Laptop for machine learning

This is the third article in a series dedicated to the study of AI and everything related to it. In the first part we discussed the theory, in the second part we prepared to solve practical tasks. Today we will be building the perfect computer for machine learning and setting up the system.

So, let's go to the selection of laptop for machine learning. I will offer three options: budget, medium and fancy.

Laptop for machine learning

If you don't want to build a new computer, you can simply upgrade your old one! Buy a Titan X or GTX 1080 graphics card (and Nvidia recently announced a more powerful GTX 1080 Ti) and install VMware Workstation or any other GPU-enabled virtual machine. Alternatively, you can put Linux as the main system, and install Windows in a virtual machine – this way you get the maximum performance for machine learning.

Then install the necessary frameworks, which we will discuss in the second section of this article, and you will have a ready-made system, and a very cheap one at that.

Video card

The central processing unit is no longer the most important part of a computer. For sophisticated gaming systems, powerful processors are needed, but not for machine learning – where Nvidia plays the role of Intel.

Although AMD cards have proven themselves in cryptocurrency mining, they have not yet matured to AI. That will change soon, but for now, Nvidia is on a roll. But don't discount Intel. It has acquired Nervana Systems and plans to launch specialized machine learning chips this year.

It has 3584 CUDA cores at 1531 MHz and 12 GB of GDDR5X video memory with a bandwidth of 10 Gb / s.

For machine learning, the number of cores and the amount of memory are important. Basically, machine learning algorithms are just a bunch of linear algebra. Imagine a huge Excel spreadsheet. A simple 4- or 8-core Intel processor simply cannot handle this amount of data.

Moving data in and out of memory is very computationally limiting, so the more memory there is on the card, the better. Therefore, Titan X is the best.

It's a pity, but only 2 cards can be bought on the official website. But for us, money is not a problem, so we will take two more from somewhere else. Yes, it's 4-way SLI! It will cost you about 360 thousand (at the time of writing – approx. Transl.), But this is the lion's share of the costs.

Any benchmarks will warn you that an SLI of more than 2 cards will not give a significant increase, but we are not playing toys (okay, let's be honest, not only playing toys)! To work with AI, you need to use the maximum possible number of video cards, so four is not the limit at all. Note that you don't need an SLI bridge unless you plan to use the system for gaming. The maximum that will be associated with graphics is plotting in matplotlib.


I advise you to take at least 16 GB of DDR4 RAM. Many people say that RAM should be 2 times more than video memory, but 16 GB should be enough for you. I would recommend the Corsair Vengeance DDR4-3000 model, a set of two 8 GB sticks will cost about $120.