Enterprise IT Made Easy

The #1 platform for building your IT infrastructure. Connect directly with top data centers, network, peering and cloud providers. We're 100% vendor-neutral.

Explore Data Centers


Streamline Your Colocation RFPs With IGNITE Project Platform

Datacenters.com IGNITE Project Platform makes it easy to configure your requirements for colocation services, select colocation providers and data center locations, and submit to your digital Colocation RFP directly to colocation providers matching your requirements.


Build Enterprise Cloud RFPs With IGNITE Project Platform

Configure your cloud servers, cloud storage, DRaaS, SaaS, UCaaS, or CCaaS solutions on the IGNITE Project Platform. Pick a category, enter your requirements for cloud services, and submit your digital Cloud RFP. IGNITE intelligently matches you to the right cloud providers.


Search Data Center Locations and Cloud Providers

Find data center locations, colocation providers and cloud service providers by searching our verified data center database including top markets, facilities and providers in the USA and Internationally. Click data centers near me to view datacenters within a specific radius.


Shop Colocation Pricing, Bare Metal Servers and Hosting Services

Get exclusive deals and discounts from top providers on colocation pricing, bare metal server pricing, dedicated servers and hosting services. Reserve your pricing online directly from the provider or customize your service using the Datacenters.com Project Platform.




data centers




marketplace products

Chris Newell

Global Consultant

Leslie Bonsett

Global Consultant

Michael Kriech

Global Consultant

Calling Data Center Consultants, Brokers and Telecom Agents

Chris Newell

Global Consultant

Leslie Bonsett

Global Consultant

Michael Kriech

Global Consultant

Join the Datacenters.com team as a data center consultant, real estate broker or telecom agent, VAR or MSP. We're always looking for elite industry professionals with a strong background in data center consulting, cloud consulting services, managed services and networking. Ask about our upcoming Datacenters.com Data Center Certification training program.

Data Center Industry Resources

Data Center Vendor List

Datacenters.com Vendor Directory is dedicated to datacenter owners and operators as a resource for sourcing vendors for critical infrastructure including UPS, cooling, construction, security, modular, hardware, storage, networking and more.

Visit Data Center Vendor List
Data Center Real Estate

Alongside manufacturing, data center real estate is one of the hottest markets in commercial real estate. Browse data centers for sale, lease and data center real estate investment opportunities on Datacenters.com. List your data center for sale privately.

View Data Center Real Estate

Trusted by Top Colocation Providers

What Our Providers and Customers Say About Us

Rackspace helps its customers accelerate the value of cloud at every phase of their digital transformation journey and Datacenters.com is a natural partner for anyone on this journey.
Vicki Patten
Vicki Patten
Our company prides itself on being the secure infrastructure company with a global platform of 50+ best in class data centers. We are happy to partner with Datacenters.com with its forward thinking, industry changing, global user experience being a great fit for our products. We are excited to be pioneers in the marketization of colocation and to be a part of the Datacenters.com Marketplace
Chad Markle
Chad Markle

Latest Data Center News

Read the latest data center news about cloud computing, technology and more.

