Need help building a computer for statistical processing

star987

Commendable
Feb 2, 2017
31
0
1,530
I am currently in the process of trying to build a desktop to run statistical analysis with SAS. I typically use multi GB files and have been struggling with my current machine. The proposed machine is listed below but I really don't have any experience building a computer so please provide any help with what I've laid out...especially with the motherboard.

CPU - i7-7700K 4.20GHZ 8MB LGA 1151 Kaby Lake

Hard Drive - 480GB HyperX Savage SATA-III 6.0Gb/s (x2 in a Raid 0)

Motherboard - ASUS Prime Z270-A w/RGB, USB 3.1, 3 PCIe x16, 4PCIe x1, 6 SATA3, 2 M.2 SATA/PCIe

RAM - 32GB (8GBx4) 3000MHz Dual Channel Memory

Video Card - GeForce GT 730 2GB GDDR3

Thanks!
 
Solution
Something like this with little compromise. You don't need SATA SSDs in RAID, you can get much greater performance from PCIe NVMe SSDs (3200MB/s vs the 800-900 or so with striped SSDs)

PCPartPicker part list / Price breakdown by merchant

CPU: Intel Core i7-6800K 3.4GHz 6-Core Processor ($409.78 @ OutletPC)
CPU Cooler: CRYORIG H5 Ultimate 76.0 CFM CPU Cooler ($56.49 @ OutletPC)
Motherboard: ASRock X99 Extreme3 ATX LGA2011-3 Motherboard ($159.99 @ SuperBiiz)
Memory: G.Skill Ripjaws Series 32GB (4 x 8GB) DDR4-2400 Memory ($199.99 @ Newegg)
Storage: Samsung 960 Evo 500GB M.2-2280 Solid State Drive ($249.99 @ B&H)
Storage: Western...

star987

Commendable
Feb 2, 2017
31
0
1,530
A laptop with the following:

CPU -i5 - 5300 @ 2.30Ghz

Hard Drive - Unknown brand SSD not Raid 0.

Motherboard - Unknown

RAM - 8GB

Video Card - Unknown

Obviously much is unknown to me. This is a Dell laptop provided to me by my work and it is not even close to being a sufficient machine for my needs.

Regarding my proposed system, I am also curious about CPUs (AMD vs. Intel and i7 vs. Xeon E7).

Thanks
 

Eximo

Titan
Ambassador
For professional number crunching you should use the X99 platform.

I guess we need to know some data though. Have you monitored performance while running your analysis? What is getting consumed the most, memory or CPU? It would be a shame to find out it is a single threaded application and waste a lot of money on hyperthreading or a hex+ core CPU. Any chance of GPU based analysis? Might take a cheap CPU and a couple of GPUs to get you a much faster result.

At work we use SAS JMP, but our standard workstation approaches $8000 in cost, so they get what they get. (8-core Xeon, Ivy Bridge-E on the older ones, Nvidia Quadro, 64GB of memory) (And we have a step up from that)
 

star987

Commendable
Feb 2, 2017
31
0
1,530
I looked that post through and there is no way I have the knowledge to build something with that much detail going into it. I was hoping for a simple (if there is such a thing) suggestion on the proposed computer I suggested and whether or not I'm wasting my time with how I have it set up.

If its possible to make those suggestions given what I have provided I would greatly appreciate it, if not that is fine as well.
 

star987

Commendable
Feb 2, 2017
31
0
1,530


Could you elaborate on the X99 platform? I'll look into it more but am not sure what exactly that means.
 

Eximo

Titan
Ambassador
Certainly faster than a laptop. But without knowing how the software will interact with the computer, no way to know.

I recall one application that actually performed worse the more CPU cores were thrown at the problem. Ideally that software needed the most powerful single core it could get. Each time another core was added it would do half the work. (never did figure that one out, went back to the vendor and everything)
 

ulises314

Reputable
May 25, 2016
87
0
4,660
You'll see about 150% speed increase in the processor alone, plus you'll have faster RAM and much faster storage (which will impact a lot your big data set work); the only part that doesn't convince me is the GPU, it's worst than the included graphics in the 7700k and even the VRAM is slower than system ram, so even if it's cheap is just money thrown away, since SAS can't use GPUs to process data you are better off with the included graphics.
 

Eximo

Titan
Ambassador
Intel has two basic classes of hardware. Consumer cpus with a max TDP of around 90W, quad core, hyperthreading. And Workstation/Server chips, more specifically the HEDT line-up. TDP up to 150W and up to 22? cores, Would have to look it up. Those are multi-thousand dollar CPUs.

