: really depends on ur budget. 32gb is a lot of ram but you sound like you need it.

case: i personally like the Fractal Design Define R3 better (a simpler case but still has lots of space inside) but the 932 is a good one too.

hth.

 

[astrostu]

Thanks, dakwar. I think I have what may be a stupid question: What exactly do the GPUs do for me? I know what they do when doing 3-D modeling. I know what they do in gaming. And I know that they're required to hook a monitor up to the 'puter. But for normal desktop applications (really in this case, Arc[stuff]), would a better GPU really benefit me?

 

[dakwar]

Well, the GPU just renders graphics on the screen at a faster rate. For gaming and 3-d architectural model analysis it would help for sure, yes. But from what little i know about the ISIS software package (i haven't actually used this software package myself) I don't think it would help much; because as far as i can tell, you wouldn't actually be modelling anything in 3D and then moving that 3D model around on the computer screen. All you'll be looking at are 2D images. Anf if all you'll be looking at are 2D images, a more powerful GPU won't actually be of any more help than a cheaper GPU.

That being said, the GPU can SOMETIMES also be used to do raw data calculations: e.g. CUDA in NVIDIA cards. But to use the CUDA feature is not trivial: the software package needs to be specially written to take advantage of CUDA and VERY FEW packages out there as far as I know take advantage of CUDA just because of how involved and finicky CUDA is. If ISIS was CUDA-enabled, you'd see maybe a 5-20x increase in computing performance of your images. But as far as I know, ISIS is not CUDA-enabled.

So with all that in mind, it sounds like your money would be better spent on RAM and a faster CPU than a good GPU. Nowadays the intel CPUs come with integrated graphics capabilities. The ivy bridge CPUs will have a slightly upgraded graphics capability built into them. From the sounds of it, if you're not gaming or 3D modelling, you probably can get away with not buying a dedicated GPU altogether. The intel integrated graphics are good enough for smooth playback of blu ray discs and 1080p files. Or, you could always build your computer without a dedicated GPU, use it for a while to see how you like it, and intall a dedicated GPU into it at that point if you're not happy with its graphics performance.

 

[Demosthenes X]

Whatever you do with a processor, stick to the K-series as you can over-clock them. Looking at Newegg, the price difference seems to be $20 between the 2600K and 2700K... I'd probably spend the extra $20 and just get the best one.

 

[astrostu]

Thanks, that makes sense. I don't think any of the software I use makes use of CUDA. And I really don't game (except for those brief bouts where I'm addicted to Civ, but that's a different issue). So long as the motherboard (or CPU?), then, supports a monitor, I can get by without a graphics card at the beginning and see how it works?

Also, I posted this on another forum. They suggested I could probably get away with significantly downgrading the motherboard. I'm now looking at the GIGABYTE GA-Z68XP-UD3P which supports the LGA 1155, 32 GB RAM, 2 PCIe 2.0, 4 SATA@6GB/s, and lots o' USB with an HDMI out as well. The only thing I DON'T see is that it supports Turbo Boost, which the Asus specifically said it did. Is this a worry? Will they anyway?

----------

Whatever you do with a processor, stick to the K-series as you can over-clock them. Looking at Newegg, the price difference seems to be $20 between the 2600K and 2700K... I'd probably spend the extra $20 and just get the best one.

Thanks for the clarification - didn't know what the "K" meant.

 

[bruinsrme]

I build my son a machine using that 932.

He loves it.

Recommendation is to spend the money on the CPU and RAM, as someone mentioned the K series.
We went with a 700W ps originally. he recently upgraded to 800+ not sure which brand tho.

The PS was replaced due to a couple of shutdowns that occurred during rendering.

 

[dakwar]

If you're going to use onboard graphics, you need a mobo with some sort of port to connect a monitor to. This gigabyte z68xp has hdmi only. If you're okay with that (i.e. if your monitor supports hdmi input, or you're willing to buy an hdmi to dvi converter cable) that yeah this cheaper mobo is absolutely good enough. Em, ... on second thought, this might be good enough for your needs: Asus P8Z68-V LE.