Visit Data Center Blog
18 Sep 2020
Top Benefits of Remote Work for Employers and Employees
Although a contributing factor, the2020 pandemic wasnt the only driving force behind the rise of remote work. It may seem like a relatively new fad, but the concept of remote work actually dates back to 1972. Thats right. Jack Nilles coined the phrase telecommuting while working on a complex NASA communications system. Its something that has picked up speed over the years.COVID-19 accelerated the need for digital transformation technologies like unified communications (UC), contact center as a service (CCaaS), and cloud infrastructure. It also caused the share prices of video conferencing and collaboration platforms likeZoom,Microsoft Teams, andSlack to skyrocket.Keeping businesses connected even during times of social distancing, work from home, and economic uncertainty ultimately became the priority of the day. Not only was it a mantra, but it was also a necessity. Many businesses accelerated network modernization, cybersecurity stance, and cloud infrastructure to meet the demands of a virtual workforce.But what was the impact on the employer and employee?In this article, I will be looking at the lessons learned from COVID-19 including the benefits of remote work for employers and employees. I will also look at how the pandemic has altered the business landscape forever.1) Work-Life Balance ProductivityWork-life balance, its a buzzword that gets circulated in recruitment and brand messaging but what exactly does it mean? Work-life balance can be summed up by the word flexibility. That flexibility occurs when employees are able to work around the demands of their life.Think about it this way A traditional setting may require employees to be in-office from 8:00 AM-5:00 PM, Monday through Friday. By the way, most jobs are no longer 8-5ers. Theyre more like 7:00 AM-7:00 PM. With a flexible schedule, employees can work throughout the day, in between errands, exercise, and kid duties. Employees can work early morning, late nights, weekends, and whenever it works for them.As long as the work gets done, the traditional 8-5 job really is no longer relevant. With a remote workforce, employees are happier and less stressed when managing school schedules, medical appointments, errands, workouts, and home projects. The result is that employees feel more accomplished and have more time to spend with family and loved ones.2) Less Commuting More FocusHow many of you like commuting to an office in rush hour traffic? If you answered yes, I want to know how you do it. Think about this Theaverage commuter spends 225 hours per year commuting back and forth to work. Thats over nine calendar days spent going back and forth to the office.Studies have shown that commuting has a significant human and financial toll on employees. It causes considerable stress on employees who sit for hours in stop-and-go traffic. Not to mention anxiety caused by road rage, seeing and being involved in accidents. The side effects of commuting include fear, anger, and stress. This leads to lower energy levels, motivation, and morale. It also leads to the following medical conditions high cholesterol, high blood pressure, depression, and increased anxiety.The question for employers is would your employees be more or less motivated and productive if they didnt have to spend unpaid hours a day in their car? Would this lead to or detract from employee productivity and longevity? Would you retain your best employees longer?3) Larger Talent Pool More OpportunityFor employers, remote work expands your reach when it comes to accessing a larger talent pool. Quick question how many candidates ignore your job posts for employment due to where you are located? Maybe youre not able to tap into the local talent you need. Remote work is amajor selling point towards attracting highly-qualified candidates from markets outside your own.For employees, remote work means better opportunities and increased chances for career growth. Cities like San Francisco, New York City, and Los Angeles may offer great jobs in technology and financial industries, but the cost of living drives many applicants to look elsewhere. They may even settle for a job that they are overqualified for or not interested in within a different market.For both employers and employees, remote work represents an opportunity on a global scale. Its a concept thats both exciting and intimidating at the same time. Instead of competing on a regional level for job opportunities and work, employees must now compete with candidates all over the world and at much lower compensation levels. For employers, this can also be a risky proposition. You never know what you are going to get when hiring employees abroad.4) Promoting Diversity in the WorkplaceThis is a huge topic of discussion today diversity in the workplace. Remote work can make it easier to hire employees from diverse socioeconomic, geographic, and cultural backgrounds. Major changes are taking place employers are quickly realizing the benefits of environments that promote diversity and inclusivity.Remote work also makes it easier to expand into new markets whether regionally, domestically, or even internationally. It automatically encourages a diverse and inclusive environment. Imagine opening a new Asia-Pacific market and hiring a remote workforce in Singapore. Hiring locally can help your business align with cultural and language barriers that would have otherwise hindered operations and put market expansion success at risk. The hiring of a remote workforce could be the answer.5) Employee Employer Cost-SavingsThis is a big one. You may not realize