Computle study questions cloud workstation pricing

Computle has published a study of cloud workstation usage in the UK architecture, engineering and construction sector, based on 51 million telemetry samples collected over 82 days.

The data suggests that professional CAD, BIM, visualisation and rendering workloads use only a small share of available GPU power during routine work, raising questions about cloud pricing models that charge as if machines run at peak output all the time.

The study covered 150 UK machines used by five customers across four Nvidia GPU classes: RTX 4000 SFF, RTX 5070, RTX 5080 and RTX 5090. It found that an RTX 5070 workstation averaged 7.9W of GPU power draw against a rated 250W thermal design power, or about 3.2% of its specification.

Across the monitored fleet, average CPU utilisation was 5%, while 85.6% of machines recorded zero utilisation at any given moment. According to the report, nearly 60% of all telemetry readings showed both CPU and GPU workloads effectively idle.

Those figures point to what Computle described as a bursty pattern, in which short periods of heavy activity are followed by long periods of low usage. In practice, many architecture and design teams actively use workstations for about 40 hours a week, even though conventional cloud contracts often charge for all 168 hours.

The findings come as architecture, engineering and construction firms continue to review hybrid working arrangements, cloud rendering workflows and energy costs after prolonged volatility in European electricity markets.

Pricing model

Using the telemetry data, Computle has introduced a pricing structure called Flex, which separates fixed workstation infrastructure charges from metered electricity use. For a 50-seat architectural studio, the company said the model could cut annual cloud workstation costs by around GBP £21,500 compared with static contracts.

The report argues more broadly that older pricing approaches reflect assumptions formed when electricity costs were lower and less volatile. Cloud workstation contracts have often bundled hardware access and energy consumption into a single fixed fee, regardless of whether machines sit idle for long periods.

"Cloud workstations were priced for a world of cheap electricity," said Jake Elsley, Chief Executive Officer of Computle. "That world ended in 2022. The gap between what silicon actually draws and what static contracts bill for stopped being a curiosity and started being a material cost problem for studios and practices."

Energy use

The study also examined emissions linked to idle infrastructure. Computle estimated that an RTX 5070 workstation running continuously on the UK grid would generate about 129 kg of CO2 a year.

It said the same workload, supplied through its renewable-backed UK infrastructure, would reduce operational scope-2 emissions to zero. While that estimate reflects Computle's own infrastructure model, the comparison underlines growing attention on the energy profile of digital tools used by design and engineering teams.

Cloud workstations have become more common as firms spread teams across multiple locations and rely on remote access to specialist software. For practices handling CAD models, BIM files and visualisation work, remote systems provide access to expensive hardware without requiring a physical workstation for every employee.

Yet the economics of that shift have become harder to ignore as electricity prices have risen and clients have demanded tighter control over operating costs. If day-to-day workloads rarely approach the upper limits of a machine's GPU and CPU, charging as though each system runs flat out around the clock may leave customers paying for power they do not use.

Computle's analysis was limited to a UK sample and a relatively small group of customers, so it does not amount to a sector-wide census. Even so, the data offers a detailed snapshot of how cloud workstations are used in one specialist market and adds to the wider debate over whether infrastructure billing should move closer to real consumption rather than nominal hardware ratings.

For architecture and design firms, that debate goes beyond hardware efficiency. It directly affects how remote working systems are budgeted, how project teams scale rendering and modelling resources, and how technology suppliers justify fixed monthly fees when much of a workstation's installed capacity may sit unused for most of the week.

At the centre of Computle's argument is the claim that the mismatch between rated power and real usage has become a commercial issue rather than a technical footnote.