Building a data analysis machine

[ikilledbill]

TLDR: Have a £3k budget for an analysis machine, need hardware recommendations. 

Hello,

I'm a complete novice when it comes to building but I thought I would give myself a challenge. 

 

I'm based in the UK and have just secured a fairly hefty research grant with a decent budget set aside for hardware. I reckon after buying a few neuro specific pieces of tech I'll have £3000 left for an analysis PC. 

 

The PC will be used to perform computational neuroscience analyses which includes machine learning, AI development, spectral analysis and quite meaty image analysis of MRI scans. 

 

As I say, I don't know much about PC hardware so I thought you guys might help me with choosing the CPU and GPU and maybe explain why you're recommending?

 

Thanks very much!

P.S. first time posting so apologies if I've not followed the rules in some way. 

[Metallus97]

wellcome 

 

so do you need CPU or GPU power or both?

do you need a anything special? 

My GF works in a MRI center to and they use tesla cards...

give us some more info about what the pc has to do and wich technologies are used 

[ikilledbill]

4 minutes ago, Metallus97 said:

wellcome 

 

so do you need CPU or GPU power or both?

do you need a anything special? 

My GF works in a MRI center to and they use tesla cards...

give us some more info about what the pc has to do and wich technologies are used 

Thanks for such a quick reply! Bare with me because I am a tech idiot. 

 

For the most part I believe we need CPU power with a lot of threads. The PC we currently use has an i7 (not sure which, but it is at least 5 years old) and 64gb ram. Even my more basic analysis in MATLAB runs for 8 hours and according to task manager uses all 64gb of ram and about 80% of the CPU. 

 

We analyse mostly EEG data in MATLAB, which involves importing ASCII files that are about 4gb in size. MATLAB keeps these files in memory while analysing which is mostly bandpass filters being applied and Fourier transforms for spectral analysis. 

 

We also run machine learning algorithms and evolutionary algorithms on huge datasets using R or MATLAB. These completely send the current PC into overdrive and usually have the CPU almost maxed out constantly for at least 15hours at a time. 

 

The GPU comes in for the MRI analysis at the moment we have some form.of Quadro in there. MRI is fairly graphics intensive because it is overlaying many slices to create a 3D model, then applying different 3D transforms and filters to the model. The GPU also takes a beating from MATLAB when we create spectrograms, but nowhere near as intense as MRI. 

[Metallus97]

ugh MATLAB i hate it ... Mathematica user here ( i am a  theoretical physicist)

 

Ok had to say that  

Now about the pc: do you need ECC ram or something like that?

in general i suggest going with amd thread ripper and a nice quadro.

give me some more info and i will pics you together something.

are we talking desktop pc or server rack?

[Metallus97]

are there parts you can move over from your old pc? For example storage? In general how much storage do you need in this system?

[ikilledbill]

After 4 years working with it I'm inclined to agree with you! I like R though. Gotta love open source 

 

Definitely desktop PC. The benefit being that we're building a new research centre in 2 years which will come with a server in the basement which we can use, which means I'll get to keep the desktop as my own. 

 

I want ECC ram yes (I think). 

 

We have a 4tb HDD that will be moved over. I'm a Linux user and our MEG analysis is performed in Linux. So I was thinking I would have separate SSD for Windows and Linux if that's possible? So boot either from their respective SSD the access the data from the shared HDD?

 

[Metallus97]

yep that is possible. 

problem with ECC: 64GB is close to 1000£. Without it’s more like 600£.

this will get expensive. I need to know what you need more. you either can get a 12 core CPU or a 16 Core... the 16 core would result in a weaker Quadro.

 

allso: how is your ram usage? Do you need the 64GB?

[ikilledbill]

I think the balance is probably important so maybe the 12 core CPU with a better Quadro rather than limit the GPU? 

 

I can move money around a little to get the ECC if it is the best option. 

 

All I can tell you about the ram is that when I have my analyses running the task manager reports that somewhere between 40 and 64gb is being used.

[Metallus97]

14 hours ago, ikilledbill said:

I think the balance is probably important so maybe the 12 core CPU with a better Quadro rather than limit the GPU? 

 

I can move money around a little to get the ECC if it is the best option. 

 

All I can tell you about the ram is that when I have my analyses running the task manager reports that somewhere between 40 and 64gb is being used.

sorry for the late answer... had to go to work

uh yes ok ill put something together. Ill do two options one with ECC and one without. This highly depends on how important and valuable the data youre working on is.