it but there are significant cost-savings associated with remote work and were not just talking about employer costs. A study showed that employees that work from home saved an estimated $2,000 to $6,500 dollars per year. This includes common costs such as fuel, vehicle maintenance, transportation, parking, and meals.Are you ready for this? Another study showed that businesses were able to save an average of $11,000 per employee annually. What costs were included in the figure above? Office and real estate costs, office supply costs, transportation subsidies, and operational and personnel costs. This makes a lot of sense, right?And when you scale it, the savings go parabolic. Its estimated thatUS businesses saved $30 billion per day by allowing their employees to work remotely.6) Environment Corporate ResponsibilityIn addition to cultural diversity and inclusion trends, environmental impact is right there at the top of the list for corporate responsibility initiatives. Reducing the impacts of operations is a huge positive in todays world. Most businesses are focused on making a difference by recycling, increasing renewable energy usage, and reducing their carbon footprint.How does remote work fit into the picture? A recent study showed that by allowing 3.9 million employees to work from home at least half of the time, it reduced greenhouse gases by the equivalent of removing 600,000 cars from the road for an entire year. It was equal to a reduction of 7.8 billion miles and three million tons of greenhouse gases.Think about this The above statistics represent only one aspect of the environmental impact which is commuting. Remote work has also shown to reduce paper consumption, air conditioning, heating, and lighting use. Another study showed that the potential environmental impact of remote work could be the same as planting 91 million new trees.7) Increased Employee ProductivityI spoke earlier about flexible schedules and work-life balance. The major lesson learned from the 2020 pandemic is that the work from home concept actually increased productivity. A study showed that65% of employees think they would be more productive working from home than in a traditional office setting. 49% of respondents said that they rely on their home office when they need to focus and get work done. It makes sense, right?Think about it How much time do you spend in non-essential meetings and fraternizing with other employees, and carrying on non-work-related conversations? All of those minutes add up to hours and even days of wasted time throughout the year.8) Accelerated Digital TransformationYou could say that the pandemic caused many businesses to advance their digital transformation initiatives by a few months or even years. COVID-19 many businesses by surprise and as a result, many were caught off guard for how to equip their employees to work from home.If anything, COVID-19 has advanced or pulled the demand forward for new technologies and especially those that enable remote workforces. Its no wonder an entire set of publicly-traded companies in the United States saw their stock prices surge by 100%, 200%, or higher as a result of increased demand for their products and services. It was called the COVID trade.Thats not all. Cybersecurity also became a major area of concern for both businesses and IT departments as remote workers operated outside the comfort of their confines. Unsecured communications and access to systems, networks, and files became a major target for hackers. Bring Your Own Device, also known as BYOD, was also another target.However, there was a silver lining for many businesses They learned to prioritize and harden their IT infrastructure and adapt to a world where remote workers need access to systems and files outside of the enterprise network. Many businesses invested in and implemented new security practices that were needed regardless of the pandemic or not.9) Happier Healthier EmployeesHeres the bottom-line point. Remote workers tend to happier and more loyal employees. Why? Part of it is trust. Some of it is flexibility. However, it all comes down to the end result. Having time to balance work and life results in an improvement in life. Employees that have time for personal errands, hobbies and interests, and workouts have less stress and are healthier.Why is employee health important? Employees that are healthy have the ability to reduce health insurance costs for the entire organization. In addition, healthier employees are sick less, recovery from illness quicker and have less of a chance of being injured on the job.Conclusion: Remote Work is Here to StayIts easy to see the benefits of remote work from the employer and employee perspective. Some may say that the benefits are directly related to talent recruiting, employee productivity, happiness, and health. For others, its about corporate responsibility and the environment, diversity, and inclusion. Yet, others may focus on the financial rewards, bottom line, and cost savings.I believe that remote work is here to stay and its a major turning point in how we all do business. We will see a huge shift in demand for commercial real estate and office space. In fact, were already starting to see it with the cancelation of large-scale office leases and the sale of newly constructed headquarters even before companies move in.What are your thoughts? Will remote work adoption continue to accelerate in the post-pandemic, COVID-19 world?Need Help with Your Digital Transformation?Do you need help with your information technology and digital transformation initiatives? Were here to help! We offer the largest selection of Video Conferencing and Collaboration, UCaaS, CCaaS, IaaS, and SaaS solutions available in the market. When you work with GCG + Datacenters.com, you have direct access to all of the top cloud providers, telecom carriers, ISPs, and network service providers.Contact me to learn more about what we can do for your business. Its free and theres no obligation.
