Machine Learning Computer To CRUSH 2021 – Let me introduce a couple of possible setups that will let you crush 2021 like a pro. First of all, it’s important to be aware of what you need. In my opinion, that’s the gist of the whole problem and I will try to focus on building a better understanding of particular options. As a disclaimer, I will say that computers for Machine Learning are generally expensive, but read this article ‘til the end because I will show you some alternatives that might be a good fit for you.
OK, so let’s distinguish a couple of problems that require different setups. If you’re going to focus on professional Reinforcement Learning, in most cases you don’t really need powerful GPUs. A crucial aspect of your models is parallelism. It means that training can occur simultaneously and a number of CPUs are what you really need. In this case, the most important part of the computer is its processor. What I found as a good balance between price and money is AMD Ryzen with 12 cores and 24 threads.
It means that you could run up to 24 simultaneous epochs for your environment. Depending on your environment and algorithm, you need a decent amount of RAM to serve it. What I recommend here, is a safe choice of two 16GB RAM T-Force Vulcan for only $140. The rest of the components aren’t that important if you’re not going to heavily use algorithms that leverage deep learning.
You can choose whatever compatible components you like and you’ll be fine – just don’t forget about the SSD hard drive, but I assume it’s standard nowadays. For Deep Learning, on the other hand, you should focus on GPUs and when it comes to other components, they just shouldn’t interrupt. The best value to price ratio in my opinion belongs to GeForce RTX 2060 Super for $400. It has 8gb of memory, 2176 cores, and more importantly, 272 Tensor Cores which are responsible for CUDA optimized operations.
With this card, it should be perfectly fine to run any kind of model for your personal projects and you could use it even for commercial use in the future. If you are able to spend more money, you can aim to buy 2 or 3 of them and they should cooperate perfectly fine as long as you take care of a processor and motherboard that can handle more GPUs.
Since the other components are less important, I will list them in a description section. Just remember to always ask a specialist whether a particular part will be compatible with your other choices. You can get free advice by calling to almost any computer parts seller. If you’re interested in tabular data, I’ve got good news for you.
You can use one of the gradient boosting on decision trees libraries like XGBoost or AdaBoost and you don’t need anything else but your laptop. You probably think, “OK, Jack, but I have no idea what kind of problems I’ll have to solve in the future, and I don’t want to spend $2000 on a universal Machine Learning computer”. And I totally get it!
I’ve got a solution for you if you think this way. Fortunately, we live in the 21st century and there’s such thing as a cloud computing. You can get access to decent hardware completely for free. It’s not gonna be a top-notch setup, but as long as you need it only for your private projects and you’re not going to start in a competition, it should be perfectly fine to use.
After some time, you’ll decide whether you need a strong, personal computer, or you prefer to use cloud solutions. I have 2 completely FREE propositions for you.
The first one is Kaggle kernel.
You can use GPUs provided by Kaggle. The big plus here is that the whole Kaggle forum is a repository of knowledge for Data Science. The big minus here is its limits.
You might face problems when you have to use them more frequently than an average user.
The second solution is AWS free-tier – in my opinion, that’s the optimal plan for everyone who doesn’t like to buy a dedicated Machine Learning workstation.
First of all, AWS and their SageMaker service are just a fantastic help for every Data Science enthusiast. You can test many algorithms 100% for free and even if you are interested in doing something that goes beyond free tier, it’s gonna cost you cents.
Literally it’s almost for free even outside of free tier, as long as you use it for your personal use and you don’t need very powerful hardware. As an additional plus, there are many built-in algorithms that you can almost plug and play without much coding to instantly see the results. Definitely check this out before buying a new computer. I wish you a fantastic day and I’ll keep my fingers crossed for your career.
reference – Machine Learning Computer To CRUSH 2021
Web enthusiast. Thinker. Evil coffeeaholic. Food specialist. Reader. Twitter fanatic. Music maven. AI and Machine Learning!