But for about $150 more than an i7-7700k you can get something like the i7-6800k, which has 6 cores instead of 4, and has nearly the same per-core performance. Also quad channel memory instead of dual.

And you can go up from there. 128GB/256GB of memory is possible, ECC memory instead of unbuffered. And as mentioned, lots and lots of cores.

When time=money and accuracy is necessary, these are the systems that professionals use.
 

star987

Commendable
Feb 2, 2017
31
0
1,530


I'm using SAS for statistical analysis. I'm not sure how to determine an issue related to cores and how the software interacts with the computer.
 

star987

Commendable
Feb 2, 2017
31
0
1,530


That is exactly what I'm using and my budget is <= $1800.
 

Eximo

Titan
Ambassador


Simple performance monitor in Windows will get you basic information. Watch the CPU cores, see if one is completely loaded, watch the memory see if it is maxed out. If your dataset doesn't exceed a few GB then getting 32GB won't help. If it is all CPU processing power, then we need to know if more cores or hyperthreading are an advantage or disadvantage.

I'm aware of SAS as a company, but they have a wide offering for many types of analysis.
 

ulises314

Reputable
May 25, 2016
87
0
4,660




SAS is multithreaded but it doesn't use the GPU, similar free software (R) can take advantange of the GPU, but AFAIK this is not what is going to be used.
 

star987

Commendable
Feb 2, 2017
31
0
1,530
So, the SAS I use is the basic version of their software where I have a blank white screen and I type my code. To the best of my knowledge it is simply called SAS 9.4. Given my clear lack of understanding in this area, is it possible to get recommendations of what to put into a computer given my needs and budget?
 

Eximo

Titan
Ambassador
Something like this with little compromise. You don't need SATA SSDs in RAID, you can get much greater performance from PCIe NVMe SSDs (3200MB/s vs the 800-900 or so with striped SSDs)

PCPartPicker part list / Price breakdown by merchant

CPU: Intel Core i7-6800K 3.4GHz 6-Core Processor ($409.78 @ OutletPC)
CPU Cooler: CRYORIG H5 Ultimate 76.0 CFM CPU Cooler ($56.49 @ OutletPC)
Motherboard: ASRock X99 Extreme3 ATX LGA2011-3 Motherboard ($159.99 @ SuperBiiz)
Memory: G.Skill Ripjaws Series 32GB (4 x 8GB) DDR4-2400 Memory ($199.99 @ Newegg)
Storage: Samsung 960 Evo 500GB M.2-2280 Solid State Drive ($249.99 @ B&H)
Storage: Western Digital Caviar Blue 1TB 3.5" 7200RPM Internal Hard Drive ($49.33 @ OutletPC)
Video Card: MSI Radeon R7 240 2GB Video Card ($51.89 @ OutletPC)
Case: Fractal Design Core 2300 ATX Mid Tower Case ($49.99 @ NCIX US)
Power Supply: EVGA SuperNOVA G3 550W 80+ Gold Certified Fully-Modular ATX Power Supply ($77.99 @ SuperBiiz)
Operating System: Microsoft Windows 10 Pro OEM 64-bit ($100.17 @ Amazon)
Total: $1405.61
Prices include shipping, taxes, and discounts when available
Generated by PCPartPicker 2017-02-02 13:36 EST-0500
 
Solution

Eximo

Titan
Ambassador
X99 CPUs don't have onboard graphics, so you do need a crappy video card. You can always bump that up to a decent card at a later date if you start using something that can take advantage of GPU computing.

Windows is optional.
 

star987

Commendable
Feb 2, 2017
31
0
1,530


They have just recently released a version supported by Windows 10, I've confirmed this to be the case directly with SAS.
 

Eximo

Titan
Ambassador


Oh they say lots of things...

Pretty much what I do for a living is proving companies like them wrong. Don't have to use Windows either, if he's using the one I think he is, then it will run on any flavor of Linux/Unix as well.

Actually we have a lot of problems with SAS having buggy releases too. Works for group A, but some critical function group B uses doesn't work and we end up holding off.
 

Lutfij

Titan
Moderator
Then that is good sign. I only hope it's reliable. I agree with ulises314+Eximo that you're going to need something outside of the Kabylake platform but I'd side with a Xeon simply because of more cores. If you can afford it that is.

Eximo,
yeah I'm one of those advocates. You can get better results if you know what you're doing on Linux.
 

star987

Commendable
Feb 2, 2017
31
0
1,530


Any reason to upgrade the speed of my ram? Any thoughts of Lutfij's comment regarding Xeon vs. the i7?