Reading a server quote: the line items that decide the five-year cost

The headline server price is useful only after every material component and service obligation has been normalised. A lower-priced server can cost more over five years if its warranty is shorter, restoration commitment is weaker, storage media is unsuitable, power redundancy is incomplete, or critical spares are supplied only at additional cost. Acquisition price remains important. Where hardware, licences and contractual terms are genuinely equivalent, it may decide the award. The problem is that two quotations described as “2U dual-socket server with five-year support” are often not commercially or technically equivalent. Start by establishing what the quoted unit contains A valid comparison requires an exact bill of material. Generic descriptions conceal differences that later become upgrade, replacement or downtime costs. For example, “1 TB RAM” does not show whether the server contains 16 x 64 GB RDIMMs or 8 x 128 GB RDIMMs. The second arrangement may leave more DIMM slots available, but channel population, supported speed and CPU topology must also be checked. Similarly, “dual-port 25 GbE” does not identify the controller, optics, cables, transceivers or supported firmware. For each quoted 1U, 2U, 4U or tower server, verify the following: Exact CPU model, quantity, core count and thermal design requirements. DIMM type, quantity, capacity, speed and population across each processor. Drive model, form factor, interface, endurance class and quantity. RAID or HBA model, cache capacity and cache-protection method. Network adapter model, port speed, optics and cables. Remote management features and any required licence tier. Rail kit, cable-management arm, bezel, power cords and rack accessories. Operating system, virtualisation, backup and management licences, including renewal terms. PSU quantity, wattage, efficiency rating and redundancy mode. Warranty duration, service level and support locations. The five-year calculation should begin only after this configuration baseline is fixed. A five-year warranty is not a complete warranty description Warranty duration does not establish how quickly a failed server will be restored or which costs are included. A quotation must define the service obligation behind the number of years. The first question is when the warranty clock starts. Dispatch, delivery, installation, commissioning and final acceptance can be separated by weeks or months. A five-year period measured from dispatch is not equivalent to five years from final acceptance. The quotation and tender conditions should then state whether parts, labour, travel and onsite attendance are included at every consignee location. Buyers should also check whether HDDs, SSDs, RAID cache-protection modules, fans, system boards and hot-plug PSUs remain covered for the complete term. Other points requiring written confirmation include: Business-hours, working-day or 24x7 coverage. Coverage for BIOS, BMC, storage-controller and device firmware updates. Warranty treatment of replacement parts. Exclusions for batteries, write endurance, accidental damage or environmental conditions. Failed-drive retention where storage media cannot leave the site. Escalation, service credit, LD or warranty extension when the committed service level is missed. GeM conditions must be read together with the category specification, Bid, RA, STC and ATC. Read the GeM General Terms and Conditions in force on your bid date: the version current when this was written set a standard warranty period unless another was specified through the applicable procurement documents, but portal conditions are revised and superseded, so the version number matters as much as the clause. The accessible version also addressed rectification, replacement and post-warranty spare supply. See the GeM General Terms and Conditions . Spectra's warranty coverage and the applicable conditions are set out at /warranty . Requirements involving onsite coverage, preventive maintenance or contracted response should also be mapped against /services . Response time and restoration time are different commitments A four-hour response promise does not necessarily mean that the server will be working within four hours. The clock may refer to acknowledgement, remote diagnosis, engineer arrival or part dispatch. The quote should identify: The event that starts the SLA clock. The severity levels to which the commitment applies. The coverage window, including holidays and non-working hours. The time for acknowledgement and remote diagnosis. The time for engineer arrival and spare delivery. The restoration or final-resolution commitment. Customer obligations for access, logs, diagnostics and remote connectivity. Published enterprise support descriptions illustrate this distinction. HPE describes onsite-response timing separately from receipt, acknowledgement and confirmation of a hardware incident. Lenovo also distinguishes onsite response from committed-repair services. These examples show why the exact SLA wording must form part of the commercial comparison, rather than relying on the phrase “four-hour support”. See the relevant HPE service description and Lenovo data-centre support options . Support must be confirmed location by location A service level available in a metro location may not apply at a district office, industrial site or remote installation. Multi-location procurement should include a PIN-code-wise support matrix rather than one general national statement. The matrix should show engineer availability, applicable response time, nearest spare depot, cut-off times and any travel exclusions. It should also identify whether a critical part is stocked locally, moved from a regional location, or ordered only after diagnosis. Spares availability determines how long a repair actually takes A five-year labour warranty has limited operational value if a failed motherboard, backplane, RAID controller or PSU cannot be obtained promptly. The quote should separate the right to request support from a measurable spare-supply commitment. For critical systems, ask for: A spare-support period measured from final acceptance. A list of customer-replaceable and field-replaceable units. Maximum dispatch and delivery times for critical parts. Regional stocking locations for the deployment sites. Advance replacement terms and failed-part return periods. Post-warranty spare prices or a defined price basis. Rules for supplying an equivalent component after the original part reaches end of sale. Confirmation that substitute hardware and firmware remain compatible with the installed cluster, array or operating environment. Initial spares may be justified for installations where restoration cannot wait for regional movement. Typical candidates include hot-plug PSUs, fans, boot drives and storage media. The correct list depends on system design, installed quantity, workload criticality and whether parts can be shared across identical nodes. The Department of Expenditure’s Manual for Procurement of Goods, 2024 treats acquisition, operation, maintenance and disposal as parts of total cost. For capital equipment, it also supports