Which HPE server do I actually need?

For 10 to 30 staff sharing files, running a domain controller, hosting accounting software and perhaps a couple of light virtual machines, you don't need a large server — you need a reliable one sized sensibly. For the very smallest offices of fewer than 10 staff, the HPE ProLiant MicroServer Gen11 is worth considering — wall-mountable, office-quiet, from AED 2,500. For most 10–30 staff offices, the ML30 tower sits on the floor and runs quietly. If you have a rack, a single-socket DL360 Gen12 does the same job in 1U. A single processor is plenty for this load; spend the budget on a second drive for redundancy and a proper backup rather than on cores you won't use.

This is the most common enterprise workload — consolidating many servers onto one physical host. The DL380 Gen12 is the default: a 2U dual-socket Intel server with the memory capacity and expansion room virtualization demands. If your VMware licence bill is a concern, the dense AMD DL345 Gen12 is the consolidation target that cuts it — and pairs with per-socket Morpheus licensing, covered in depth in the VMware licence guide linked below.

For SQL Server, PostgreSQL, MySQL or Oracle supporting a business application, two things matter most: fast storage and enough memory to cache the working set. The DL380 Gen12 with NVMe storage handles the large majority of business databases comfortably. Use NVMe rather than older SAS or SATA — database performance lives and dies on storage latency. Size the memory so the active part of the database stays in RAM; going to disk for every query is what makes a database feel slow.

ERP software runs on fast storage more than it runs on core count. Whether the platform is SAP Business One, Microsoft Dynamics 365, Odoo, or a similar system, the bottleneck is almost always storage latency — not processor speed. For up to 50 concurrent users, a DL360 Gen12 with NVMe SSDs handles the workload without overspending. For larger estates — multiple companies on one instance, heavy reporting, or more than 50 users running concurrent sessions — step up to a DL380 Gen12 for the extra memory capacity and storage bays. Dubai's trading companies, construction groups and logistics operators run this configuration — it is the most common ERP server deployment we see in UAE business.

In-memory platforms like SAP HANA keep the entire dataset in RAM, so the deciding spec is total memory capacity. The four-socket DL580 Gen12 scales to 16 TB of DDR5 in a single 4U chassis — enough for large HANA deployments. This is the one workload where the big four-socket machine is the right answer rather than overkill. For smaller HANA instances a well-specified DL380 may suffice; the dividing line is the dataset size, which we'll size against your SAP requirement.

Virtual desktop infrastructure is memory-hungry and, increasingly, GPU-assisted. Running on Citrix Virtual Apps and Desktops or VMware Horizon, a standard task-worker desktop typically needs 2–4 vCPUs and 4–8 GB RAM per session — so a server with 512 GB RAM handles roughly 64–120 sessions before reaching capacity. For standard task-worker desktops, a DL380 Gen12 with generous memory hosts many sessions per server. For graphics-accelerated desktops — designers, engineers, CAD — you'll want GPUs, which is where the DL380a Gen12 with its dedicated GPU capacity comes in. The planning metric is users-per-server and the per-user memory and (where needed) GPU allocation, which depends on the platform and what those desktops run.

AI inference, model serving and GPU compute need a server built around the GPUs, not one with a card squeezed in. The DL380a Gen12 is HPE's dedicated chassis for this — up to ten double-wide GPUs with dedicated power and cooling domains, so the cards run at full power reliably. For occasional or light GPU work, a DL380 Gen12 with up to three double-wide cards may be enough; for sustained AI workloads the DL380a is the purpose-built answer. GPU selection and count follow the model you're running.