But if you're going to wait for ivy bridge, you might as well also see what sorts of new mobos will be released alongside it.

 

[astrostu]

If you're going to use onboard graphics, you need a mobo with some sort of port to connect a monitor to. This gigabyte z68xp has hdmi only. If you're okay with that (i.e. if your monitor supports hdmi input, or you're willing to buy an hdmi to dvi converter cable) that yeah this cheaper mobo is absolutely good enough. Em, ... on second thought, this might be good enough for your needs: Asus P8Z68-V LE.

But if you're going to wait for ivy bridge, you might as well also see what sorts of new mobos will be released alongside it.

The spare monitor I have is HDMI/DVI/VGA. So, yeah. And to clarify -- if I were to "outgrow" that motherboard, I can always get a new one and swap it out, right? Like, if I end up with 3 USB 3.0 things to connect?

 

[astrostu]

Just got this in from the advisor who teaches ArcGIS: "All I know is that when I run arc on machines with crappy graphics cards the draw/refresh rate sucks."

 

[wywern209]

Just got this in from the advisor who teaches ArcGIS: "All I know is that when I run arc on machines with crappy graphics cards the draw/refresh rate sucks."

then you will need a faster nice graphics card because of the size of images you will be working with. 39mpix and such sounds like big files to me. i would get a quadro with lots of vram.

 

[astrostu]

then you will need a faster nice graphics card because of the size of images you will be working with. 39mpix and such sounds like big files to me. i would get a quadro with lots of vram.

I'm going to post this to the ArcGIS forum and see what they say about the GPU. Most of the basemap images I'm using are 4 GB.

 

[astrostu]

The guy who manages one of the pieces of software I want this for (the crater detection stuff) said that the ability to actually use the DDR3-1600 RAM would significantly help speed up the code. The Intel i7-2600K only supports 1333 MHz RAM. The Ivy Bridge chips, which are set for the end of the month release, will support 1600. The i7-3770K, while top-of-the line when it'll be released, is still set for an MSRP of just above the i7-2600K ($332). So now I'm thinking that should be my chip.

 

[SporkLover]

.

HDD: sure. i'd suggest external hdd for backup as opposed to two 1tb internal drives.

I actually prefer internal drives for backup. My experience with external is that the drives inside aren't usually meant to be used for long extended periods.... And when you do find an external setup that uses good drives you are paying out the nose for them.

Sooooo, I go with MOBOs with on board raid controllers and setup mirror array...... or Striping/mirror array for gaming rigs. when/if a hard drive fails you don't lose usage of your computer; you put in a replacement drive and repair the array. With this being a work machine, I would definitely recommend this route.

 

[velocityg4]

The guy who manages one of the pieces of software I want this for (the crater detection stuff) said that the ability to actually use the DDR3-1600 RAM would significantly help speed up the code. The Intel i7-2600K only supports 1333 MHz RAM. The Ivy Bridge chips, which are set for the end of the month release, will support 1600. The i7-3770K, while top-of-the line when it'll be released, is still set for an MSRP of just above the i7-2600K ($332). So now I'm thinking that should be my chip.

In that case you may want to consider building with an LGA 2011 board. You can use quad channel memory kits instead of dual channel for a much bigger boost. With the size of the data files you are using I would consider larger hard drive as well. Possibly a RAID 5 array for speed, capacity and redundancy. Skip the sound card. Unless you are an audiophile or sound engineer with a $5000+ sound system, onboard is far more than enough.

For about $1850 you could do:
i7 3820 3.6Ghz 4 core $310
8GBx4 Corsair RAM $240
Rosewill 650W 80plus gold certified PSU $100
GeForce GTX 560 (2GB) $190
Gigabyte GA-X79-UD3 Motherboard $240
Windows 7 Professional OEM $140
3x1TB Seagate Barracuda 7,200 RPM $300
128GB Crucial M4 SSD $150
Lite On Blu Ray $60
Nice Case $100

All through Newegg.com

For an extra $290 you can get a 6 core i7 3930K 3.2Ghz. If what you are doing taxes the CPU then I would definitely consider the upgrade. Especially since it can be overclocked. For video cards I would avoid the Radeon 7000 series as they do not do well on GPGPU tasks (hindered by AMD on purpose to sell Fire cards). Though it does not sound like you need an uber expensive workstation card anyways. Which you could add later on if needed.