10 Sep 2020
Datacenters.com Launches Network Services on the IGNITE Project Platform®
Datacenters.com, the leading technology platform and marketplace for colocation, cloud and connectivity, announces the launch of network services to the Datacenters.com IGNITE Project Platform.Many businesses today are focused on wide area network (WAN) technologies as the need for reliable, secure, and scalable network infrastructure demand accelerates with digital transformation. Cloud services, Internet of things (IoT), and edge computing are fueling demand for network modernization. These technologies, paired with large, global IP backbones are enabling greater resource management, employee empowerment, customer experience and insights, and speed to market for products and services for businesses worldwide.With IGNITE Network Projects, IT professionals can design their networks and view available providers that match their requirements. Users can access powerful network configurators for software defined WAN (SD-WAN), multi-protocol label switching (MPLS), and private line networks. Based on the users location and technology requirements, Datacenters.com intelligently matches users with the right network service providers. Network proposals are uploaded directly by providers to the IGNITE Project Platform making it easy for users to compare all of their options.Datacenters.com has added 31 of the top network providers including ATT, Verizon, CenturyLink, Aryaka, CATO Networks, Zayo, Cloudgenix, China Telecom, and Colt Technology Services with the launch of IGNITE Network Projects. All of the providers have been on-boarded and trained on this platform.Key Features:- First digital RFP for enterprise network services- Users can design their SD-WAN, MPLS, and private line technologies online- Proprietary algorithm provides intelligent matching based on network requirements- Users can select all or specific network services providers to participate in their RFP- 31 top network service providers have been on-boarded and trained on the platform- Proposals and pricing for networking solutions are delivered directly through the platformQuotes:High-performance networks like SD-WAN and MPLS are critical for todays digital enterprise, said Michael Allen, VP of Solutions Engineering at Datacenters.com. These solutions represent a core solution offering for us and were excited to launch the first fully automated, intelligent digital RFP tool to help users architect their network infrastructure.Its amazing to think that we have onboarded and trained 184 providers on the IGNITE Project Platform in 2020, said Heidi Humphreys, President of Datacenters.com. Theres a lot of excitement around the launch of the IGNITE Network Projects both internally and from our partners that were an integral part of this launch.This is just the beginning for the IGNITE Project Platform, said Michael Price, VP of Software Engineering at Datacenters.com. We are continuously adding data and upgrading the platform to improve our intelligent matching algorithm and provide a better experience for Datacenters.com users and providers.###About Datacenters.comDatacenters.com is the #1 technology platform and marketplace connecting buyers and sellers of colocation, cloud, connectivity, managed services, and related IT services. We are dedicated to one thing helping IT professionals research, purchase, and manage their technologies across a diverse range of solutions, providers and vendors.Since 2014, Datacenters.com has attracted more than 1.6 million visitors. The platform provides detailed information on 335 providers,2,937 data center facilities, and 173 marketplace products globally.Datacenters.com is the sister company of Global Consulting Group, Inc. (GCG), a leading technology services distributor and IT consulting company headquartered in Englewood, Colorado.Learn more at Datacenters.com and follow us on LinkedIn.
8 Sep 2020
Top Questions to Ask When Writing a SD-WAN RFP?
To write or not write an SD-WAN RFP? That is the question. In all seriousness, an SD-WAN request for proposal (RFP) must include comprehensive questions that dive into vendor service features, capabilities, architecture, cost-savings, and proof of concepts (PoCs).When it comes to new technologies like SD-WAN, IT departments and enterprise network teams must write their software-defined WAN RFP in a way that encourages vendor and service provider transparency across a multitude of features within the SD-WAN marketplace. In this article, I will discuss the key questions that your business must include in writing their SD-WAN RFP.Unmanaged or Managed SD-WAN ConsiderationsThe most common consideration in purchasing from a vendor or SD-WAN provider is the approach. Will you be selecting do-it-yourself (DIY) or a fully managed SD-WAN solution? This is a distinction that your business will need to determine right away as it impacts on-premise equipment and network connectivity options.Just a few years ago, IT Departments would typically default to the telecom carrier or network service provider for their wide area network (WAN) edge devices and infrastructure. This included relying on the provider to deliver the equipment such as routers and switches from networking vendors such as Cisco or Juniper. However, the landscape has changed with SD-WAN