As an Amazon Associate and (affiliate of other merchants) we earn from qualifying purchases without any extra cost to you when bought through our links below.
If you are into machine learning and looking for one of the best desktop computers for it but cannot find any, you will need to know what specific configuration you should look for in the system in the first place.
This is an article specially created for that purpose that will let you know about the best configurations and the different aspects to look at in a desktop computer system that will serve all your purpose.
- The processor of the computer should come with huge processing strength because it will have to handle a lot of data. It should ideally be an Intel Core i5 of the latest generation or an AMD equivalent.
- You will need to store a lot of data as well and therefore the storage space and functions should be high and reliable. It should preferably not be less than a 256 GB SSD or a 1 TB HDD with reasonably high data transfer speed.
- A minimum 16 GB of RAM is required in a desktop to be used for machine learning because you will need to work on a lot of complex algorithms all through. Make sure that it has high speed and also keep on upgrading it down the road.
- While working on neural networks during machine learning you will need to render 3D models which will need a proper graphics card in your desktop computer. Using anything less than the RTX variant may affect your jobs adversely.
- As for the motherboard, it may be a costly or a budget model but it should certainly come with faster versions of PCI Express lanes to support high-end GPUs that you will need to run on your system for machine learning.
In This Article
What is Machine Learning?
It is a part of the data sciences which constitute a vast number of concepts, while machine learning mainly focuses on Artificial Intelligence (AI) and how a computer running on it can be made to learn and evolve from its default state without any sort of further programming, based on a sample algorithm.
This will further enhance our computing experience and allow data scientists to achieve those tasks easily which either required extensive programming or such programming those are not possible singularly.
It primarily emphasizes on prediction and thus has found its applications in various other fields as well.
Machine learning thus requires much knowledge and patience. You must have a firm grip on different concepts of computer programming, complex algorithms, mathematics, statistics, etc.
But all of these will require a platform on which you can learn and also run different applications related to your work. Thus, a huge raw computational power is a bare necessity.
A good CPU and GPU, a large RAM, and storage capacity other related components will also be needed. These also include a good cooling system as you cannot allow so many high-end components to heat up your PC and ultimately damage it.
And most obviously a better power supply since a computer with so many high-end specs is sure to require a lot of power to function. You must know that if you are planning on going deep in this field, even PC builds are not going to be cheap.
But that doesn’t mean that you won’t be able to build one for yourself without emptying your bank account. The main thing to keep in mind is that there should be enough space for expansions in the future, probably more than any other sort of build and that means keeping provisions for multi CPU and GPU setups as well along with all the usual aspects like RAM, Storage, etc.
This article is a collection of the best configurations that one may have in their PCs for machine learning. Like our other articles, this too has something for buyers of the different budget preferences, and so all are welcome here.
We have tried to place only the products which provide the maximum value for your money and hope that you will be helped.
Best Desktop Configurations for Machine Learning
For Budget Configuration for Machine learning:
- Processor: Intel core i5-9600KF
- Hard Disk: WD Blue 1 TB
- SSD: Crucial MX500 250GB
- RAM: Corsair Vengeance Blue 16 GB
- Graphics: ASUS ROG STRIX GeForce RTX 2060 SUPER
- Motherboard: Gigabyte B365M
- PSU: Corsair CX Series 750 Watt
- OS: Ubuntu Linux 19.04
For Professional Configuration for Machine learning:
- Processor: AMD Ryzen 7 3800X
- Hard Disk: WD Black 2TB Performance
- SSD: Samsung 970 PRO SSD
- RAM: Skill Trident Z Neo Series 32GB
- Graphics: EVGA GeForce RTX 2070 SUPER XC
- Motherboard: MSI B450 Tomahawk Max
- PSU: Thermaltake Toughpower 1200W
- OS: Ubuntu Linux 19.04
For High-end Configuration for Machine learning:
- Processor: AMD Threadripper 2920 X
- Hard Disk: Seagate IronWolf 4TB
- SSD: Samsung 860 PRO SSD 1TB
- RAM: HyperX Fury 64GB
- Graphics: MSI GAMING GeForce RTX 2080 Ti
- Motherboard: GIGABYTE X399 AORUS Xtreme
- PSU: EVGA SuperNOVA 1600 G2
- OS: Ubuntu Linux 19.04
The processor can be called the heart of any computing system since it is responsible for anything you can do on your PC or any such device.
For machine learning, a massive processing strength is a must and this is why many prefer using GPUs in place of CPUs. But we believe that no matter how strong, a GPU was never built to do the things a CPU does and that is why you are suggested to keep both in your build.
While the overall cost will be affected you will be able to use your PC for other purposes as well and since you will already be spending a lot, the later will be preferred.