Considering that you are using this to make money. If you will make more money with a faster computer by getting more work done you shouldn't quibble over a couple hundred dollars. Especially since the software you are using sounds quite expensive, a few hundred becomes a small percentage of the whole. As it stands $70 more for a superior system is quite small, $360 for an even bigger boost is only 20% more. These price differences are assuming you originally factored the cost of Windows. You may even consider a second SSD for Linux.

If you think you may need more memory in the future you may consider the ASUS Sabertooth X79 for an additional $70. It has 8 RAM slots supporting 64GB.

 

[astrostu]

Okay, so now I'm trying to understand LGA 1155 versus 2011 and future support/expandability. From what I see, most of the i7 chips now are 1155 with the very high-end being 2011. All of the 1155 that I see on the Ivy Bridge roadmap are 1155. So I was thinking that's what I should go with.

Now I'm wondering if, in reality, the Ivy Bridge first offerings are 1155, but that future ones will be 2011? It looks like the Sandy Bridge chips up until last November were all 1155 and then came out the chips that were LGA 2011.

Also, on quad vs. dual channel RAM. I didn't think to look at this. Does the RAM have to be any different? And is it basically like multi-threading the RAM so it'll access faster?

In the benchmarks I looked at, there doesn't seem to be too much of a difference between the 3820 and the 2600K except in a very few specific applications, and it is $40 cheaper and apparently very over-clock-able. Motherboards are more expensive, though, but I could conceivably get one that goes up to 64GB RAM if I went with the LGA 2011.

 

[astrostu]

Alright, I'm getting the LG drive now 'cause I could use it in my Mac even if I don't do this. But NewEgg's $10 off (14%) ends tonight.

 

[throAU]

The major bottlenecks in most PCs that are common across most applications are RAM and IO throughput. 16gb (or in a desktop, 32gb) of ram is cheap (compared to total build cost), no point getting less these days...

I'd go for a core i7 of current generation (if you can wait a little while, wait for ivy bridge) to get the full CPU feature set, clock speed will depend on your budget - be aware that there is diminishing returns in terms of $/performance. A middle of the road i7 will likely get 85-90% of the total system performance of a top of the line i7 at much lower cost...

Does your application need 3d video? If not, blu-ray will work with most onboard video these days.

If you can find a motherboard that will do SSD caching (Z68 chipset), a small SSD (e.g., 60gb) to use for disk cache will help your IO a lot.


edit:
sounds like your app may need 3d video then. i'd contact the software vendor directly, and ask if they have a hardware compatibility list for the app. many 3d productivity apps have a recommended/certified hardware list

 

[astrostu]

The major bottlenecks in most PCs that are common across most applications are RAM and IO throughput. 16gb (or in a desktop, 32gb) of ram is cheap (compared to total build cost), no point getting less these days...

I'd go for a core i7 of current generation (if you can wait a little while, wait for ivy bridge) to get the full CPU feature set, clock speed will depend on your budget - be aware that there is diminishing returns in terms of $/performance. A middle of the road i7 will likely get 85-90% of the total system performance of a top of the line i7 at much lower cost...

Does your application need 3d video? If not, blu-ray will work with most onboard video these days.

If you can find a motherboard that will do SSD caching (Z68 chipset), a small SSD (e.g., 60gb) to use for disk cache will help your IO a lot.


edit:
sounds like your app may need 3d video then. i'd contact the software vendor directly, and ask if they have a hardware compatibility list for the app. many 3d productivity apps have a recommended/certified hardware list

As I mentioned (and you edited), I'm looking to run three programs specifically, but I want expandability/upgradability as these datasets grow. And I just remembered a fourth, actually. I have one application that needs to rapidly be able to display over 500 million datapoints, and on my laptop it takes several seconds each time, which is as much time as it takes me to perform the task, so the time taken is doubled which means this task will go from taking 6 months to 1 year.

