GPU server

Definition of

GPU servers are servers with at least one GPU (Graphics Processing Unit) that offer the performance and speed required to perform computing-intensive tasks.  

‍

Technology

Thanks to the parallel architecture of a GPU, which was originally developed for graphics and video processing, a GPU server can handle several tasks simultaneously at a speed that significantly exceeds the capabilities of a CPU-based server.

GPUs are throughput-optimized and specialized developments that complement CPUs. Instead of a few resource-intensive cores with a high clock frequency to perform computationally intensive tasks, GPUs use thousands of resource-saving cores that are optimized for parallel processing of these processes.

‍

‍

Application scenarios for GPU servers:

• High Performance Computing
• Industrial automation
• Conversational AI
• Drug research
• Financial services and the economy
• AI/deep learning training
• Healthcare
• Business intelligence and analytics
• Climate and weather modeling    
• Big data analysis pipelines
• 3D animations and simulations (e.g. modeling the folding of protein chains)
• Hash cracking (e.g. password recovery)
• Crypto mining
• AI inference and machine learning
• Cloud Gaming
• Media/video streaming
• Industrial automation
• Intelligent medical expert systems
• Scientific virtualization
• Edge Computing
• Micro Data Center
• Virtualization
• Oil and gas production
• Research institute
• Technologies for autonomous vehicles
• Molecular dynamic simulation

‍

Different types of GPU servers:

Rack servers with smaller or larger form factors differ mainly in terms of their density and expandability.
‍

1U and 2U GPU servers

GPU servers with smaller form factors have been specially developed for high performance density. The disadvantage is their performance compared to servers with larger form factors. They are often used because they are more cost-effective and take up less space in the server rack.

This design is easy to maintain, extremely portable and easy to scale (performance can be scaled by using several servers).

In servers with a smaller form factor, the GPUs are usually installed horizontally for space reasons. There is also little space for PCIe slots and storage, although there is the option of expansion using a PCIe expansion kit or JBOD housing.

‍

4 U, 5 U or 8 U GPU rack server

Larger GPU rack servers are designed for workloads that require greater performance. They have more space for storage and additional expansion slots so that additional PCIe cards can be used to improve data processing performance. The additional space is also required for better air circulation to prevent overheating.

In larger form factor servers, GPUs are installed vertically, with the additional space for the power connectors at the top of the card rather than at the back.

‍

Liquid-cooled universal GPU systems

Liquid-cooled systems for high-density AI infrastructures are essential for larger applications. Liquid cooling systems are significantly more energy efficient than conventional cooling systems.

‍

Customizable size of GPU server:

When sizing, we take into account the desired product functions as well as your current and future business requirements. The optimal server configuration for each application depends on the target workloads, the server's use cases and the desired speed.

GPU servers can be configured for specific target workloads such as video rendering, deep learning training, inference, big data analysis and high-performance computing (HPC).

GPUs consume a lot of power and generate a lot of heat. They are larger than CPUs and require additional space for power connections. The server housing must not only be large enough to accommodate the desired number of GPUs. Suitable ventilation must be planned to prevent overheating.

‍

Advantages

Compared to a CPU-based server, GPU servers offer significantly higher performance, more flexibility and better utilization of CPU resources.  

GPU servers are essential for a scalable AI infrastructure.

‍

Contact

We offer GPU servers in various form factors and configurations. With our manufacturer Supermicro, we offer high-performance computing solutions for a wide range of demanding applications. Our technical experts are available for further questions and support in every project phase.