Equipment modelling

The target method reconstructs equipment from components, then allocates impacts to uses. Old Open Data equipment factors are obsolete for this calculation. A representative product LCA remains useful for calibration, validation or as a compatible specific inventory.

Reference models provide context; their reuse depends on boundary compatibility.¹

Component-based principle

Digital equipment combines processors, memory, storage, displays, batteries and structural elements:²³

$$I_{equipment_i}=I_{cpu_i}+I_{ram_i}+I_{storage_i}+I_{display_i}+I_{battery_i}+I_{structure_i}$$

Transparent

Each component can be traced to its source.

Parametric

The impact varies with the equipment’s characteristics.

Modular

A component can be refined independently within the model.

Comparable under conditions

Boundaries, units, characterisation methods and data quality must be harmonised before comparison.

Unknown components must remain explicit gaps or documented residuals.

Choosing the representation

Approach	Application	Description	Traceability
Component model	Known/reconstructable configuration	Detailed, without automatic accuracy guarantee	Source for each component
Hybrid	Partial configuration	Components plus residual	Both parts documented
Parametric archetype	Family known, exact configuration unknown	Representative configuration	Explicit assumptions
Aggregate factor	Compatible equipment-level inventory	Aggregated	Source and covered boundary

Component reference catalogue

The target equipment catalogue groups CPUs, accelerators, memory, storage, displays, batteries and structures. CPUs and accelerators are still published under cloud in the current snapshot; moving them is a future migration, not an already available path.⁴

Characteristics describe hardware. Factors and models describe its impacts with units, phases, sources and versions. Configurations assemble these elements. The cloud catalogue retains provider offerings and mappings to hardware.

An accelerator card separates die, memory and auxiliary elements. A complete-card inventory replaces the elements it covers; it is not added to them. The die proxy used in some illustrations remains a subtotal.

Accelerator card components

Accelerator card

MemoryMemory

Computedie

MemoryMemory

Circuit board · power supply · cooling

A full-card inventory replaces the components it covers; it is not added to them. This schematic represents no specific architecture or impact shares.

Assessing impacts

Component models

Evidence differs between models. Historical CPU, RAM and SSD coefficients are documented by Boavizta. Node, display, chemistry and residual coefficients used in the exploratory simulator include D4B scenarios. A contextual publication does not validate an individual coefficient.⁵⁶⁷⁸

Component	GWP model, kg CO₂ eq.	Inputs	Status / provenance
CPU die	$(die_{mm^2}\times0.0197+9.14)F_{node}F_{pkg}$	Die area, node, packaging	Historical Boavizta base; node/package multipliers are hypotheses
RAM	$((capacity_{GB}/density)\times2.20+5.22)F_{pkg}$ per module	Capacity and documented physical density	Historical Boavizta model; technology compatibility required
SSD	$(capacity_{GB}/density_{NAND})\times2.20+6.34$ per drive	Capacity, NAND density	Historical Boavizta model
HDD	31.11 per drive	Drive count	Historical fixed estimate; suitability depends on the equipment
Display	$area_{cm^2}C_{technology}+base$	Diagonal, technology	D4B scenario, not calibrated here against a product LCA
Battery	$capacity_{kWh}F_{chemistry}$	Capacity and chemistry	D4B scenario; contextual battery studies do not validate the chemistry table
Structural elements	Equipment-type residual	Device type	Supplier reports and explicit estimates where available

A complete graphics card includes more than a CPU-style die estimate: memory, board, cooling, casing and other processes may be material. See the specialised component chapters.

Embodied impact

Annual allocation

$$I_{emb_i}=\frac{\sum_c I_{c,emb_i}+I_{structure_i}}{RL}$$

Component and structural inventories are total embodied impacts in $U_i$; $RL$ is the retained lifetime in years. The result is $U_i/\text{year}$. The component sum excludes the structural residual when that residual is added separately.

Manufacturing geography

Manufacturing depends on processes and their energy supply. Wafers, assembly, memory, boards, displays and batteries may involve different countries. Market shares do not identify a particular product's supply chain.

Geography is specified by stage when known. Otherwise, the assessment retains the selected inventory's geography and identifies the gap. Prefilled exploratory countries are not observed production locations. The electricity factor of the usage country cannot replace total manufacturing impact.

Manufacturing versus use

Usage country determines operational electricity factors. Changing it does not relocate manufacturing. Geographic adaptation of manufacturing requires the underlying process and electricity inventory.

