Data Science Rig - Single Core Single Thread

[jpauk]

Hi All, 

 

I am attempting to build a rig for Data Science. My main question here is which processor would be the best for single thread work/price. Python and R are usually the most commonly used languages in the Data Science community. However, they only use a single core and thread unless you code in the Multi threading and Multiprocessing. Most data crunching is done before optimization and production of an application, hence I am chewing through some good size data on a single core before I put any efficiency in the code. The next main question is that Google's Tensor Flow, which is a machine learning library is able to make use of a GPU, is becoming more commonly used. So some moderately priced GPU advice would be helpful as well.  

 

I have been watching Linus tech tips for a couple years now and the videos that include the Intel Pentium G3258 dual core processor seemed intriguing for the data science purpose due to the high clock speeds and the price. However, with the release of the 7th gen Intel Core i3-7350K I am leaning toward the newer product. Also this is a machine that my IT department will have control over, so any risky overclocking is out of the question. 

 

There are a number efficient pieces of hardware I will be adding the the building including an M.2 SSD over PCI Express 3.0. Here is a what I have in my New Egg shopping cart as of now. Someone please stop me if something isn't going to work. I am also thinking running Ubuntu (or Peppermint) rather than windows.

 

LINUS!!!!!!!!! NEW BUILD IDEA

 

My Build

-CPU: Intel Core i3-7350K

-Mother Board: ASUS Z170-E ATX

-Memory: 64GB CORSAIR Vengeance DDR4 (4 X 16)

-GPU: ASUS GeForce GTX 1060-DUAL 3GB

-Power Supply: CORSAIR CX-M series 750W ATX

-CPU Cooler:Cooler Master RR-212X

-M.2: Toshiba OCZ RD400 128GB PCI Express 3.0

-SSD: 2 Sandisk SSD Plus 1Tb

-Case: Fractal Design Define R5 Black Window Silent ATX Midtower

[reniat]

I'd just like to point out that python CAN do true (e.g. non-time shared) concurrency. It's not simple and there is no shared memory between threads, but it's possible. If you're looking to make learning models in python, it is possible to do concurrently. I've not used R though.

 

I don't know a ton about data science professionally (I took a graduate course in college, that's it) but I'm pretty sure there are ways to get use of multiple cores. Lots of people I know who do use data science have clusters of large machines to do their crunching, and many research domains are taking advantage of AWS services through amazon for cloud computing. 

 

tl;dr im no expert, but are you SURE you only need single core performance when teaching models? That seems really backwards to me. 

[Shiv78]

A 7700K is probably your best bet, it has reached over 7GHz on liquid nitro with only 1 core enabled. I'd get a Kraken X62 with a 7700K if your budget allows

[jpauk]

reniat, 

 

You are absolutely correct. Once models are production ready, or real training is to be done, the full spectrum of technology will be implemented. However, in development there are a bunch of tasks like data collection and cleaning that I am just doing on my machine and not looking to write perfectly efficient code. So until I put applications onto production server, a modest machine will do the trick. My current quad core 2.3 GHz laptop isn't overly exciting to work on.  

[DragonTamer1]

No matter what you get I would wait until after the full ryzen releas to see if prices change. The 7350K is horibly overpriced and may come down in price once the competition has returned. It would even be worth waiting for the quad core ryzen to see if it can force its price down even more.

 

For what you are doing I think that the 7350K would be the perfect processor if its price is right. For a GPU an RX 460 or GTX 1050 would be a good pick.

[Mihle]

7700k is the best single core performance I think if you ignore price.

[fjdhdhcyfhf]

32 minutes ago, jpauk said:

reniat, 

 

You are absolutely correct. Once models are production ready, or real training is to be done, the full spectrum of technology will be implemented. However, in development there are a bunch of tasks like data collection and cleaning that I am just doing on my machine and not looking to write perfectly efficient code. So until I put applications onto production server, a modest machine will do the trick. My current quad core 2.3 GHz laptop isn't overly exciting to work on.  

i would get the i3, it will be so much faster than the laptop

[done12many2]

Did someone say single thread?  

 

 

 

 

[imreloadin]

11 minutes ago, done12many2 said:

Did someone say single thread?  

 

 

 

 

Damn son, how high do you have that overclocked lol?

[done12many2]

Just now, imreloadin said:

Damn son, how high do you have that overclocked lol?

 

5.4 GHz on that run.  My daily overclock is 5.3, which put's single-thread around the 230 mark.  Still snappy.  

[reniat]

what is your cooling situation?

[done12many2]

Just now, reniat said:

what is your cooling situation?

 

Full custom loop, delidded chip, but it's all ambient.  No sub ambient stuff.

 

Loops in signature.

[Lays]

18 minutes ago, done12many2 said:

Did someone say single thread?  

 

 

 

 

 

 

 

5 minutes ago, imreloadin said:

Damn son, how high do you have that overclocked lol?

 

 

[done12many2]

Just now, Lays said:

 

 

I'll bet you a $100 that you can't rerun that again right now and post it?  

 

That's right, you'd have to pull the board and hook up all the DICE stuff to do that.  lol  

 

Ambient rules!!

[Lays]

2 minutes ago, done12many2 said:

 

I'll bet you a $100 that you can't rerun that again right now and post it?  

 

That's right, you'd have to pull the board and hook up all the DICE stuff to do that.  lol  

 

Ambient rules!!

I can go beat yours tho on ambient

[done12many2]

Just now, Lays said:

I can go beat yours tho on ambient

 

Oh, you said your sub ambient can beat mine on ambient?  I won't argue with that buddy.  Now go find your own sandbox to play in.  

[Lays]

1 minute ago, done12many2 said:

 

Oh, you said your sub ambient can beat mine on ambient?  I won't argue with that buddy.  Now go find your own sandbox to play in.  

No I mean I can go beat your score right now on my ambient setup lol

[done12many2]

I guess you ain't working tonight?