I've contacted the guy who maintains one bit of code and he says it's all about RAM and access speed. The other command-line software is HD read/write and CPU (projecting Gpx images) and RAM to a much lesser extent. I've contacted the main software vender for the other, visual application, and am waiting. I posted this on another board, too, and happened to get a guy who responded who specs out machines for ArcGIS stuff ... he said that the graphics card definitely will be taken advantage of, but I clarified my use of that application (as in NOT using the 3D analyst parts) and am waiting for a response.

And now that I just remembered the fourth bit, I'm going to e-mail them and see what they say regarding GPU usage with displaying data, too.

 

[throAU]

Definitely look into SSD caching then. The Z68 chipset does it, and I would hazard a guess that the new Z77 chipset (just released in advance of ivy bridge) will too.

I'd also suggest going for more drives - not for space, but for IO throughput.

Rather than 2x 1tb drives, consider 4x500gb, or even 4x1tb to be able to run RAID10 instead of RAID1. Or maybe even 4x256gb SSD in RAID 0 and a 1tb SATA backup drive?


See if you can get an idea on whether your app is going to be CPU bound or IO bound. From the sounds of it (heavy HD usage) it may be IO bound, in which case throwing more CPU at the problem may not be any use if the IO can't keep up...

 

[velocityg4]

Okay, so now I'm trying to understand LGA 1155 versus 2011 and future support/expandability. From what I see, most of the i7 chips now are 1155 with the very high-end being 2011. All of the 1155 that I see on the Ivy Bridge roadmap are 1155. So I was thinking that's what I should go with.

Now I'm wondering if, in reality, the Ivy Bridge first offerings are 1155, but that future ones will be 2011? It looks like the Sandy Bridge chips up until last November were all 1155 and then came out the chips that were LGA 2011.

Also, on quad vs. dual channel RAM. I didn't think to look at this. Does the RAM have to be any different? And is it basically like multi-threading the RAM so it'll access faster?

In the benchmarks I looked at, there doesn't seem to be too much of a difference between the 3820 and the 2600K except in a very few specific applications, and it is $40 cheaper and apparently very over-clock-able. Motherboards are more expensive, though, but I could conceivably get one that goes up to 64GB RAM if I went with the LGA 2011.

From what you have been saying it sounds like your uses are those specific circumstances where the 3820 beats the 2600. Especially on the memory throughput front as the quad channel creams the dual channel. The quad channel is just four matching sticks. Only LGA 2011 boards can take advantage of this. The 6 core 3930K would definitely trounce the other CPU's.

LGA 1155 and 2011 are two different lines. LGA 1155 is consumer grade costs are cut by not offering the level of throughput that LGA 2011 allows. This entails fewer PCI-e lanes and memory channels. There are more differences but these are the main points. LGA 1155 is the successor to LGA 1156 and LGA 2011 is the successor to LGA 1366.

LGA 2011 also allows for multiple CPU sockets and Xeon CPU's. If you wanted you could get a dual socket LGA 2011 board and use two quad, hexa or octo core Xeon CPU's with 16 RAM slots and support for up to 512GB RAM.

 

[astrostu]

@throAU - I think I/O will be the main bottleneck, but there also comes a point where this stuff requires a lot of CPU power. I mean, manipulating several gigapixels and doing math on every single 32-bit pixel takes awhile. I'll look into the RAID idea. As for the chipset, based on what just came out from NewEgg last night, the Z77 are only LGA 1155. I think this makes sense given the new Intel chips coming out are only LGA 1155 ... for now.

