Best configuration for data science using R or Python

[hfahmy]

1. Budget & Location

Budget is roughly 2,000 USD (can exceed somewhat if necessary).

 

2. Aim

System will be used form mining, crunching, visualising and modelling large datasets. The software that I'm using is R with RStudio. As per my understanding, R doesn't rely on multicore to do its heavy work and doesn't look like thats going to change in the future. I might in the future move to Python, which doesn't out-of-the box rely on multicore as well. Currently I also use MS Excel every now and then to open  large data dumps and make small tweaks here and there.

 

3. Monitors

Two 24-inch LED monitors.

 

4. Peripherals

System will be running Windows 10, no other special peripherals will be used as of now.

 

5. Current Setup

At the moment, I'm working on a laptop with an Intel i5 (not sure about L3 cache size) with 8GB RAM and a 256 SSD. With this setup, the main bottlenecks in my workflow are mostly in having to wait for the data to load RAM and running some extensive calculations.

 

I'm assuming that upgrading to an i7 with more cores would not yield much efficiency as R doesn't rely on multicore processing. Should I aim at increasing the CPU clock speed and increasing L3 cache size to make data transfer between RAM and processor more efficient?

 

And any other specification that I should focus on to increase the efficiency?

[CostcoSamples]

Even if you can't make use of multiple cores, or even hyper threading, the i7 6700k still has the highest clock speed, likely making it the fastest for single thread processing.  But for the biggest cache size you're looking at the x99 platform such as i7 6800 (15 mb cache).  AMD CPUs tend to have larger cache sizes, but their single threaded processing is not great.

 

Sounds like you need a lot of RAM though - would 64 GB be suitable?

 

Would you benefit from a very fast PCIE SSD such as Samsung 950 pro or Intel 750 series?

[CostcoSamples]

Here is a first crack at a build.  I'm assuming you don't need a GPU.  Monitors, OS, and peripherals not included.  $1,315

 

http://pcpartpicker.com/list/PX3mkT

 

The integrated Intel graphics are fine for general office use, but I've never tried it for multi monitor setup.

[hfahmy]

On 8/12/2016 at 9:42 PM, CostcoSamples said:

Even if you can't make use of multiple cores, or even hyper threading, the i7 6700k still has the highest clock speed, likely making it the fastest for single thread processing.  But for the biggest cache size you're looking at the x99 platform such as i7 6800 (15 mb cache).  AMD CPUs tend to have larger cache sizes, but their single threaded processing is not great.

 

Sounds like you need a lot of RAM though - would 64 GB be suitable?

 

Would you benefit from a very fast PCIE SSD such as Samsung 950 pro or Intel 750 series?

Yeah, I think this is where I should focus most. Thanks will have a look.

[Xineas]

47 minutes ago, hfahmy said:

Yeah, I think this is where I should focus most. Thanks will have a look.

My two cents. I went with a relatively quiet case. Put in a good motherboard with the usual display connectors you need for running 2 screens off of the integrated graphics. Put in a CPU with a high clock speed to begin with. Pair it with the 280mm AIO cooler, put the 4 fans in a push-pull configuration, hook those up to the fan splitter and overclock the CPU to something stable, 4.8GHz should be possible. Intake the air from the front with the radiator there, exhaust out the back, so there's some airflow through the case to make sure the other components on the motherboard get some fresh air too. 64GB of RAM is enough for any application you can think of and coincidentally the max supported by this motherboard. The 950 Pro is fast as hell and 512GB should be enough for most things, you can always drop an extra drive in.

 

PCPartPicker part list: http://pcpartpicker.com/list/DdcBcc

 

CPU: Intel Core i7-6700K 4.0GHz Quad-Core Processor  ($319.99 @ Newegg) 
CPU Cooler: Corsair H110i GTX 104.7 CFM Liquid CPU Cooler  ($99.99 @ Newegg) 
Motherboard: Asus Z170-A ATX LGA1151 Motherboard  ($137.98 @ Newegg) 
Memory: Corsair Vengeance LPX 64GB (4 x 16GB) DDR4-3000 Memory  ($373.96 @ Amazon) 
Storage: Samsung 950 PRO 512GB M.2-2280 Solid State Drive  ($317.99 @ B&H) 
Case: Corsair Carbide 400Q ATX Mid Tower Case  ($99.99 @ Newegg) 
Power Supply: Corsair 750W 80+ Platinum Certified Fully-Modular ATX Power Supply  ($134.99 @ Newegg) 
Operating System: Microsoft Windows 10 Home OEM 64-bit  ($89.00 @ B&H) 
Case Fan: Corsair Air Series AF140 Quiet Edition 67.8 CFM  140mm Fan  ($17.50 @ OutletPC) 
Case Fan: Corsair Air Series AF140 Quiet Edition 67.8 CFM  140mm Fan  ($17.50 @ OutletPC) 
Case Fan: Corsair Air Series AF140 Quiet Edition 67.8 CFM  140mm Fan  ($17.50 @ OutletPC) 
Case Fan: Corsair Air Series AF140 Quiet Edition 67.8 CFM  140mm Fan  ($17.50 @ OutletPC) 
Monitor: Acer GN246HL 24.0" 144Hz Monitor  ($199.00 @ B&H) 
Monitor: Acer GN246HL 24.0" 144Hz Monitor  ($199.00 @ B&H) 
Other: Fan Splitter ($12.49)

