Novice: which build - threadripper vs ryzen

Deep Learning
1

I’m new, but am hoping to start a masters program in addition to doing work for my own business applications. I’m interested in augmented reality, natural language processing, and simulations.

I don’t have any of these components yet. For the first 2-3 years I don’t expect to have more than 2x GPUs unless there’s a strong business need - in which case I could also utilize cloud offerings.

Between these two builds, which would you choose? The threadripper with it’s future compatability for 4x GPUs, or the ryzen with it’s low cost and support for 2x GPUs? The cost is equal between the two.

Threadripper: https://pcpartpicker.com/user/EricCBohn/saved/QPPjpg

Ryzen: https://pcpartpicker.com/user/EricCBohn/saved/qfnWbv

2 Likes
2

Your top list is private.

3

First, actually getting two GPU’s to work together is… tough.

Second, you don’t need threadripper.

Spend the extra $500 on a rtx 2080 ti for the extra 3gb of vram or shoot for a quattro.

edit; you also have way too many fans and your hard drive is way too small

edit2; and get 64 gb of ram

4

+1 for extra ram

5

I made the Threadripper build public, sorry about that.

I used the fans because I don’t like the lighted ones the case comes with. I may end up with a different case.

I have a 3TB HDD that I can put in this.

What’s a Quattro?

6

Sorry, I thought they named it quattro, not Quadro.

It’s nvidia’s flagship for machine learning (RTX 6000). Expensive though, about $4k a card.

Get one of intels cheaper 1.5-2TB NVME’s, they’re plenty good and faster than a spinner.

Buy a different case if you don’t like the fans vs spending $60+ on fans.

7

I had been using Tim Dettmers guides mainly when selecting my builds. In the guides, he says the RTX 2070 is the best to start with, and to get twice as much RAM as your video card supports (rounding up to get 32).

https://timdettmers.com/2018/12/16/deep-learning-hardware-guide/

https://timdettmers.com/2019/04/03/which-gpu-for-deep-learning/

The 2080 Ti had seemed more than I would need, but I’m curious why we have two people so far are suggesting 64gb ram and the 2080 Ti over the 2070 super? It’s a deviation from the guides and my proposed builds that I wasn’t expecting

Also, the single GPU suggestion is surprising. A part of me thinks it could be good experience and fun to learn to parallelize across GPUs, but another part of me is happy to hear that a single gpu is enough for my needs because of the cost savings.

1 Like
8

64GB RAM: Because pandas uses exponentially more memory in larger dataframes.

I could easily, easily use 128GB/256GB on a 4GB sqlite db

Actually I’ll also respond regarding the Super; they’re a crappy branded card for gamers. Anyone who does anything with the card, like video editing or ML is going to want 11gb of vram. Just get a 2070 instead of a super and save $$ if you’re going the cheap route.

9

Wondering if anyone else has any opinions on this. The cost is equal between the Ryzen and Threadripper builds.