Lifetime correction

The $TL/RL$ correction applies only to a factor already normalised per time unit, not again to total manufacturing. See Time and ageing.

Operational impact

$$\begin{aligned} P_{total}&=P_{cpu}+P_{ram}+P_{storage}+P_{display}+P_{other}\\ E_{elec}&=P_{total}D_{usage} \end{aligned}$$

Power in kW multiplied by duration in hours gives kWh. Supply losses and missing components contribute separately only when absent from the power boundary. A complete wall-power measurement already includes its component consumption.

Convert electricity using the relevant electricity impact factors.

Worked laptop reconstruction

Fictitious scenario: 14-inch laptop, 150 mm² CPU with a 7 nm scenario multiplier, 16 GB RAM, 512 GB SSD, LCD and 56 Wh NMC battery.

Component	Applied scenario	Manufacturing GWP, kg CO₂ eq.	Approximate share
CPU	$(150\times0.0197+9.14)\times1.8$	21.8	14.7%
RAM	$((16/1.0)\times2.20+5.22)\times1.1$	44.5	30.0%
SSD	$(512/48.5)\times2.20+6.34$	29.6	19.9%
LCD	$540.36\times0.07+3.0$	40.8	27.5%
Battery	$0.056\times65$	3.6	2.5%
Structure	Laptop residual assumption	8.0	5.4%
Total	Unrounded sum	148.3	100%
Annual allocation over five years		29.7 kg CO₂ eq./year

Interpretation

The 16:9 screen area is $14^2\times2.54^2\times144/337\approx540.36$ cm². Values are rounded. The CPU multiplier 1.8, RAM density 1 GB/cm² and DDR5 packaging 1.1, display/battery factors and structural residual are scenario assumptions. Their contribution shares do not establish universal priorities or the footprint of a commercial laptop.

Cross-validation against equipment factors

A valid comparison needs the same configuration, lifetime, lifecycle stages, LCIA method and unit. A difference from a family average neither validates the component model nor proves that hardware generation explains the gap.

The comparison identifies the exact product LCA, version, configuration, boundaries and unresolved differences. Earlier non-reproducible numerical comparisons are retained in the refactoring archive rather than used as validation evidence. An independent calibration campaign remains necessary.

Hybrid model

The residual covers unmodelled board, casing, power supply, cables and assembly. An explicit description of its contents prevents overlap; see Structural elements.

When a component model is unsuitable

Missing data

Absent information remains identified as missing. A comparable reference can supply an estimate, with its source and justification. This substitution is a modelling choice, not automatic hardware detection. Known data take precedence.

A documented representative configuration, called an archetype, can fill missing characteristics or represent the whole device when detail is insufficient.

Situation	Response
Component model unknown, type known	Documented same-type archetype
Component and type unknown	Explicit residual where justified; otherwise gap
Power unknown	Representative family estimate, with operating assumptions
Lifetime unknown	Documented nominal or expected lifetime
No detailed equipment data	Compatible aggregate equipment factor

Assumptions and limitations

• Component decomposition is useful only when its inputs are supportable.
• Historical coefficients require technology and unit checks; CPU die area is in mm², whereas some memory models use cm².
• No statistical validation of D4B scenarios is claimed. Missing non-climate inventories are not zero impacts.
• Structural residuals remain an area for improvement.

Footnotes

1. UBA, Green Cloud Computing, 2021; verify component-specific units. See the documentary corpus. ↩

2. Kelly et al., 2021, lithium carbonate/hydroxide LCA: battery context, not validation of simulator chemistry factors. See the documentary corpus. ↩

3. Ellingsen et al., 2014, LCA of a Li-Ion Battery Vehicle Pack. ↩

4. Digital4Better, data catalogue, snapshot 77cae4a, current paths data/cloud/cpus.* and data/cloud/accelerators.*. The target organisation is a D4B convention. ↩

5. Boavizta documentation. Online documentation, unpinned version. ↩

6. Tannu & Nair, 2023. The Dirty Secret of SSDs: Embodied Carbon ↩

7. Historical component models, documented hardware configurations, versioned digital4better/data cloud CPU/accelerator records and product LCA reports where a compatible comparison is possible. Digital4Better data catalogue, snapshot 77cae4a. ↩

8. Digital4Better, ADEME Base Empreinte and Digital4Better data.Evolving catalogue; pin a commit for reproducible factors. ↩