Total: $2046.38

 

Prices include shipping, taxes, and discounts when available
Generated by PCPartPicker 2016-08-13 14:41 EDT-0400

[hfahmy]

Thanks @Xineas, this is configuration is almost what I'm look for. Can you suggest the changes that would be required to this config if I want to upgrade the CPU to Intel® Core™ i7-6800K Processor (15M Cache)?

[pyrojoe34]

R loves RAM, I eat through my 32GB and a 50GB page file when running large datasets and random forests. You're right that R isn't inherently multi-threaded but there are plenty of scripts you can use to force multi-threading (dosnow, parallel, etc). I would look into learning to use those as they will save you a lot of time. Since RAM is relatively cheap I would get at least 32GB and better yet, consider getting 64GB. If you go with an x99 system like I did you can get a quad-channel kit (for faster bandwidth) and still have room to upgrade in the future if you want to add more.

 

The other thing to favor is drives. Get some fast SSDs (multiple is better since you can optimize data flow to never read and write to the same drive at the same time, this will help with loading things to RAM faster). Maybe one M.2 for OS/programs/page file and one or two SATA SSDs as working drives (one reserved for cache, and one for the data). Then you can always get a slower large HDD for storing the data when it's not being used.

 

Are you going to use any CUDA accelerated scripts/tools?

 

I think a 6800k (or 5820k) is a good choice (make sure get get the most out of single core speeds by doing some mild OCing, not too much though since you don't want errors to pop up in executing your scripts) paired with 64GB of quad-channel RAM (maybe like 3000Mhz or so), an M.2 SSD, 2 SATA SSDs, and a large HDD. I'm not sure why the guy above recommended 144hz monitors since (unless you're gaming) that's completely useless for your work. Stick with cheaper monitors and put the money into other things. You'll also likely want Win 10 Pro, not home. I got a key on Kinguin for like $30. Only get a good GPU if you're sure you'll use CUDA or if you also want to game, otherwise get a lower end one and save the money.

[pyrojoe34]

5 hours ago, hfahmy said:

Thanks @Xineas, this is configuration is almost what I'm look for. Can you suggest the changes that would be required to this config if I want to upgrade the CPU to Intel® Core™ i7-6800K Processor (15M Cache)?

Here you go:

PCPartPicker part list: http://pcpartpicker.com/list/HMfh9W
Price breakdown by merchant: http://pcpartpicker.com/list/HMfh9W/by_merchant/

CPU: Intel Core i7-6800K 3.4GHz 6-Core Processor  ($428.99 @ SuperBiiz)
CPU Cooler: Corsair H110i GTX 104.7 CFM Liquid CPU Cooler  ($120.00)
Motherboard: Asus X99-A II ATX LGA2011-3 Motherboard  ($229.99 @ SuperBiiz)
Memory: Corsair Vengeance LPX 64GB (4 x 16GB) DDR4-3000 Memory
Storage: PNY CS1311 240GB 2.5" Solid State Drive  ($69.99 @ B&H)
Storage: PNY CS1311 240GB 2.5" Solid State Drive  ($69.99 @ B&H)
Storage: Samsung 950 PRO 512GB M.2-2280 Solid State Drive  ($315.62 @ B&H)
Case: Corsair Carbide 400Q ATX Mid Tower Case  ($99.99 @ Newegg)
Power Supply: Corsair RMx 750W 80+ Gold Certified Fully-Modular ATX Power Supply  ($104.99 @ Newegg)
Other: Windows 10 Pro OEM ($30.00)
Total: $1469.56
Prices include shipping, taxes, and discounts when available
Generated by PCPartPicker 2016-10-11 06:17 EDT-0400

 

 

I left out a few things here:

 

GPU: with x99 you have to buy one but depending on your intention that can vary greatly. Either you won't use CUDA and can grab one for $50, or you will use CUDA and may want to dish out a bit more ($100-300). Let me know your intentions and I can try to recommend something.

 

Monitors: dual monitors are very nice, and for coding, one that swivels 90deg to vertical can be very nice. Don't bother with high refresh rates like the guy above listed, it's useless unless you're gaming. Save your money for other parts or to get a larger screen at higher resolution (1440p or 4k) which can very very nice when looking at a lot of text. I use a 43" 4k TV that is actually really nice to code on and gives lots of screen space. Maybe consider 2 27" 1440p monitors? If you're set on 2x 24" monitors then just get some regular 1080p 60Hz monitors and call it a day.

 

HDD: if you start to notice you need more disk space for long term storage grab a 3-5TB HDD for $100-150 in the future.

