How we organized the delivery of GPU server for AI and machine learning implementation: from client request to rent with buyout option

Innovative artificial intelligence (AI)-based solutions can bring great value to businesses. Automating routine operations, improving customer service, high-performance computing, code generation, creating virtual reality applications, performing complex tasks – in whatever capacity artificial intelligence is used, it provides additional benefits and allows you to set yourself apart from competitors. With such a powerful tool at its disposal, a company can not only strengthen its position in the market, but also make a serious leap forward.

After recognizing its potential, many enterprises have already started integrating artificial intelligence into their business processes. Accordingly, there is a need to purchase or rent special hardware: powerful AI server equipped with high-performance, energy-efficient graphics processing units (GPUs) and capable of handling complex tasks such as processing complex AI algorithms or analyzing big data.

This case is dedicated to one of the first projects related to AI servers. It was successfully completed. The client was satisfied and plans to further increase capacity with our help.

About the project and objectives

We have been working with a major gambling platform specializing in online casinos for four years. Its IT infrastructure is hosted in our data center and maintained by our company. Recently, a client needed to deploy an AI server to simulate internal business solutions, and they contacted us with a corresponding request.

The request was unconventional for the gambling company, as its core business activities had not previously involved the use of artificial intelligence hardware.

«The project was small and not very complex, but it was the next step in a long-term partnership and gave us a new opportunity to demonstrate our team's professional approach and specialist expertise» – Olga Boujanova

Initially, it was assumed that this server would be the first and a special test platform would be purchased for it. In case of successful testing and integration, the company plans to expand its capacities and create a separate IT infrastructure for working with artificial intelligence.

Video card is key element of AI server

Special attention had to be paid to the selection of video cards, because they played a key role in calculating the cost and performance of the AI server. Since NVIDIA products are currently at the peak of popularity, the client wanted to use video cards manufactured by this particular company. In addition, they conducted preliminary tests using old NVIDIA models, and even they performed well.

We were given a rough specification of the AI server, to which we added a video card model that was suitable in terms of power and cost. Since the model was somewhat outdated, we decided to consult HPE, where we planned to purchase hardware for this project. The vendor recommended more up-to-date and powerful video cards, including NVIDIA L40 48GB PCI Express.

Configuration selection

When communicating with the vendor, we were convinced that the specification provided by the company did not quite meet the stated objectives. Therefore, our technical specialists prepared several configuration options and offered the client a choice. Having studied the options, they approved modification with 48 GB video card. This model is not the latest generation, but it is up-to-date and optimally suits the specific situation in terms of price/performance ratio.

AI server configuration best suited for specific tasks

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

We submitted the adjusted technical requirements to our commercial department, which formed a proposal based on it. The management of the gambling company quickly agreed on everything, and the project was put into operation.

According to technical specifications, we ordered the required components from the vendor and built production server based on HPE hardware with a configuration capable of meeting the high computing power requirements of AI modeling for a gambling company.

Why HPE server hardware was selected

The client selected server hardware from HPE, a leading manufacturer of products for the IT industry. The choice was primarily driven by the need to address current business modeling issues. But it was also made with the expectation of meeting the growing needs for computing resources in the future.

Technical innovations applied in design of DL380a

HPE DL380a Gen11 4DW features fourth-generation Intel® Xeon® Scalable processors and is optimized for GPU accelerators. These processors deliver exceptional computation power and productivity due to a larger number of cores and advanced CPU functionality compared to third-generation servers. DDR5 and PCIe Gen5 RAM increases bandwidth and improves I/O performance. Combined, this results in balanced I/O performance on all CPUs, so that smooth and efficient operation is guaranteed even under heavy loads.

The latest version of HPE iLO integrated management system allows you to securely configure, monitor, and remotely manage your server hardware from anywhere in the world. In addition, HPE iLO provides all critical information about the major components of the server. The system administrator can quickly get data on device inventory and power management, reports on temperature, firmware, and “health” status.

What distinguishes HPE DL380a is also the engineering design for the GPUs used in it. The manufacturer has proposed a new approach to supporting GPUs. They are moved to the front of the chassis, which contributes to better air exchange and cooling. To increase uptime, each GPU is provided with a separate power supply. The updated server design allows for up to eight single GPUs or up to four dual GPUs, as well as the ability to directly communicate between individual microprocessors via high-speed connection. By combining the available GPU memory, the communication speed is increased.

Optimal AI server for business needs

Upon completion of the project, the gambling company received a server that combines innovative technology, modern design and special functionality for artificial intelligence.

The server configuration allows for a significant increase in performance and the ability to process AI models with 100+ billion parameters. The manufacturer took into account the specifics of the AI appliance and created a server with characteristics that allow it to efficiently perform a wide range of tasks in the field of artificial intelligence: from predictive maintenance of hardware to training neural networks for speech recognition.