@velocityg4 - Okay, good. I can read basic benchmarks, but the ones I found were like 50 different measurements and I was just lost. And it looks like I could do the 3820 chip + different motherboard for it to take advantage of the quad channel throughput for just about $100 more. It sounds like it's really not just the specs and number of ports I need to pay attention to, but some of these other details ... this is why I love these forums. But let's not go crä-Z with cores and RAM. I am on a budget at the moment.

Would I also be correct in my guessing, then, that Ivy Bridge LGA 2011 chips will be forthcoming, but in the meantime, overclocking the 3820 should give me comparable or better response than the new Ivy Bridge due to the memory throughput?

 

[throAU]

Would I also be correct in my guessing, then, that Ivy Bridge LGA 2011 chips will be forthcoming, but in the meantime, overclocking the 3820 should give me comparable or better response than the new Ivy Bridge due to the memory throughput?

If it is for work, and you actually care about the calculation results: don't overclock.

Especially given you are talking about 6 months+ of work?

Overclocking is fine for gaming or stuff you don't really care too much about.

However, once you start working outside of spec, there are no guarantees that any calculations you do are correct. Do you want to spend a further 6-12 months re-running your job to ensure you get the same result back?

An OC'd CPU may run fine/stable, may run games and benchmarks just fine (perhaps with an occasional un-noticable graphical glitch, etc), but if, for example there's a heat/clock related issue in a part of the CPU that isn't directly involved in memory access (e.g., SSE register, floating point unit, etc) - it may run perfectly "stable" (i.e., not crash, etc) while giving dodgy calculation results.


If you can't determine whether the app is IO or CPU bound, I'd suggest putting a strong IO subsystem in (RAID10, SSD, etc), plenty of RAM (it is cheap), on a good quality motherboard, and go for a low/mid CPU of the current (i.e. ivy bridge, or sandy bridge) generation - that will fit on the current socket standard - and upgrade CPU if required later.


I guess another question you need to ask is how well threaded is your app?

If it can scale well with thread/core count - you should be looking at Xeons IMHO. Sure, a core i7 may beat a single Xeon head-to-head on gaming benchmarks, etc - but Xeons can be paired and you can get double the core count or more. i7s can't.

The other xeon advantage is ECC memory. If your job is going to take MONTHS and is business related, then I'd consider ECC memory to be important, to ensure that no bit-errors are incurred in (thus corrupting/invalidating) your results.

I'm not an expert on xeon motherboards/chipsets/etc (we buy pre-built workstations for that here), but it is certainly something to consider.

Yup, xeons are a bit more expensive, but there are reasons - and by the sounds of it they may be relevant to your intended purpose...

 

[astrostu]

If it is for work, and you actually care about the calculation results: don't overclock. Especially given you are talking about 6 months+ of work? Overclocking is fine for gaming or stuff you don't really care too much about.

However, once you start working outside of spec, there are no guarantees that any calculations you do are correct. Do you want to spend a further 6-12 months re-running your job to ensure you get the same result back?

An OC'd CPU may run fine/stable, may run games and benchmarks just fine (perhaps with an occasional un-noticable graphical glitch, etc), but if, for example there's a heat/clock related issue in a part of the CPU that isn't directly involved in memory access (e.g., SSE register, floating point unit, etc) - it may run perfectly "stable" (i.e., not crash, etc) while giving dodgy calculation results.

Talking about several years of work to do on this machine, though with the type of stuff I'm doing, I'd know right away if something was going wrong. But, point taken -- I won't worry about overclocking. But it is nice to know that it'd be available.

If you can't determine whether the app is IO or CPU bound, I'd suggest putting a strong IO subsystem in (RAID10, SSD, etc), plenty of RAM (it is cheap), on a good quality motherboard, and go for a low/mid CPU of the current (i.e. ivy bridge, or sandy bridge) generation - that will fit on the current socket standard - and upgrade CPU if required later.

I'm definitely going for ≥32 GB of RAM. If I end up getting a motherboard that supports it, I'll do 64 GB.

