Categories

CPUs

Server CPU selection comes down to balancing core count, clock speed, and platform compatibility. Workloads that scale across many parallel processes — virtualization, containers, databases — benefit from higher core counts even at a lower per-core clock speed. Latency-sensitive or single-threaded workloads benefit more from higher clock speeds on fewer cores. Always confirm socket and chipset compatibility with your target motherboard or server platform generation before buying, since server CPU generations are not cross-compatible the way some desktop platforms are. Dual-socket builds also need matched CPU pairs — check with our team before ordering if you are upgrading an existing dual-socket system.

131 products

GPUs

Server GPUs fall into a few distinct roles: AI/ML training and inference acceleration, virtual desktop infrastructure (VDI) graphics, and general compute offload. Training workloads benefit from the highest memory bandwidth and VRAM capacity available, since large models are often memory-bound rather than compute-bound. Inference and VDI workloads can typically use lower-tier cards, prioritising per-user density over raw throughput. Rackmount server GPUs are almost always passively cooled and rely on the chassis's own airflow — confirm your server's cooling design and available PCIe slot power before ordering, since passively cooled cards will overheat in a chassis without adequate directed airflow.

0 products

Memory

Server memory is sold primarily as RDIMM (Registered DIMM) or LRDIMM (Load-Reduced DIMM). RDIMM is the standard choice for most single- and dual-socket servers and offers a good balance of capacity, speed, and cost. LRDIMM supports higher maximum capacities per channel and is typically used in memory-dense configurations where every DIMM slot is populated with large modules. Almost all server memory is ECC (Error-Correcting Code), which detects and corrects in-memory data corruption — this is a requirement, not an option, for production server workloads. When upgrading an existing server, match the existing generation (DDR4 or DDR5), speed rating, and rank/width where possible, since mixing incompatible modules can force the whole channel down to the slowest common speed or prevent the system from booting.

40 products

Network Cards

Server network cards (NICs) are specified by port speed, port count, and connector type. 1GbE remains adequate for management and lighter workloads; 10GbE is now the common baseline for production server traffic; 25GbE and above suit storage networks, high-throughput virtualization hosts, and clustered workloads. Dual- and quad-port cards allow for link aggregation (higher combined throughput) or active/passive failover, which is standard practice for production servers where network uptime matters. Connector type matters too: RJ45 (copper) is simpler to cable but limited in reach and speed compared with SFP+/SFP28 (fibre or DAC), which most 10GbE-and-above deployments use.

0 products

Power Supply

Server power supplies are specified by wattage, redundancy configuration, and efficiency rating. Redundant configurations (commonly 1+1) let a server keep running if one PSU fails, which is standard for production servers and effectively mandatory for anything hosting customer-facing workloads. Efficiency is rated under the 80 PLUS programme (Bronze through Titanium) — higher ratings mean less power wasted as heat, which matters both for running costs and for cooling load in a dense rack. When replacing or upgrading a PSU, match the exact form factor and connector pinout for your chassis; server PSUs are rarely interchangeable across different server platforms or even across generations from the same vendor.

0 products

RAID Card

RAID controllers (and HBAs for non-RAID pass-through) manage how multiple drives combine into logical volumes. Hardware RAID cards offload parity calculation from the host CPU and typically include a cache with battery or flash backup, protecting in-flight writes from power loss — an important consideration for RAID 5/6 arrays, where a write hole without cache protection can cause data loss on unexpected power failure. Software RAID avoids the extra hardware cost but consumes host CPU cycles and does not offer the same write-cache protection. Match the controller's supported RAID levels and drive interface (SAS/SATA/NVMe) to your chosen drives before ordering, and confirm cache battery/capacitor health if buying a refurbished card.

0 products

Servers

Enterprise servers come in three form factors: rack, tower, and blade. Rack servers (1U–4U) suit data centre and colocation deployments where density matters; tower servers work well for smaller offices without rack infrastructure; blade servers pack the highest density but require a compatible chassis. Beyond form factor, weigh single- vs dual-socket CPU configurations against your workload — virtualization and database hosts typically benefit from dual sockets, while lighter application or edge workloads run well single-socket. Look for redundant power supplies and hot-swappable drive bays if uptime is critical, and confirm memory and drive capacity headroom for future growth before committing to a platform.

29 products

Storage

Enterprise storage splits along two axes: interface (SAS, SATA, NVMe) and media (HDD, SATA/SAS SSD, NVMe SSD). SATA drives are the most cost-effective option for bulk, sequential-access storage. SAS drives add dual-port redundancy and typically higher sustained performance, making them a common choice for RAID arrays in production servers. NVMe drives connect directly over PCIe and deliver the lowest latency and highest IOPS, suited to databases and other I/O-intensive workloads. For SSDs specifically, check the endurance rating (DWPD — drive writes per day) against your workload's write intensity: read-heavy workloads can use lower-endurance, lower-cost drives, while write-heavy database or logging workloads need higher-endurance drives to avoid premature wear.

21 products