 

 

Changes from the other guy:

CPU/mobo: x99 system for quad-channel and more cores, slower single core performance but if you code efficiently you will be better off. To be honest though, most programs/code will not use all this power, the only reason I use that much is because I run machine learning algorithms on massive datasets (tens to hundreds of millions of data points with very large forests and many nodes). Even then, most of the script barely uses 10% of the CPU. It's mostly cross validation and training that takes more power.

Fans: Don't waste your money unless you're a silence freak, the case and cooler will have more than enough already.

PSU: Platinum PSU is a waste of money, For $50 less you get the gold which will do everything exactly the same at 1-2% less efficiency (not noticeable unless you're running a data center with 100s of computers)

OS: Get Win 10 Pro instead of Home, you may end up wanting/needing the extra features as a programmer (for example, I like to use remote desktop to work off my desktop from my laptop if I'm not at home). Also get it from someone like kinguin for $30 (I did that and had no problems)

Cooler: The H110i GTX should not be $200, They are usually $120 and I got mine for $99. Wait for the price to go back to normal or get a different cooler.

 

[hfahmy]

@pyrojoe34, that spec is better for suited for my needs. I won't be using CUDA soon, may be sometime in the future. But I'll need 2 x HDMI out at least. So, what would you suggest for a GPU?

[pyrojoe34]

45 minutes ago, hfahmy said:

@pyrojoe34, that spec is better for suited for my needs. I won't be using CUDA soon, may be sometime in the future. But I'll need 2 x HDMI out at least. So, what would you suggest for a GPU?

Hard to say, it's probably not worth getting an expensive one on the off chance that you do some CUDA in the future, better to get a cheaper one and upgrade if you decide you want the extra power later.

 

Will be doing any gaming, video editing or 3D work with this build?

[hfahmy]

31 minutes ago, pyrojoe34 said:

Hard to say, it's probably not worth getting an expensive one on the off chance that you do some CUDA in the future, better to get a cheaper one and upgrade if you decide you want the extra power later.

 

Will be doing any gaming, video editing or 3D work with this build?

Nope, just data science and other work. Definitely no gaming, video editing or 3D modelling work.

[pyrojoe34]

1 hour ago, hfahmy said:

Nope, just data science and other work. Definitely no gaming, video editing or 3D modelling work.

Then grab a cheap used card like a 950, 750/760, or 650/660 (the cheapest you can find that's not too old) for <$50 on Craigslist or something (you can even go ultra-cheap like a 730/740). An Nvidia card with even a small amount of CUDA cores will even allow you to test CUDA code and see if you want to use it or not and if it's worth an upgrade in the future. However, if you go with the i7-6700k build instead you don't even need a GPU since it has one built in and you can just use the motherboard outputs. To be honest 99% of the code most people do will not be able/need to use even most of a 6700k so that would be enough if you go that route too.

[brob]

I have a slightly different take. An i7-6700K because it has the highest base clock of any Skylake cpu. An after market cpu cooler and z170 motherboard mostly to allow for lightly overclocking the cpu if that should ever be desired. Think of this as a bit of "future proofing". High speed, lower latency memory to improve load and access times. A fast M.2 ssd, again to improve load / save times. A fairly low-end gpu to improve display speeds and off-load graphics processing.

 

Rather than two 1080 monitors, consider going with single a 3840x2160 unit. Has a slightly larger "desktop", (i.e. can display a little more than two 1920x1080 monitors), while occupying a little less desk space. Very nice if one ever has to display a very wide spreadsheet.

 

PCPartPicker part list / Price breakdown by merchant

CPU: Intel Core i7-6700K 4.0GHz Quad-Core Processor  ($328.99 @ SuperBiiz) 
CPU Cooler: Noctua NH-U12S 55.0 CFM CPU Cooler  ($59.50 @ OutletPC) 
Motherboard: Asus Z170M-PLUS Micro ATX LGA1151 Motherboard  ($119.99 @ SuperBiiz) 
Memory: G.Skill TridentZ Series 32GB (2 x 16GB) DDR4-3200 Memory  ($184.99 @ Newegg) 
Storage: Samsung 950 PRO 512GB M.2-2280 Solid State Drive  ($314.99 @ SuperBiiz) 
Video Card: EVGA GeForce GTX 1060 3GB 3GB GAMING Video Card  ($198.99 @ SuperBiiz) 
Case: Phanteks Enthoo EVOLV MicroATX Mini Tower Case  ($119.99 @ Newegg) 
Power Supply: SeaSonic S12II 430W 80+ Bronze Certified ATX Power Supply  ($44.99 @ Newegg) 
Operating System: Microsoft Windows 10 Home OEM 64-bit  ($85.70 @ My Choice Software) 
Monitor: Asus PB279Q 27.0" 60Hz Monitor  ($498.99 @ SuperBiiz) 
Total: $1957.12
Generated by PCPartPicker 2016-10-26 08:44 EDT-0400