and the platform is now the main consideration. This represents a shift from connectivity being the primary consideration.In writing your RFP, it will be important to carefully consider available platforms that consist of both hardware and software applications. We will dive into this subject in more detail later and cover key questions to ask.How to Start Writing Your SD-WAN RFP? Determine RequirementsTheres no doubt that SD-WAN offers a broad set of features, which can have a positive impact on your application performance, optimization, and security posture. The creation of a feature matrix is often an ideal place to start when writing your RFP. This provides a great starting point for your team as to the features that are most important to your business and technology stack.Here are a Few SD-WAN Features to Consider:Dynamic Routing This is a feature that ensures that traffic uses the best path possible depending on the business need. For example, mission-critical or delay-sensitive applications.Quality of Service (QoS) This SD-WAN feature assesses the granular application treatment across the user profile. This includes application type and business demand.Link Steering Remediation When an outage occurs, failover conditions are usually set as up or down. SD-WAN offers enhanced capabilities to sense packet loss and increased latency or jitter. This allows SD-WAN to make intelligent route selection based on circuit performance.Application Performance Monitoring Certain SD-WAN vendors offer detailed packet analysis and monitoring to assess traffic at the application and user levels. This can be extremely helpful in shaping your enterprise network.Next-Generation Security For many IT Departments and business, cybersecurity is a top priority. This is especially true with the network. Many SD-WAN vendors offer next-gen firewall services as a part of their SD-WAN solution offerings. Some do not.Network Function Virtualization (NFV) This is an important feature in most SD-WAN platforms. NFV simply asks whether the WAN edge device is available has virtualized capability within a cloud-based environment.Zero-Touch Deployment This is a feature that many businesses find extremely helpful. Zero-touch simply means that your IT team can turn-up services without needing to interact with physical hardware or networking gear. This results in fast and efficient delivery of SD-WAN services at the edge.Automation Orchestration A major benefit of SD-WAN is the quick and effortless installation of services using a management GUI. Make sure to evaluate the different automation and orchestration GUIs to find one that is user friendly and reliable.WAN Optimization Even though WAN optimization is typically delivered as a separate service offing, SD-WAN technologies frequently offer the ability to optimize and cache network traffic.Recommended SD-WAN Service-Based Questions to AskYouve created your service feature matrix, discussed important technology and business drivers, and now youre ready to organize your SD-WAN RFP into standard RFP sections. However, lets first start by asking the right questions!What is the SD-WAN vendors elevator pitch?Requesting an elevator pitch means that the vendor understands and has a high-value proposition. We are the only SD-WAN provider that can [or] our SD-WAN platform provides Common pitches include cloud-based NFV, global support, domestic vs. International footprint, network optimization, next-gen firewalls and granular QoS.Does the SD-WAN vendor sell a stand-alone service?Its important to know whether or not SD-WAN providers offer a standalone service or if they have other capabilities such as network connectivity and security solutions available. In many instances, a vendor might be known for a specific capability that primarily drives their sales activity. However, vendors may also offer valuable competencies outside their core offerings, such as firewall provisioning.Does the SD-WAN vendor provide support for third-party circuits?This is an important consideration. Does the SD-WAN vendor provide third-party circuit support? A major benefit of SD-WAN is the ability of the enterprise to use multiple circuit types and providers in building their SD-WAN or hybrid network architecture.Many businesses will migrate from MPLS networks to internet-based services. In this scenario, network teams may need to run double connections for a certain period of time. In fact, an MPLS Network may still be a component of hybrid network architecture. This may require the business vendors WAN edge to terminate private-based services.Does the SD-WAN vendor meet your global support and coverage requirements?If your business requires International internet connectivity, you will need to assess the SD-WAN vendors point-of-presence (PoP) coverage and IP footprint to understand the impacts it may have on application performance. Certain providers operate a substantial global network which includes specific PoPs for private and public internet traffic. SD-WAN focuses on application performance, but latency and jitter often arise when deploying International, global network services.SD-WAN Vendor-Based Questions to AskWhat are the SD-WAN vendors unique solution features?Certain SD-WAN vendors are known for their features or specialties. For example, this may include wireless deployments, security, out-of-the-box configuration and installation, 4G/5G access, and cloud support. Determine which service features are relevant and assess them against your business requirements.How does the SD-WAN vendor deliver its architecture?The promise of SD-WAN technologies is to provide an architecture based on WAN edge