Evaluating Microsoft Azure as an Alternative to On-Premises HPC

To enhance efficiency, satisfy customer expectations, and ensure their high-performance computing (HPC) systems are future-ready, manufacturers are actively seeking innovative solutions to optimise their computing environments. But a key challenge for manufacturers has been the lack of an independent, self-led comparison between high-performance computing (HPC) capabilities and traditional on-premise solutions. While HPC offers clear engineering benefits, the cost analysis is complex. This is where MTC were able to help.  

Providing an impartial analysis, MTC explored real-world manufacturing simulations to benchmark key performance indicators (KPIs) and develop a model that helps manufacturers make informed choices about their HPC infrastructure. Highlighting the potential for manufacturers to streamline operations, accelerate product market entry, and unlock revenue faster. 

Use Cases  

To ensure a relevant and practical comparison, MTC selected two distinct simulation scenarios:  

Fluid-to-fluid heat exchanger with intricate internal and external shapes, employed for Use Case 1 

Analysis  

MTC developed a unified cost model to assess the viability of Microsoft Azure HPC versus on-premise computing, accounting for factors like scalability, computational speed, and utilisation. The model serves as a roadmap for manufacturers considering cloud-based simulations, offering insights from standalone modelling to fully integrated digital twins.  

• Cost – A detailed cost model was developed, factoring in annual run costs, data transfer expenses, and utilisation rates.  
• Usability – The ease of integration and accessibility of each solution for manufacturers.  
• Elasticity – The ability to scale computing resources on demand.  
• Speed – Processing efficiency and turnaround time. 

Analysis  

MTC developed a unified cost model to assess the viability of Microsoft Azure HPC versus on-premise computing, accounting for factors like scalability, computational speed, and utilisation. The model serves as a roadmap for manufacturers considering cloud-based simulations, offering insights from standalone modelling to fully integrated digital twins.  

• Cost – A detailed cost model was developed, factoring in annual run costs, data transfer expenses, and utilisation rates.  
• Usability – The ease of integration and accessibility of each solution for manufacturers.  
• Elasticity – The ability to scale computing resources on demand.  
• Speed – Processing efficiency and turnaround time. 

Our findings revealed several key insights

Optimising Cost & Utilisation 

While an efficiently managed on-premise HPC setup can achieve 75–80% utilisation, Azure's reserved offering proves more cost-effective for those with lower on-premise utilisation rates. To handle peak demands flexibly without additional machine reservations, combining Azure's Reserved and OnDemand services can further enhance cost-effectiveness.
 

Cloud Migration 

For customers utilising a hybrid on-premise and cloud approach, frequent large data transfers back to on-premise storage can significantly reduce cost savings, especially for high-volume, fast simulations. However, as many customers transition to cloud-based solutions, this issue may be temporary.

To mitigate the high cost of data transfer, simulation files could be post-processed and stored in the cloud, provided the processes permit. In the use cases examined in this study, transferring the final simulation files to on-premise storage incurred an additional cost of £35,000 per year.
 

Balancing Speed & Cost 
 

More computing power doesn’t always mean a linear cost increase. Running simulations with too few cores can be both slow and expensive, while increasing core count strategically—e.g., from 32 to 96 vCPUs—can dramatically reduce run times with only a modest cost increase.

 

Conclusion 

In MTC's use cases, on-premise computing proves more economical only when utilisation exceeds 87% for fluid simulations and 72% for mechanical simulations—levels that many manufacturers find challenging to sustain consistently. Nonetheless, the cost analysis indicates that the total cost of ownership for both on-premise and Azure HPC solutions is comparable. 

When assessing their HPC capabilities, customers should consider more than just cost and thoroughly evaluate their specific requirements. A structured approach ensures the selection of technology that best fits their needs. 
This study provides a systematic method for analysing digital infrastructure, empowering manufacturers to make informed decisions that align with their operational and cost considerations. By adopting this approach, businesses can optimise their computing environments to enhance efficiency, meet customer demands, and future-proof their HPC capabilities.