Drives are another issue. They get expensive quickly. First, if I go with something like the ASRock X79 Extreme6/GB motherboard, the specs say it supports 4 SATA@3Gb/s drives and 5 @6Gb/s. SATA RAID support is 2 SATA3, and 4 SATA2. My reading of RAID 10 is that it needs 4 drives. If I still want to do "one" SSD for OS and software, and "one" HDD for data, then a RAID 10 on the data would need to be relegated to the SATA2 (3 Gb/s) connectors, and I could only do a RAID 0 or 1 with the SSDs on the OS/software drives. And this would require 6 hard drives ... if I went with the planned 1TB WD drives for data and the 1/8TB Crucial drives for OS/software, we're talking $860 for this, which is around 45% of what I want to spend on the whole system. I'm not trying to discount your advice -- if I had a grant to pay for this, I'd totally do it and yell at the purchasing people until they agreed with me.

I guess another question you need to ask is how well threaded is your app?

If it can scale well with thread/core count - you should be looking at Xeons IMHO. Sure, a core i7 may beat a single Xeon head-to-head on gaming benchmarks, etc - but Xeons can be paired and you can get double the core count or more. i7s can't.

ArcGIS claims their software is multithreaded, but for what I do in it, it is not.

Igor's main interface is not multithreaded, but much of the code I've written within it is. But half of the time spent is waiting for it to draw to the screen, so it's a read from RAM and CPU issue (hence maxing out those ...).

ISIS is not at all multithreaded, but every year for the past three years at a conference I go to I remind them that they need to be working on it. After all, ISIS is image processing, projecting pixels, and that task was what multithreading was MADE for. What I can do is have multiple instances of it running on different image sets, sorta the poo' man's multithreading.

And this crater detection software was written in the late 90s. It's maintained, but no real work is being done on it. When I've wanted to "multithread," I've done what I do in ISIS. Problem is that it's this software that's the RAM hog. I want to analyze gigapixel images with it, but on a simple "small" 40 megapixel image, it soaks up 9 GB of RAM. Hence why I want to max out the RAM on this machine.

The other xeon advantage is ECC memory. If your job is going to take MONTHS and is business related, then I'd consider ECC memory to be important, to ensure that no bit-errors are incurred in (thus corrupting/invalidating) your results.

I'm not an expert on xeon motherboards/chipsets/etc (we buy pre-built workstations for that here), but it is certainly something to consider.

Yup, xeons are a bit more expensive, but there are reasons - and by the sounds of it they may be relevant to your intended purpose...

The job will take months (years, really), but it's not like I'm setting up a single rendering job or simulation to go for a few months. It's more I'm analyzing craters, and there are hundreds of thousands of them that need to be done interactively and individually (ArcGIS, Igor, crater detection); or I'm processing images to do analysis on (ISIS). So I think I'm okay with not getting into ECC memory.

 

[throAU]

with regards to storage: obviously cost is a factor - just throwing options out there...

I'd certainly look into the new intel chipsets though; the ability to use a small (<60gb) SSD for caching on a RAID array (of spinning disks) will be a big win - for not a massive cost. I'd certainly consider that rather than using a small SSD for boot/OS - you likely reboot once every couple of weeks, who cares how long boot takes? Caching your application data will be a bigger win by far.


I wouldn't be overly concerned about using SATA2 ports for spinning disks, but if you need more SATA3 ports you can add them via a PCI-e card - pretty sure they're reasonably cheap.


All this does depend on the read/write characteristics of your app - if it is largely streaming sequential IO then SSD won't be so much of a win, but if it is a fairly random access pattern with small transfer sizes, irrespective of SATA3 or SATA2, SSD (even for cache) will be a big win due to the lack of head movement or rotational latency influencing seek time - if you throw random 4k reads at a spinning disk, performance goes out the window (say, 75 IOs per sec for a typical 7200 rpm SATA disk is typical - i.e., 300 KB/sec worst case transfer rate for random IO). With no moving parts, SSD can handle say 10,000-40,000 4k IOs per second....

for random IO, the SSD is so far ahead it isn't funny.
for sequential - not so much...