Please note that we are not talking about multi CPU builds here as they are very costly and usually not required even by most experts.
The best entry-level CPU would be the Intel core i5-9600KF. It has got 6 cores and 6 threads. It is not that beast of a processor, but with a 4.6 GHz turbo speed, it can handle a good deal of work pressure. The idea is to pair it with a good GPU in order to obtain the maximum performance.
Then we have the AMD Ryzen 7 3800X. It is an octa-core processor and the reason we choose it is because it provides better performance than its comparable Intel competition, at a much lower price. When you are building a professional desktop, the more you save the better it is!
The best CPU today for machine learning in an affordable premium price range is the AMD Threadripper 2920 X. It is better than the elder Threadripper 1920 X straight up and can go against the Intel Xeon processors as well.
It has 12 cores and 24 threads. The base clock speed is 3.5 GHz and can be boosted to 4.3 GHz. The main features are that it supports quad memory and has enough PCI-E slots for installing even 4 GPUs. Thus, it is the ultimate machine learning CPU.
Hard Disk Size
Machine learning will require you to store a lot of files. In order to save some money and add some more reliability to your build, we suggest you do not go for a full SSD build. An SSD is not inseparable unless you are working on complex datasets, but keeping it will allow some smoothness to your device.
For budget buyers, we recommend the WD Blue 1 TB. It is the usual 3.5 inched drive and rotates at a speed of 7200 rpm. There may be cheaper alternatives, but this is the fastest and most reliable at this price.
Next, for more performance-oriented users there is the WD Black 2TB Performance. Like the one above, it is similarly dependable and fast. Also, you get a 5-year warranty and as long as you have a backup of your data, there shouldn’t be any troubles.
The most pricey of this list of hard drives would be the Seagate IronWolf 4TB. It offers good performance and the 4 TB of storage should be enough for you to store your designs and experiments for a long period of time.
The minimum size preferred for the purpose we are talking about here is about 256 GB and anything near it would be good for entry-level builds. You may opt for Crucial MX500 250GB. It has read and write speeds ranging from 500-550 MB/s ideally and is one of the most bought SSDs.
As you go higher and into deeper algorithms, you will need to upgrade to a 512 GB SSD for better speeds. Here comes the Samsung 970 PRO SSD. It has a NvMe M.2 interface and can have the highest of 3500 MB/s for reading and 2700 MB/s for writing.
The highest that your system will require is 1 TB. You already have a large Hard disk at this point and there is no point in spending more on an SSD.
Also, SSD’s above this range don’t come that cheap. Thus, the Samsung 860 PRO SSD 1TB would be ideal. It not only provides decent reading and writing speeds but also a 5-year warranty.
Your PC will need to work on complex algorithms and neural networks. Only processing won’t do if you don’t equip it with enough RAM. If you will be working on a multiple GPU setup, then your system will require more RAM.
A minimum of 16 GB must be kept even for the base level builds. As you progress on the field, you will need to upgrade it. Keep the maximum RAM of your system to about 64 GB, and you shouldn’t have a problem.
Although, if your system is a top-end one, you may even go for 128 GB, that would be exaggerating.
- Corsair Vengeance Blue 16 GB (For learners)
This is the one that the low-end systems should be using. These are the cheapest but stable memory modules that you will find and although come at a lower speed, it can be overclocked.
- G.Skill Trident Z Neo Series 32GB (For professionals)
- HyperX Fury 64GB (For experts)
The slightly high cost of the HyperX Fury is justified by its performance. It is built for high-performance PCs and you should be able to make the most use of the four 8 GB modules when you have enough space in your rig.
There are other variants with different memory arrangements as well if you want two modules rather than four.
Another matter to take note of before choosing the RAM is that machine learning will require dual-channel memory, and the higher the speed of your RAM the better it is.
While most systems should be good with it, applying a 4 channel memory will need you to choose other components more attentively as not all processors and motherboards support it. Therefore it would be wise to invest in faster RAM modules even if it costs you a few more dollars.
This is the most important part of this build since graphics is of utmost necessity to render 3D models and working on neural networks.
Not only this, but the GPU should also be able to give a boost to your CPU as well in various instances since they have a lot more cores and processing capability, even than the high-end CPUs. Depending on your budget, you may have multi GPU setups as well.
While a single GPU in your system is usually enough for professional purposes, you will require a multi GPU setup for more high-end work. Your entire budget will largely depend on the graphics card you choose and thus you cannot make any mistake here.
We have tried to include the GPUs that offer the best performance in the given price tag. Thereby we will not be talking about extremely costly GPUs like the Titan series from NVIDIA since they are only suitable for highly expensive systems.
Rather we are concerned with builds that are affordable, even though expensive. The lowest recommended capacity of graphical storage is 8 GB and so our GPUs start from there.
