System Components – What You Should Know Before Buying
System components are the parts that make up a server, workstation, or other device in IT infrastructure: processors, RAM, motherboards, graphics cards, RAID controllers, network cards, and TPM security modules. Their selection and mutual compatibility determine the performance and stability of the entire environment. For IT departments and those responsible for hardware purchases, choosing server-class components means fewer unplanned downtimes, a longer hardware lifecycle, and a lower total cost of ownership (TCO). In this guide, we discuss the types of components available from Senetic, their manufacturers, and the criteria to consider when assembling or upgrading servers and workstations.
What are system components?
System components (computer parts) are all the hardware elements necessary for the operation of a server, workstation, or other device in IT infrastructure. The most important of these are:
- Processor (CPU) – performs calculations and processes data.
- RAM – stores data for running applications and processes.
- Motherboard – connects all components and ensures communication between them.
- Graphics Card (GPU) – generates images and, in servers and workstations, also accelerates calculations, including tasks related to artificial intelligence.
- RAID Controllers – manage disk arrays and protect stored data.
- Network Cards – provide communication with the corporate network and other servers.
- TPM Modules – are responsible for the hardware protection of cryptographic keys.
How do professional components differ from consumer ones?
The difference comes down to specific design features:
- Server processors support ECC memory and usually offer 8 to 12 memory channels per socket, while desktop processors typically have 2 channels.
- RDIMM and LRDIMM memory buffer control signals, allowing more modules to be installed on one channel without degrading signal quality.
- Professional graphics cards have ECC error-correcting memory and drivers certified by software manufacturers for CAD/CAM and BIM applications.
- Server-class RAID controllers are designed for continuous operation, feature a high MTBF (mean time between failures) rate, and protect cached data in case of power loss.
Why is it worth choosing certified components?
Certified solutions reduce the risk of incompatibility and data loss. Before purchasing, it is advisable to check:
- QVL (Qualified Vendor List) compatibility lists and supported processor lists published by motherboard and server manufacturers,
- memory module compliance with JEDEC specifications,
- FIPS 140-3 (or earlier FIPS 140-2) certificates for TPM modules.
This allows the IT department to verify each component based on technical documentation.
Types of system components
The selection of components depends on the role of the device: a database server requires a different configuration than a workstation for 3D rendering or a file server. Below we discuss the most important categories of components available from Senetic.
The server processor market is dominated by Intel and AMD. Modern server processors offer from a few to over a hundred cores per socket and support both single and multi-processor configurations. An important feature of server platforms is RAS (Reliability, Availability, Serviceability) functions, such as error detection and correction or isolating damaged components without stopping the system. When choosing a processor, it is worth paying attention to:
- Socket – the processor must fit the motherboard socket. Intel and AMD platform sockets, and often also the next generations of processors from the same manufacturer, are not interchangeable.
- Number of cores and clock speed – server processors usually have lower clock speeds than desktop ones, but a larger number of cores and a large cache provide higher throughput for multi-threaded loads, such as virtualization or databases.
- Memory support – the number of channels, maximum RAM capacity, and standard (DDR4 or DDR5).
- Number of PCIe lanes – determines how many expansion cards and NVMe drives can be connected directly to the processor.
DDR4 memory is still used in servers and earlier generation computers, while DDR5 is the standard for new platforms. Modules must be compatible with the motherboard: DDR4 and DDR5 have differently placed notches (keys), so they are not interchangeable. The following types of memory are used in servers and workstations:
- RDIMM (Registered DIMM) – buffers address and control signals, allowing more memory slots to be populated without losing stability.
- LRDIMM (Load-Reduced DIMM) – additionally buffers data lines, reducing the electrical load on the memory controller. It allows the use of modules with the largest capacities.
- UDIMM (Unbuffered DIMM) – unbuffered memory used in workstations and entry-level servers, often in ECC versions.
A larger amount of RAM allows for more virtual machines, containers, or database processes to be handled simultaneously. Senetic offers server modules with ECC error correction in various capacities.
Professional graphics cards are designed for stable operation in engineering applications, rendering, visualisation, and AI calculations. When choosing a card, it is worth considering:
- VRAM capacity – determines the size of 3D scenes and AI models that can be processed. In professional cards, memory usually has ECC error correction.
- Memory bandwidth – affects the speed of processing large datasets, e.g., during inference of language models.
- ISV certified drivers – ensure stable operation in CAD/CAM, BIM applications, and digital content creation.
- AI acceleration – specialised cores (in NVIDIA cards: tensor cores) and support for lower precision formats accelerate training and inference of models.
- PCIe interface – full-size cards use the PCIe x16 connector, while some compact models operate on fewer lanes.
- Power, cooling, and dimensions – the most efficient cards consume several hundred watts and occupy two or more slots, requiring an appropriate power supply, airflow, and space in the case.
The motherboard is the foundation of the system. Server motherboards, compared to desktop ones, offer:
- a greater number of PCIe slots,
- support for one or two processors,
- integrated management controller (BMC) with IPMI or Redfish support, enabling remote administration,
- support for ECC RDIMM and LRDIMM memory.
