MainDedicated ⁄ How business can implement AI and machine learning without excessive costs: GPU server configuration selection and rent with buyout option

How business can implement AI and machine learning without excessive costs: GPU server configuration selection and rent with buyout option

Applying technologies based on artificial intelligence (AI) can significantly improve business efficiency and create conditions for its transition to a qualitatively new level. Many enterprises have already evaluated the potential of artificial intelligence and started to implement it in their activities. But they are faced with the need to modernize server hardware, as ordinary servers are unable to cope with the increased workloads associated with the use of AI because upgrading IT resources is an expensive endeavor. If a company doesn’t have an opportunity to invest a large sum of money in one installment, and it is necessary to integrate AI and machine learning in the near future, the solution is to rent a powerful server equipped with high-performance graphics processing units (GPU) with the buyout option. The economic feasibility of this solution for a limited budget is confirmed by the increasing demand for such services.

New AI cluster for gambling platform

The gambling company has been hosting its infrastructure in Frankfurt, at one of Germany’s leading service providers, for four years. When the need arose to integrate AI into its business processes and a separate server was required, it was only natural that the client turned to their regular service provider.

The company aimed to optimize internal procedures through business modeling. Artificial intelligence was given a key role in this process. Accordingly, the Business Development Department set a complex task for the service provider: to calculate the cost of a GPU server suitable for AI and deep machine learning, to agree with the supplier on the terms of delivery and to provide the hardware on a long-term rent with the buyout option.

The server was to be the first in a new AI cluster, and after successful trials on a specially purchased test platform, the client planned to increase capacities.

Manufacturer and configuration are the most important criteria for selecting AI server hardware

The client already had a ready-made AI server configuration, but it included outdated models of NVIDIA video cards, on which preliminary testing was conducted. The tests gave excellent results; NVIDIA video cards were fully satisfactory in terms of GPU power and price, so the company did not consider other manufacturers. But the service provider had doubts about the expediency of using these particular models, so the technical engineer asked the vendor for advice. The doubts were confirmed; it turned out that there are more up-to-date versions of NVIDIA with powerful GPUs capable of speeding up procedures related to complex calculations, precision computations, big data processing, machine learning and other AI tasks many times over.

The service provider offered several hardware configuration options with new generation video cards. The client selected the best price and performance option with NVIDIA L40 48GB PCI Express:

  • HPE DL380a Gen11 4DW.
  • 2x Intel Xeon-Gold 6444Y 3.6GHz 16-core.
  • 8x HPE (1x64GB) Dual Rank x4 DDR5-4800.
  • 2x HPE 7.68TB NVMe.
  • Intel E810-XXVDA2 Ethernet 10/25Gb 2-port SFP28.
  • 4x NVIDIA L40 48GB PCIe.
  • 4x HPE 1800W-2200W.

The adjusted configuration was approved and the project went into operation. It should be noted that this hardware configuration was best suited for the anticipated workloads. The AI server immediately included four power supplies and the maximum number of video cards with high computing power.

To perform parallel tasks, the client needed a large storage with the possibility of expansion. The server-provider added two NVMe disks at once: for the first time this number will be enough, and additional disks can be added later if necessary.

The video card played a leading role in the selection of hardware, but the choice of server hardware manufacturer was also of great importance. The choice is influenced by many factors, primarily the specifics of the task, budget, performance requirements, and scalability.  HPE ProLiant DL380a Gen11 server equipped with fourth-generation Intel® Xeon® Scalable processors is the most suitable in this particular situation.

DL380a server features

  • With more cores and improved CPU performance over the previous generation, DDR5 RAM, and PCIe Gen5, AI server delivers high processing power, performance, and smooth operation even under heavy loads.
  • Unique front-loaded GPU design has increased processor density. Up to four dual and up to eight single GPUs can be used simultaneously. This number is ideal for natural language processing tasks, deep neural network machine learning, and other AI workloads. In addition, due to structural changes it provides efficient cooling, reducing risk of overheating and triggering of forced thermal throttling that significantly degrades performance.
  • Integrated HPE iLO management system provides secure remote server management and remote status monitoring of all critical server components.

In this technical configuration, HPE ProLiant DL380a Gen11 server is capable of supporting complex AI applications, processing large data sets, including visual information. This makes it an ideal solution that fully meets the needs of a gambling company not only for now, but also for future growth.

How to rent a server with GPU for machine learning with buyout option and save up to 20% of the base cost

The entire IT infrastructure of the company is maintained by this service provider and located in one of Frankfurt’s leading data centers (Tier III level). The secure network connection has already been built, the redundant power supply has been secured, and the system has high fault tolerance and reliability. The client was satisfied with everything, so they decided to place the new AI server in the same data center.

Thus, the project included a whole range of services:

  • assistance in configuration selection;
  • negotiations with the vendor and organization of delivery;
  • assembly and preliminary configuration of server hardware;
  • network connection in accordance with Tier III data center security regulations and rules;
  • providing the client with administrative access to resources via a secure VPN;
  • 24/7 technical support and monitoring.

It is also important to note that the server provider helped not only to build a high-performance AI server with GPU for machine learning, but also provided it on very favorable terms for the client.

Thanks to a long-standing partnership with the manufacturer, the service provider has the opportunity to purchase hardware at preferential prices and offer discounts to its regular clients. For example, typical HPE AI-server configuration costs about €100,000, the presented configuration with 4 top NVIDIA GPU, suitable for machine learning, will cost ~ €140,000 if purchased at the regular price. The gambling company managed to save about 20% including the cost of maintenance. At the same time, they did not have to pay the full amount at once, as the GPU server was rented with the buyout option under the contract with the service provider.