The ASUS ROG STRIX GeForce RTX 2060 SUPER is a good GPU for you if you are low on budget or are a beginner in this field of data science. But no worries this is better than most other options present in this price range. It will be better if you use a dual GPU setup, but of course, your budget must cooperate in this.
Next, the EVGA GeForce RTX 2070 SUPER XC is the ideal choice among affordable video cards that have a great graphical computing capacity. Like the one before it too has an 8 GB storage requirement, and a triple fan setup so makes sure you choose the case and motherboard accordingly.
Finally, the champ of its segment, the MSI GAMING GeForce RTX 2080 Ti. It is 11 GB and also the best between the three GPUs we have mentioned here.
It can handle even the complex renderings, and using 4 of these in your premium PC build will ensure it lasts for a long time before you have to make any major upgrades.
With the support of 4 monitors, programming is ever smooth. Just remember to have spare PCI-E slots to employ more of these.
The motherboard, as you know will store all the major components. It does not matter if you choose a very costly one or a budget-friendly one, there is never going to be a difference in your performance in either.
But what does make a difference is the number of Express lanes your motherboard has and their versions (v3, v4, etc). The more PCI-E lanes you will have, the more GPUs you can run.
For example, most costly modern GPUs perform best at 16 PCI-E lanes and while they may also run at 8, their operating speed will be decreased.
The v5 express lanes are the fastest ones yet, but you will be well to go for a v3 if not a v4 version as they offer intermediate speeds and are pocket-friendly.
Having more PCI-E slots on your motherboard makes your system future-proof and ensures that when making upgrades, a motherboard replacement would not be necessary unless you have a hardware failure in this regard.
We have arranged the motherboard line up based on the CPU collection seen above. This means that the first one is the cheapest, while the last one will support AMD’s Threadripper CPUs.
If you are a data scientist or are learning to become one, you will soon realize that the Linux OS is far more suited for your purpose. This is because it has more software support and runs a greater number of applications for machine learning than Windows or Mac OS.
Also, there are other features it offers than the other two and thus it would be more useful for you. Although, if your PC is not solely for machine learning, you may consider Windows 10 Pro and Mac OS X systems as well.
As you may notice, by the time you buy a Power Supply your system already has a lot of components and that means the power requirement has also increased.
A PSU thus becomes one of the most relevant hardware units of your build. More GPUs will need power and this, in turn, will require a PSU with a higher wattage.
While a wattage of around 700W is necessary for even the basic of builds, there are many costly and high capacity ones available in the higher segments. If you use 4 GPUs, then a 1500W PSU should be enough.
- For Lower TDP systems: Corsair CX Series 750 Watt
- For Medium TDP systems: Thermaltake Toughpower 1200W
- For Higher TDP systems: EVGA SuperNOVA 1600 G2
So many hardcore parts will soon heat up and wear down real fast if you don’t provide sufficient cooling to it. A question that arises here is what kind of cooling would your PC require?
The answer is simple and it depends upon the overall specs of the build. Air cooling is best for single GPU arrangements while using it for more than one GPU will lead to a slight performance drop.
Water cooling comes to the rescue in these cases, but it is most costly and requires maintenance as well. Now you must decide whether you are comfortable with spending more for liquid cooling or are fine with the performance drop and would rather use air cooling.
So what you choose is up to you since we cannot guess the system you are building, but we suggest the second option if you are not so much serious about the performance. But options for both are given below for your utility.
The final touch to your build is the right case that not only allows efficient cooling but also leaves space for necessary expansions. An attractive look to your workstation wouldn’t be so bad either, right? Check out the cases we have selected for you that should give you enough options to choose from.
- NZXT H710i
- Fractal Design Define R6
- Cooler Master MCB-Q500L
- CORSAIR Carbide SPEC-Omega
- be quiet! Dark Base PRO 900
A high resolution display is a perquisite when one is working with graphics. Tasks like deep learning are all about graphics and so for the best output of your experiments and models, you need a decent monitor.
Now whether you want a multi-monitor workspace is your decision and it does in fact increase your productivity. Or you might want to go for a wide angle monitor so that the lengthy programs and algorithms can be viewed comfortably.
Whatever your preference may be we hope that you would not have to wander elsewhere after taking a look at the monitors we have mentioned below. These have been selected on the basis of screen size, quality of resolution and your budget, whatever they may be.
Keyboard and Mouse
In order to type the difficult algorithms and long programs, a comfortable keyboard and mouse are a bare necessity.
Choosing a good keyboard, even if it’s costly will let you finish up hours of coding without the slightest discomfort. A responsive set of keys and a similar mouse are thus very relevant for any modern day programmer.