Custom Build for Machine Learning

bcmenen

Hi friends,
I few years ago I built myself a solid gaming PC and it is still smooth as butter. For my next build, I want something that will be a little more geared toward programming and machine learning. I am aware that there is a lot of overlap here, but it is becoming hard to store both games and data programs on the same machine. Does anyone out there have any input?
Thanks for your help!

TheITDad

Do you have a budget?  Because from a quick look pc's built for machine learning can be quite expensive.

bcmenen

when the time comes I am looking at about a $3,000 budget. More I am wondering if a dual GPU set up is worth it, if a Ryzen 9 3900x is strong enough, things like that, or if I should just get some upgrades to what I have now. Currently I am working with a Ryzen 7 2nd gen (my current MB does not support 3rd gen). I feel like upgrading too many parts would be overkill and it would be easier just to sell my rig as it is and start from scratch. Thoughts?

TheITDad

@bcmenen That sounds like a good idea to me with selling current rig and dumping that into your budget for this build.  Because you can always use it for gaming and perhaps do a dual boot with one drive being for a linux based machine learning system and a basic windows 10 gaming rig or however you wish to do that.  From what I know and what I can find dual GPU's is kinda the minimum unless you go for a ridiculous gpu like a Titan or higher but with your budget your looking more at like dual 2070 supers or maybe trying to find some used 1080TI's (but the used market can be hit or miss).  The Ryzen 3900x is probably the best CPU for it currently with current gen TR's being out of budget.  Was kinda hard to make the list in microcenters builder which is normally pretty good.  But here is a pcpartpicker list that comes to 3260 before tax.  I made sure to have two drives to try to separate the gaming and machine learning.  But dual 2070 supers should definately be a good start and the price of the 3900x is hard to beat.

PCPartPicker Part List: https://pcpartpicker.com/list/QkDdvW

CPU: AMD Ryzen 9 3900X 3.8 GHz 12-Core Processor  ($418.89 @ B&H) 

CPU Cooler: be quiet! Dark Rock Pro 4 50.5 CFM CPU Cooler  ($86.46 @ B&H) 

Motherboard: MSI MEG X570 GODLIKE EATX AM4 Motherboard  ($649.99 @ B&H) 

Memory: G.Skill Ripjaws V 64 GB (4 x 16 GB) DDR4-3600 Memory  ($325.99 @ Newegg) 

Storage: Samsung 970 Evo Plus 1 TB M.2-2280 NVME Solid State Drive  ($199.99 @ Adorama) 

Storage: Samsung 970 Evo Plus 1 TB M.2-2280 NVME Solid State Drive  ($199.99 @ Adorama) 

Video Card: NVIDIA GeForce RTX 2070 SUPER 8 GB Video Card (2-Way SLI)  ($499.99 @ Best Buy) 

Video Card: NVIDIA GeForce RTX 2070 SUPER 8 GB Video Card (2-Way SLI)  ($499.99 @ Best Buy) 

Case: Corsair Air 540 ATX Full Tower Case  ($139.98 @ B&H) 

Power Supply: EVGA SuperNOVA G2 1300 W 80+ Gold Certified Fully Modular ATX Power Supply  ($241.60 @ B&H) 

Total: $3262.87

bcmenen

thanks for the info! Question, for this setup, esp with the multiple graphics cards, is there another piece of hardware I will need or will the two cards just run as they are set up? I have seen machine learning builds with multiple GPUs but there doesnt seem to be a ton of information out there about how they actually work together

TheITDad

I believe you will need an SLI bridge which I forgot to include but they aren't terribly expensive from what I know, they are about 120 bucks.  This is a link to a 4 slot one https://www.microcenter.com/product/513187/evga-geforce-rtx-nvlink-sli-bridge---4-slot-spacing.  There are 3 spacing ones which really just depends on how much space is between your pcie slots.  And the bridge lets them work in unison I believe but I am not 100% so hopefully someone else with more knowledge could chime in.