devices with access to a software management servers. Alternatives to consider in an NFV deployment are where the technology is based on x86 virtualized instances. Does the solution have the capability to mix and match NFV and hardware options?Can the SD-WAN vendor be separated from connectivity?Certain vendor services offer an option with SD-WAN and connectivity built-in specifically for use their software-based capability. This includes dedicated SD-WAN private network, PoP locations to drop off traffic to the regional Internet Service Provider (ISP), or straightforward public IP connectivity. Can the SD-WAN solution be separated from connectivity or will you be pigeonholed into a specific method?Does the SD-WAN vendor offer a proof-of-concept (PoC)?Demoing SD-WAN platforms and having the vendor help you build a PoC is an excellent way to determine the capability and performance of the SD-WAN solution. Certain SD-WAN vendors offer demo hardware for a period of time, frequently with pre-sales resources and engineering to assist with network configurations.What is the cost associated with the vendors SD-WAN product?A major benefit of SD-WAN is the associated cost-savings. This is especially true when migrating from an MPLS network. IT Departments and networking teams can expect a certain amount of cost reduction relating to their network infrastructure. The key is to realize how these cost-savings are provided by the SD-WAN vendor. Can the vendor highlight their solution cost-savings versus competitors and other WAN solutions like MPLS?Need Help with Your SD-WAN RFP?Datacenters.com has developed a digital RFP to help with configuring and selecting the right SD-WAN vendors for your network infrastructure. Its called the IGNITE Network Project Platform and you can check it out by clicking here.In addition to providing online SD-WAN configuration and a list of SD-WAN vendors that are intelligently matched to your WAN requirements, Datacenters.com offers white glove concierge services from internal network solution engineers and architects.Thats not all. You can actually complete your RFP online through IGNITE Network Projects and receive direct from provider proposals and pricing for your SD-WAN solution. This allows for side-by-side comparisons of SD-WAN vendors, services, and pricing.Get started by clicking here and start building your SD-WAN solution now!
1 Sep 2020
Data science experts have compared the time and monetary investment in training the model and have chosen the best option
Comparison (benchmark) of GPU cloud platforms and GPU dedicated servers based on NVIDIA cards.This article represents comparative experience in training a model on different GPU platforms: Google, AWS, and a domestic Dutch hosting provider HOSTKEY. Data science experts from Catalyst have compared the time and monetary investment in training the model and have chosen the best option.Material prepared by the Catalyst team.The opinions of the authors expressed herein may not represent those of the publisher.High-performance GPUs are in high demand today in all areas of the IT industry, in scientific projects, in security systems and other areas. Engineers, architects, and designers need powerful computers for rendering and 3D animation, for training neural networks and data analysis, real-time analytics, and for other tasks involving a large amount of computation. In general, high-performance GPU servers are a very popular solution on the market.Year over year, methods and applications for Artificial Intelligence (AI) and Machine Learning (ML) are becoming increasingly widespread in all fields. These technologies have already revolutionized the field of medical diagnostics, financial instrument development, the creation of new data processing services, as well as facilitating the implementation of numerous projects in the fields of basic science, robotics, human resource management, transport, heavy industry, urban economy, speech recognition, people and content delivery facilities. For example, NVIDIA encouraged owners of powerful GPU computers to donate their processing power to help combat the COVID-19 coronavirus pandemic.GPUs as a serviceAs many modern machine learning tasks use graphics processors, people are increasingly faced with one overarching question: which one should I use? To answer this question, you need to understand the cost and performance indicators of the different GPUs. Currently, there are a number of providers that offer virtual, cloud (GPUaaS) and dedicated GPU servers for machine learning.Dedicated servers are the best option when you need computing power on an ongoing basis, and when maintaining your own machine is prohibitive or simply impossible. Some providers rent out GPU servers on a monthly basis, and if you have the right workload that can keep the server busy in terms of hours and minutes, then the prices can be very attractive indeed. However, in most cases, virtual servers (VDS / VPS) with by-the-hour or minute billing or cloud services are still the preferred choice as they are usually cheaper and more flexible.Types of GPUs used in the dedicated GPU server pools of four leading providers (according to Lliftr Cloud Insight, May 2019)According to the latest Global Market Insights report, the global GPUaaS market will reach $7 billion by 2025. Market participants are developing GPU solutions specifically for Deep Learning (DL) and AI. For example, the NVIDIA Deep Learning SDK provides high-performance GPU acceleration for Deep Learning algorithms and is designed to create off-the-shelf DL frameworks.GPUaaS companies focus on technological innovation, strategic acquisitions and