«A server in this configuration is an ultra-scalable platform that can easily handle the specific workloads inherent in AI» – Volodymyr Marchenko

During preliminary negotiations, in addition to the main request, the client mentioned the need to solve parallel tasks and store a large amount of data. We took this into account and made it possible to add several more NVMe disks. But in terms of the number of video cards the specified set reached the limit, as the main focus was on them. In this particular case, further expansion will be made only by increasing the number of servers.

Comprehensive services

In addition to building and delivering AI servers, we provided the client with full range of services including:

• colocation in Tier III data center  with rent of a server rack;
• redundant power supply (such redundancy guarantees continuous operation of the IT infrastructure even in the event of a power failure);
• network connection.

As part of the project, we also performed initial default hardware preconfiguration, provided continuous monitoring of IT infrastructure status and technical support. The company’s system administrator was provided with independent remote access via secure VPN, which allows for convenient and secure resource management.

This server was the first step in the gambling company’s long-term plan to expand its computing capacities. In case of successful integration, the client plans to acquire additional capacity by renting at least one more VPS server with GPU and similar characteristics.

Rent GPU server for machine learning: solution with perspective

The custom build we made was fully compliant with the current tasks of the gambling company. For other businesses, similar AI server rental can be an initial solution that can be gradually expanded by adding GPUs or other accelerators as needed, rather than adding a finite number of GPUs at once. This solution provides flexibility, scalability and performance required to support most complex AI applications not only today, but also in the future.

Below are a few examples of possible HPE DL380a Gen11 AI server configurations with different video cards.

Option 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

Option 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

Option 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

Option 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

Option 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

All DL380A configurations come with 8 SFF NVMe U.3 slots that can be replaced with 12x EDSFF NVMe, if necessary. It is also possible to add licenses for operating systems: Red Hat Linux, SUSE Linux, and Windows Server.

Benefits gained by the gambling company from implementation of the project

The result of our work on this project was a high-performance AI server configured for today’s business needs and with an eye on the future. The client was able to save a lot of money and also received other benefits from the cooperation.

1. Investment distributed over time. The AI server was purchased on a leasing basis. This financial scheme implies payment in installments, allowing business to avoid one-step large expenses.
2. Hardware discount. Since our company is in a long-standing partnership with HPE, the vendor provides us with hardware on special terms. We, in turn, make good discounts for clients. For example, according to the usual price list AI server price is higher than €100,000, while the price of the configuration offered to the gambling company started from €140,000. Thanks to our partnership with the vendor, we were able to make our regular client a very favorable offer.
3. Comprehensive service. Together with the AI server, we provided all the services required for its trouble-free operation: from preconfiguration to technical support. Taking into account the cost of services, AI server cost was about 20% less compared to the base AI server price from the manufacturer.
4. Up-to-date models to order. We do not keep hardware in stock, but order it on request, which allows us to minimize risks associated with the rapid obsolescence of technology. While this may take a little more time, the result is that we guarantee delivery of up-to-date and optimally suited hardware at a favorable price. We take care of problems in case of delivery delays, providing replacement stock and migration assistance if necessary.
5. Reliability and security in accordance with requirements for Tier III data centers. The purchased AI server was added to the existing IT infrastructure of the gambling company, where a secure network connection was already built. For new clients who rent racks from us, for the first time, it should be noted that all resources are placed according to the security rules and regulations of the data center. This ensures reliable power supply and stable functioning of hardware. As for network protection, access to the administration console is provided via secure VPN.

Organizational aspects and nuances

CloudKleyer technical engineer and project manager participated in the project implementation. The work was coordinated with the managing director and technical architect. HPE specialists were also involved for consulting. On the side of the gambling company, the chief engineer and employees from three departments were involved. The request for supply of an AI server came from the Business Development Department, and the specialists of the IT Infrastructure Department and the DevOps Department, who are responsible for the implementation of business modeling in the company, were directly involved in the work on the project.

The project implementation was completed successfully and on time. In general, from the request to the launch of the new server, the work took less than 4 weeks.

This project was small and no serious issues were encountered during its implementation, but there were some nuances. For example, when the client approached us, they already had a clear idea of what they needed and formulated the task, but they did not take into account that technologies and components are losing relevance. We constantly follow the latest developments and take into account the rapid changes in the market, so we offered alternatives based on our experience and recommendations of the vendor. As a result, we selected the AI server configuration that was the most reasonable in terms of performance, delivery time and price.

Ready for the future: start your AI project with CloudKleyer

We strive for flexibility and innovation while sticking to the budget allocated for the project. Thus, our team is always ready to provide clients with state-of-the-art solutions that precisely match their goals, even if it requires extra effort on our part.

If you are interested in developing your business with artificial intelligence, take advantage of our support. We will help you select the ideal technical solution that will allow your company to reach a new level of work with big data and machine learning. We will take care of all stages of implementation: from organizing the delivery to setting up and launching the hardware, as well as providing you with dedicated GPU server rental on favorable terms and conditions.

Olga Boujanova

Consultant on server hardware and data center organization