The motherboard format must be compatible with the case. Workstation motherboards are most often ATX, micro-ATX, or E-ATX formats, while server motherboards also include SSI-EEB, SSI-CEB formats or designs dedicated to specific cases. For servers mounted in a rack, the height of the case (e.g., 1U or 2U) must be considered, which limits the height of expansion cards and cooling systems.
RAID controllers manage disk arrays and protect stored data. Modern controllers support SAS, SATA, and NVMe drives, have their own cache with data loss protection in case of power failure, and offer features such as:
- Hot Spare drives,
- changing RAID levels during operation,
- expanding array capacity without shutting down the server,
- load balancing.
RAID controllers are installed in PCIe slots. When choosing, it is worth checking the number of internal and external ports, the generation of the PCIe interface, and the supported types of drives.
The network cards available include models with bandwidths from 1 GbE to 25 GbE. Cards designed for servers offer, among other things, RDMA support for low-latency environments, SR-IOV virtualisation, and offloading network tasks from the processor. When choosing a card, it is worth considering bandwidth, the number of ports, the type of connector, and compatibility with existing switches.
The TPM (Trusted Platform Module) stores cryptographic keys and enables secure system boot and disk encryption. Many motherboards and servers have a connector for a removable TPM module, but these connectors are not standardised, so the module must be matched to the specific model of the motherboard or server.
Manufacturers of system components
Senetic offers components from many manufacturers. Filters on the category page allow you to quickly find products from a selected brand:
- NVIDIA – professional graphics cards for workstations and servers, designed for CAD/CAM design, rendering, visualisation, and AI calculations.
- Intel – server processors and network cards. Selected Intel server processors have built-in accelerators that speed up AI tasks.
- Kingston and Samsung – DDR4 and DDR5 memory modules with ECC correction (RDIMM, LRDIMM).
- HP and HPE – OEM components for HP workstations and HPE servers, including memory, processors, and controllers with confirmed compatibility.
- Lenovo – parts and components for Lenovo servers and workstations.
- ASRock – motherboards in ATX and micro-ATX formats for workstations and servers.
- Broadcom – RAID controllers and network cards.
Frequently Asked Questions
How do server processors differ from desktop ones?
Server processors support ECC memory, offer more memory channels (usually 8–12 compared to 2 in desktop processors), more PCIe lanes, and multi-processor configurations. They also have RAS functions, such as error detection and correction and isolating damaged components. Their clock speed is usually lower, but a greater number of cores and larger cache provide higher performance under multi-threaded loads. Desktop processors, on the other hand, offer high single-thread performance at a lower price.
Is ECC memory necessary for a server?
Yes, in the vast majority of applications. ECC memory corrects single-bit errors in real-time and detects multi-bit errors. In database, virtualisation, and AI data processing environments, even a single error can corrupt a record in the database or the result of calculations.
How to check motherboard compatibility with a processor?
You should verify the socket type, chipset, and the list of supported processors published by the motherboard manufacturer. This list also provides the minimum BIOS/UEFI version required to support the given processor. A compatible socket does not guarantee support – a newer generation processor may require a firmware update or may not be supported at all. For RAM, you should check the QVL list.
What is the purpose of a hardware RAID controller?
A RAID controller creates disk arrays at the hardware level: it supports RAID levels 0, 1, 5, 6, 10, 50, and 60, has its own cache, and offers features such as Hot Spare, changing RAID levels during operation, or protecting the cache in case of power loss. The calculations related to managing the array are performed by the controller's processor, not the server's processor, allowing the server's resources to remain available for applications.
When is it worth choosing RDIMM modules over UDIMM?
RDIMM modules are worth choosing in servers where many memory slots will be populated. Buffering address and control signals allows more modules to be installed on a channel without signal degradation. UDIMM is suitable for smaller configurations, e.g., in workstations and entry-level servers, and is cheaper. The type of memory is indicated by the motherboard manufacturer – most server motherboards support only RDIMM and LRDIMM, and RDIMM and UDIMM modules cannot be mixed in the same system.
What is the purpose of the TPM 2.0 module?
The TPM 2.0 (Trusted Platform Module), compliant with the ISO/IEC 11889:2015 standard, provides a hardware root of trust. It stores cryptographic keys, supports secure system boot (Secure Boot), and enables disk encryption, e.g., using BitLocker in Windows or LUKS in Linux.
Can a server be upgraded with RAM from another manufacturer?
Yes, provided that the new modules have the same type (RDIMM or LRDIMM), standard (DDR4 or DDR5), speed, and number of ranks as the modules already installed. Identical capacity is also recommended. Mixing modules from different manufacturers may reduce memory speed to the level of the slowest module or cause instability. Before purchasing, it is advisable to check the QVL list from the server or motherboard manufacturer, and to test the configuration before production deployment.
What graphics cards are suitable for a workstation?
For CAD/CAM design and 3D rendering, professional graphics cards with ISV certified drivers and ECC memory are best suited. The VRAM capacity should be matched to the complexity of the projects: for everyday engineering work, mid-range models are sufficient, while for advanced visualisations and large assemblies, cards with larger memory should be chosen. In the case of AI calculations and local execution of large language models (LLMs), VRAM capacity, memory bandwidth, and support for AI acceleration are crucial.