considering initial spares and the present value of AMC charges. The evaluation formula and measurable inputs should be disclosed in the tender rather than introduced after bid opening. Drive capacity is not a drive specification Two drives with the same capacity and interface can have materially different workload limits, endurance, error characteristics and warranty terms. “2 TB SATA HDD” or “3.84 TB SSD” is therefore insufficient for a five-year comparison. HDD class must match the workload An HDD line item should identify the manufacturer part number or specify an objectively equivalent class. Relevant details include recording technology, rotational speed, interface, sector format, rated annual workload, intended duty, non-recoverable read-error specification and manufacturer warranty. Published product data shows why this matters. A Seagate desktop HDD listing has specified a 55 TB-per-year workload limit, 2,400 power-on hours per year and a two-year warranty. WD Red Plus NAS drives have been specified at 180 TB per year with a three-year limited warranty, while a WD Ultrastar DC HC590 data-centre model has been rated at 550 TB per year with a five-year limited warranty. These are model-specific examples, not specifications for every drive in each brand or family. See the Seagate desktop HDD information , WD Red Plus data sheet and Ultrastar DC HC590 data sheet . For surveillance and VMS servers, drive selection should account for camera count, stream bitrate, retention period, analytics activity and RAID rebuild traffic. Continuous recording and AI-assisted video analysis are not equivalent workloads. A capacity calculation alone does not establish drive suitability. SSD warranty can end on write endurance An enterprise SSD described as having a five-year warranty may also be limited by total bytes written or drive writes per day. The applicable warranty can end when either the calendar period or endurance allowance is reached. The quote should state: Exact SSD model and usable capacity. SATA, SAS or NVMe interface. 2.5-inch, U.2, M.2, E1.S or E3.S form factor. DWPD and the corresponding TBW or PBW at the quoted capacity. Read-intensive, mixed-use or write-intensive classification. Power-loss protection and end-to-end data-protection features. Hot-plug and dual-port capability where required by the architecture. Whether media wear-out is covered by the server warranty. Samsung’s published PM893 specification, for example, applies a five-year or 1 DWPD condition, whichever is reached first. This illustrates why the warranty period cannot be read independently of the expected write volume. See the PM893 product specification . Expected write utilisation should be calculated before bid finalisation. If the application is likely to exceed the quoted SSD endurance, the five-year cost must include higher-endurance media or planned replacement. Two PSUs do not automatically provide power redundancy A line item showing 2 x 1100 W hot-plug PSUs proves only that two modules are supplied; it does not prove that the fully configured server can continue operating after one PSU or one input feed fails. The technical evaluation should confirm that the maximum expected configuration remains within the capacity of one surviving PSU. This calculation must include CPUs, DIMMs, drives, GPUs, accelerators, NICs and other PCIe cards, not only the base chassis. The quote should identify whether the system operates as 1+1 redundant power or as 2+0 combined capacity without redundancy. If upstream supply failure is within scope, each PSU should connect to a separate PDU or UPS path. Two PSUs connected to the same PDU do not protect against failure of that PDU or its upstream source. Also verify: Supported redundancy mode at the quoted peak load. Input voltage and power-cord type. PDU socket and current compatibility. PSU efficiency rating at relevant utilisation levels. Behaviour after loss and restoration of one input feed. Whether power capping or performance reduction is required after PSU failure. Whether the monitoring platform reports PSU and grid-redundancy status. Power supplies with higher wattage may be required after adding GPUs or high-power CPUs. If the initial quote omits that requirement, a later accelerator upgrade can also require replacement PSUs, cables and power-distribution changes. Maintenance, licences and renewals belong in the evaluated price Five-year cost should include all payments necessary to keep the quoted configuration supported and usable. Depending on the deployment, these may include warranty extensions, AMC, operating-system subscriptions, virtualisation licences, backup licences, management software, security updates and per-socket, per-core or per-device renewals. A practical comparison should show the cash outflow for each year and the cumulative five-year amount. For a formal TCO evaluation, future AMC and other measurable charges may be converted to Net Present Value using the rate and method declared in the tender. Do not add an informal downtime estimate after bids have opened. If restoration risk, energy consumption or maintenance cost will affect commercial ranking, the evaluation formula, evidence requirements and assumptions should be stated in advance. A five-year comparison sheet should capture these figures For each compliant quotation, record the following on a common basis: Complete acquisition price for the accepted bill of material. Installation, commissioning and migration charges. Warranty start point, duration and covered components. Location-wise response and restoration commitments. Required initial spares and post-warranty spare terms. AMC or warranty-extension charges for each uncovered year. Software and subscription renewals. Expected storage replacement arising from capacity growth or endurance. Power accessories and electrical changes required for redundancy. Training, preventive maintenance and health-check charges where applicable. End-of-term data sanitisation, media retention and disposal obligations. Any item marked “optional”, “as applicable”, “standard warranty” or “support included” should be clarified before commercial comparison. Such wording transfers an undefined cost or obligation into the operating period. Questions to settle before issuing the purchase order Is the quoted configuration identified down to manufacturer model or measurable equivalent specification? Does the five-year period start from final acceptance? Are parts, labour, travel and onsite attendance included at every deployment location? Is the SLA for acknowledgement, arrival, repair or restoration? Which critical spares are stocked, and where? Are HDD workload class and SSD endurance suitable for the actual application? Can one PSU carry the complete configured load? Are the two PSUs connected to independent upstream power paths? Are all firmware, management and software entitlements included for five years? Have AMC, renewals and predictable replacement costs been included in the evaluated amount? Server configurations and support requirements can be reviewed through /products and /services . For a bill-of-material review or a location-wise support requirement, use /contact .