mergers to strengthen their market position and to stake out market share. Key players in the GPUaaS market include AMD, Autodesk, Amazon Web Services (AWS), Cogeco Communications, Dassault Systems, IBM, Intel, Microsoft, Nimbix, NVIDIA, Penguin Computing, Qualcomm, Siemens, HTC. For Machine Learning, the most popular GPUs are AWS, Google Cloud Engine, IBM Cloud, and Microsoft Azure.You can also take advantage of the power of graphics accelerators from less well-known providers their GPU servers can be used in any number of projects. Each platform has its pros and cons. It is believed that high-performance servers with powerful GPUs allow you to quickly achieve your goals and get meaningful results, but are the expensive iterations worth the money, and in what cases?Learn and learn some moreThe process of training a model is more computationally expensive and it requires a lot more resources than is required to execute an already trained model. If the task does not imply some kind of ultra-high load, such as face recognition from a large number of cameras, then just one or two graphics cards of the GeForce GTX 1080 Ti level are enough for most jobs a key element in managing that project budget. The most powerful modern Tesla V100 GPUs are typically used just for training.Moreover, the working version of the model might only take up a fraction of the resources of the GPU meaning that one card can be used to execute several models. However, this approach does not work when training because of the nature of the memory bus device, and also an integer number of graphic accelerators is used for the training of each model. Also, the learning process is often a set of experiments to test hypotheses, when a model is repeatedly trained from scratch using a number of different parameters. These experiments are conveniently run in parallel on adjacent graphics cards.A model is considered trained when the expected accuracy is achieved, or when the accuracy no longer increases with any further training. Sometimes bottlenecks arise when training models. Ancillary operations like preprocessing and downloading images may take up a lot of processor time. This means that the server configuration was not properly balanced for this particular task: i.e. the CPU is not able to feed the GPU efficiently.This applies as well to dedicated and virtual servers. In a virtual server, the problem can also be caused by the fact that one CPU is actually serving several GPUs. The dedicated processor is at full capacity, but its capabilities may still be insufficient. Simply put, in many tasks the power of the central processor is important as it is needed to process data in the first stage of the task. (see graphic below)Simplified model training scheme.Training on multiple (N) GPUs is as follows:A set of pictures is taken, divided into N parts and decomposed by accelerators;In each accelerator, its error rate is calculated for its own set of pictures as well as its future gradient (the direction that the coefficients must go so that the error rate becomes smaller);The gradients from all the accelerators are collected by one of them and added up;The model takes a step in the averaged direction on one accelerator and learns;A new, improved model is then sent to the rest of the GPUs, and it cycles back to step 1 and repeats.If the model workload is very light, then the execution of N cycles on one GPU is faster than their distribution on N GPUs followed by the combination of the results. The savings can be enormous. Indeed the financial outlay can be significant when engaging in active experiments with a model. For instance, when testing many hypotheses, the calculation time can sometimes last days, and consequently, companies sometimes spend tens of thousands of dollars a month on renting Google servers.Catalyst to the rescueCatalyst is a high-level library that allows you to conduct Deep Learning research and develop models faster and more efficiently, reducing the amount of boilerplate code. Catalyst takes over the scaling of the pipelines and provides the ability to quickly and reproducibly train a large number of models. Thus, you can do less programming and focus more on hypothesis testing. The library includes a number of the best industry solutions, as well as ready-made pipelines for classification, detection and segmentation.Recently, the Catalyst team has partnered with Georgia State University, the Georgia Institute of Technology and the Emory University Joint Center for Translational Research in Neuro-imaging and Data Science (TReNDS) to simplify model training for neuro-imaging applications and provide reproducible Deep Learning studies in the field of brain imaging. Working together on these important issues will help better understand how the brain works and improve the quality of life of people with mental disorders.Returning to the topic of the article, Catalyst was used in all three cases to check the speed of the servers: on the HOSTKEY, Amazon, and Google servers. Lets compare how it went.Ease of useAn important indicator is the usability of the solution. Decidedly few services offer an intuitive virtual server management system with all the necessary libraries as well as instructions on how to use them. When using Google Cloud, for example, you have to install libraries yourself. Heres a subjective assessment of the web interfaces:Service Web InterfaceAWSAnti-leader: all configurations are encoded with p2.xlarge-style names to understand what hardware is hidden under each name, you need to go to a separate page, and there are about a hundred of such namesGoogle Average rating, not a very intuitive setupGoogle