Maybe give this a read: https://www.pugetsystems.com/labs/articles/Advantages-of-ECC-Memory-520/

 

Oh and allso: Do you need the System ASAp or can you wait until August? There THreadripper 2 will be released and well offer more power for less money

[ikilledbill]

5 hours ago, Metallus97 said:

sorry for the late answer... had to go to work

uh yes ok ill put something together. Ill do two options one with ECC and one without. This highly depends on how important and valuable the data youre working on is.

Maybe give this a read: https://www.pugetsystems.com/labs/articles/Advantages-of-ECC-Memory-520/

 

Oh and allso: Do you need the System ASAp or can you wait until August? There THreadripper 2 will be released and well offer more power for less money

No worries! You could have never answered again and I'd still be much better off than yesterday. 

 

I've had a word with my colleague and he offered up some money left over from his grant so we can bump the budget to £4k. I say given the importance of some of our patient data we want to minimise errors so the ECC is the best choice. I never knew Kingston was a good brand! Only know them from USB sticks. 

 

Oh there's definitely no rush. My plan was to finalise the build plan by the end of August. Get everything ordered and signed off through September and do the build in my annual leave at the end of October. 

[Metallus97]

16 hours ago, ikilledbill said:

No worries! You could have never answered again and I'd still be much better off than yesterday. 

 

I've had a word with my colleague and he offered up some money left over from his grant so we can bump the budget to £4k. I say given the importance of some of our patient data we want to minimise errors so the ECC is the best choice. I never knew Kingston was a good brand! Only know them from USB sticks. 

 

Oh there's definitely no rush. My plan was to finalise the build plan by the end of August. Get everything ordered and signed off through September and do the build in my annual leave at the end of October. 

That sound realy nice!

Idd suggest  to wait for Threadripper 2 then. With 4k youre defenetely able to pick up a TR2 with 64GB of EEC and a nice Quadro.

I have put a rig togetther now. But hold of on buying, this is just a gereral idea with wich we can work when TR2 comes out.

 

[Metallus97]

PCPartPicker part list: https://pcpartpicker.com/user/Metallus/saved/bLjpGX
Price breakdown by merchant: https://uk.pcpartpicker.com/list/YFKfFt/by_merchant/

CPU: AMD - Threadripper 1950X 3.4GHz 16-Core Processor  (£654.97 @ Box Limited)
CPU Cooler: Noctua - NH-U14S TR4-SP3 140.2 CFM CPU Cooler  (£71.88 @ CCL Computers)
Motherboard: Asus - PRIME X399-A EATX TR4 Motherboard  (£274.97 @ Amazon UK)
Storage: Samsung - 970 Pro 512GB M.2-2280 Solid State Drive  (£199.97 @ More Computers)
Storage: Western Digital - Gold 4TB 3.5" 7200RPM Internal Hard Drive  (£159.18 @ PC World Business)
Storage: Western Digital - Gold 4TB 3.5" 7200RPM Internal Hard Drive  (£159.18 @ PC World Business)
Video Card: PNY - Quadro P5000 16GB Video Card  (£1462.90 @ Amazon UK)
Case: Corsair - 750D Airflow Edition ATX Full Tower Case  (£129.99 @ Amazon UK)
Power Supply: Corsair - HX Platinum 750W 80+ Platinum Certified Fully-Modular ATX Power Supply  (£129.99 @ Corsair UK)
Optical Drive: Asus - DRW-24F1ST DVD/CD Writer  (£17.19 @ Kustom PCs)
Other: Crucial CT4K16G4RFD4266 DDR4 64 GB (4 x 16 GB) DIMM 288-Pin 2666 MHz/PC4-21300 CL19 1.2 V Registered ECC Memory Kit - Green  (£708.00 @ Amazon UK)
Total: £3968.22
Prices include shipping, taxes, and discounts when available
Generated by PCPartPicker 2018-08-07 10:37 BST+0100

 

This is my first idea.

Cupple of notes about it:

1. As I said wait untill we can upgrade it to TR2

2. I included a DVD writer. I dont know but where my sister works they still have to write thier Data to DVDs sometimes. If you dont need it you can ditch it. Or if you need a bluray writer: 50£

3. I included dual Workstation grade HDDs. I suggest putting thm into a RAID1 or 5 for data safety. I dont know where and how you store your data... this will be much better than consumer Harddrives but should still not be considdered "super save".

4. The SSD is for the OS and program so everything will run fast and snappy

5. I got you 4x16GB Sticks of ram, potentialy you can later upgrade to 128GB

 

Can I write you a private message? So we can discuss there, when TR2 get released? I am afraid to forget you or not find the thread annymore