Smart IT asset management: what benefits the client gained from the project implementation

  1. An innovative server solution that opens new horizons of development in the field of AI and machine learning at an attractive price. The service provider delivers customized hardware, taking into account the specific needs of the client’s business. This approach allows providing clients with up-to-date models within the budget allocated for the project, which eliminates the rapid obsolescence of IT resources and helps to stay within financial constraints. At the same time, the service provider assumes responsibility for probable delays in delivery, providing the company with replacement stock and support in the process of forced migration at its own expense.
  2. The opportunity to add an expensive and very promising IT asset, avoiding a large one-time expenditure and saving at least 20% of the price offered by the manufacturer to ordinary customers.
  3. Turnkey delivery and a set of professional services save the client’s time and their own resources.
  4. Integration of the AI server into the existing IT infrastructure, which has been running smoothly for four years. The security and high fault tolerance of the hardware is guaranteed by the data center standards, including, for example, dual power supply, which ensures that the system continues to operate even in the event of a power failure.
  5. Secured VPN connection for system management access provides network security.

Participants, timing and nuances of implementation

At least 10 specialists participated in the project. The team of the gambling company consisted of representatives of three departments: IT Infrastructure, DevOps, and Business Development, as well as the chief engineer. The project manager, technical engineer, technical architect and managing director of the service provider took part in the work. Consultative assistance was provided by the vendor’s specialists.

As for problems, there were practically none in the process of implementation. Only a small nuance is worth mentioning. When the service provider received the request, it already included the desired configuration of server hardware and technical specifications. However, formulating the task, employees of the gambling company overlooked the rapid obsolescence of technology. And here the service provider was flexible and showed high customer focus, offering several alternatives that were more modern and within budget. As a result, the gambling company received a dedicated AI server with GPU, which is 100% suitable for artificial intelligence and machine learning, and most importantly, designed to meet the growing needs of the business.

GPU server rental is the optimal option for AI and machine learning implementation

A custom AI server building and renting with a buyout option can be a great solution for getting started with artificial intelligence. The pros are not only in the fact that the enterprise gets the opportunity to pay for expensive hardware in stages, but also in the scalability and performance of such resources.

In the described case, the client added specific models of video cards in maximum quantity, which was conditioned by the current business goals. In this case, the capacity increase will be carried out due to the buildup of AI servers (VPS or VDS) with similar parameters. But you can go the other way: first take the minimum configuration and add components as your business expands or use other GPU video cards. And to optimize costs when implementing AI and machine learning, you can rent a GPU server with a buyout option.

Alternative configurations of HPE DL380a Gen11 server designed to work with AI

Example #1

  • DL380a Gen11 4x DW GPU
  • 2U, 4x PSU
  • 2x Intel Xeon Platinum 8462Y+
  • 8x 64 GB (512 GB RAM)
  • NVME boot card
  • 2x 7.68 NVME
  • 4x GPU NVIDIA L40 48GB

Example #2

  • DL380a Gen11 4x DW GPU
  • 2U, 4x PSU
  • 2x Intel Xeon Platinum 8462Y+
  • 8x 64 GB (512 GB RAM)
  • NVME boot card
  • 2x 7.68 NVME
  • 4x GPU NVIDIA L40S 48GB

Example #3

  • DL380a Gen11 4x DW GPU
  • 2U, 4x PSU
  • 2x Intel Xeon Platinum 8462Y+
  • 16x 64 GB (1024 GB RAM)
  • NVME boot card
  • 2x 7.68 NVME
  • 4x GPU NVIDIA H100 80GB with NVLink

Example #4

  • DL380a Gen11 4x DW GPU
  • 2U, 2 x PSU
  • 2x Intel Xeon Platinum 8462Y+
  • 8x 64 GB (512 GB RAM)
  • NVME boot card
  • 2x 7.68 NVME
  • 8x GPU NVIDIA L4 24GB

Example #5

  • DL385 Gen11 GPU 12 EDSFF
  • 2U, 4x PSU
  • 2x AMD EPYC 9554 3.1GHz 64-core
  • 16x 64 GB (1024 GB RAM)
  • NVME boot card
  • 2x 7.68 NVME EDSFF
  • 4x GPU NVIDIA L40 48GB

Each modification provides 8 SFF NVMe U.3 slots with the option to replace them with 12 EDSFF NVMe slots. Additionally, there is an option to add OS licenses including Windows Server, SUSE Linux Enterprise Server and Red Hat Enterprise Linux Server.

Key factor for successful integration of artificial intelligence into business

The integration of artificial intelligence into business processes is attracting more and more companies seeking to innovate and improve their services. In order to maximize the potential of AI, it is necessary to have hardware specifically adapted to work with it, which implies the availability of hardware with the appropriate configuration. In many cases, leasing, i.e. long-term rent of a dedicated GPU server from a service provider with buyback option, is a good solution in terms of functionality, costs and prospects.

By partnering with a reliable service provider, a business can get not only professional technical support, but also flexibility in IT resource management, which is especially important in a dynamic market. The provider, in turn, is able to work more closely with the client, offering innovative solutions that can be quickly adapted to meet changing business requirements.

Such cooperation allows the client not only to save on initial capital investments, but also to optimize operating costs, as leasing payments often include hardware maintenance and upgrades. In addition, the use of artificial intelligence requires appropriate knowledge and skills, so the service provider can offer staff training or even completely take over the management of the AI infrastructure.

With this approach, a company can fully focus on its core business, delegating technical aspects to professionals and ensuring the smooth operation of its AI systems. This creates the conditions for innovation and continuous development, which is a key success factor in today’s economy.

Article author

Olga Boujanova

Consultant on server hardware, network and cloud technologies

Case study:How to move to a new data center with minimal costs by leasing equipment?

Inline Feedbacks
View all comments