Colab Jupyter Notebook AnalogHOSTKEY Web service not usedIn fact, to start using these tools, it takes about 2060 minutes (depending on your qualifications) to prepare install the software and download the data. As for the convenience of using pre-released, ready-to-use instances, the situation here is as follows:Service Software SuiteAWSThe absolute leader, each instance contains every popular version of machine learning libraries as well as a pre-installed NVIDIA-Docker. There is a Readme file with a brief instruction on how to switch between the myriad of versions.Google Comparable to HOSTKEY, but you can mount your Google-drive and thus have very quick access to the data set without needing to transfer it to the instance.Google Colab Offered for free, but once every 610 hours the instance dies. The software is completely absent, including a Docker + NVIDIA-Docker2 along with the ability to install them, although you can do without them. There is no direct SSH connection, but quick access to your Google drive is possible.HOSTKEYThe server can be provided both without software and already fully configured. For self-tuning, you can use the software in the form of a Docker + NVIDIA-Docker2. For an experienced user, this is not a problem.Results of the experimentA team of experts from the Catalyst project conducted a comparative test of the cost and speed of training a model using the NVIDIA GPU on servers from the following providers: HOSTKEY (from one to three GeForce GTX 1080 Ti GPUs + Xeon E3), Google (Tesla T4) and AWS (Tesla K80). The tests used the standard ResNet architecture for these kinds computer vision tasks, which determined the names of artists from their paintings.In the case of HOSTKEY, there were two accommodation options. First: a physical machine with one or three video cards, Intel Xeon E52637 v4 CPU (4/8, 3.5 / 3.7 GHz, 15 MB cache), 16 GB RAM and 240 GB SSD. Second: a virtual dedicated VDS server in the same configuration with one video card. The only difference was that in the second case, the physical server had eight cards at once, but only one of them was given over to the full disposal of the client. All the servers were located in their data center in the Netherlands.The instances of both cloud providers were launched in Frankfurt am Main data centers. In the case of AWS, the p2.xlarge type was used, equipped with Tesla K80, 4 vCPU, 61 GB RAM. Part of its computer time was spent on data preparation. A similar situation occurred in the Google cloud, where an instance with 4 vCPUs and 32 GB of memory was used, to which an accelerator was added. The Tesla T4 card, even if it was focused on inference, was chosen because by performance in training it is somewhere between the GTX 1080 Ti and the Tesla V100. In practice it is still sometimes used for training as it costs much less than the V100.Comparison of the machine learning times and the costs of model training on different platformsThe model was used in two versions: heavy (resnet101) and light (resnet34). The heavy version has the best potential for achieving high accuracy, while the light version gives a larger performance increase with slightly more errors. Heavier models are usually used where it is critical to achieve maximum predictive results, for example, when participating in competitions, while light ones are used in places where it is necessary to achieve a balance between accuracy and processing speed, for example, in loaded systems that process tens of thousands of requests daily.Training cost, USD / training time, heavy modelx-axis time, sec y-axis cost, USDThe cost of training a model on a HOSTKEY server with its in-house default cards is almost an order of magnitude cheaper than in the case of Google or Amazon, though it takes slightly more time. Furthermore, this is despite the fact that the GeForce GTX 1080 Ti is far from the fastest card for solving deep learning tasks today. However, as the experimental results show that, in general, inexpensive cards like the GeForce GTX 1080 Ti are no so worse than their more expensive counterparts in terms of speed, regardless of their significantly lower cost.Training cost, USD / training time, light modelx-axis time, sec y-axis cost, USDIf the card is an order of magnitude cheaper, then in performance on heavy models it is only 23 times inferior, and you can accelerate the whole process by adding more cards. In any case, this solution will be cheaper than Tesla configurations in terms of the cost of training the model. The results also show that the Tesla K80 in AWS, which demonstrates high performance in double-precision computing, shows a very long time at single precision worse than the 1080 Ti.ConclusionThe general trend is this: cheaper consumer-level graphics cards provide overall better value for money than Teslas expensive GPUs. The deficiency in the sheer speed of the GTX 1080 Ti when compared to Tesla can be compensated for by an increase in the number of cards, thus eliminating any advantage in using the Tesla.If you plan to perform a computer-heavy task, servers with inexpensive GTX 1080 Ti are more than enough. They are suitable for users who plan long-term work with these resources. Expensive Tesla-based instances should be selected only in cases where model training takes little time and you can pay for work by the minute of training.Finally, one should also consider the simplicity of preparing the environment on the platforms installing libraries, deploying software and downloading data. GPU servers both real and virtual are still noticeably more expensive than classic CPU instances, so long preparation times may lead to increased cost.