The demand for user-friendly, high-performance computing (HPC) cloud solutions with preinstalled finite element analysis (FEA) and computational fluid dynamics (CFD) software is rapidly growing due to the increasing complexity of engineering challenges.
Since traditional on-premise HPC solutions often present limitations in scalability and cost-efficiency, cloud-based HPC solutions now offer a compelling alternative, empowering engineers to efficiently run resource-intensive simulations without the hassle of manual setup, license management, or hardware configuration.
This article compares leading HPC cloud providers and evaluates their unique features, pricing structures, scalability, and software integrations to help professionals select the optimal solution for their engineering simulation needs.
Rescale
Rescale is recognized as a leader in adopting HPC cloud solutions for engineering simulations. The company was founded in 2011 based on the real-world engineering challenges faced by its founders during their work as engineers at Boeing and secured early investments from prominent business leaders, including Sam Altman, Jeff Bezos, and Richard Branson.
In 2025, Rescale secured $115M funding, bringing its total funding to $260M. Its client portfolio features such well-known companies as Airbus, Nissan, and Samsung.
Rescale is integrated with the most extensive range of commercial and open source FEA software packages from such vendors as Abaqus, Ansys, Hexagon, Siemens, OpenFOAM, Simvia, and many others with detailed documentation and step-by-step onboarding guides for all software supported.
Rescale’s pricing information has not been officially disclosed for the past decade. Historical data from 2013 and 2015 indicates rates of 0.1-0.3 $/core/hour for on-demand instances and 0.05-0.15 $/core/hour for low-priority ones.
However, I also found its recent 2022 presentation with pricing. For example, a system equipped with an AMD EPYC 7763 (Milan) processor at 2.4 GHz, 448GB memory, 1920GB storage, and 96 cores is prised as 0.0778-0.0813 $/core/hour or 7.46-7.80 $/hour, slightly depending on the tariff - On-Demand Economy (ODE) or On-Demand Priority (ODP).
It was very insightful for me to find out that Rescale was built based on challenges encountered by its founders while working on optimizing Boeing 787's wing design. This resonates a lot with my own experience, as I have also worked on Boeing projects related to 787's wing (structural analysis of equipment for 787's wing structural testing).
Qarnot
Founded in 2010 in France, Qarnot has distinguished itself thanks to innovative approach to HPC infrastructure, positioning it as a pioneer in sustainable computing. The company utilizes distributed servers equipped with advanced heat recovery technology, which significantly reduces energy consumption by eliminating the need for traditional, energy-intensive cooling systems.
Qarnot raised €35M funding in 2023 from European investors (adding to funds invested earlier - €2.5M in 2016 and €9.5M in 2020).
In 2025, Qarnot secured additional funding from the European Union to develop an independent EU-based cloud HPC solution. This initiative is designed to provide European businesses with seamless access to large-scale HPC resources while ensuring compliance with the EU’s stringent data protection and sustainability regulations.
Over recent years, Qarnot has actively adapted its platform for engineering simulations. it is integrated with leading proprietary and open source FEA software - Ansys, Abaqus, Star-CCM, Converge CFD, Comsol, OpenFOAM, Code_Aster, Code_Saturne, Elmer, FreeFem, etc.
Qarnot pricing structure stands out for its affordability and transparency. For example, a system featuring an AMD EPYC 9654P, 512 GB memory, and 96 cores is priced at 2.26 or 4.26 €/hour (depending on task priority) - 0.0235 or 0.0443 €/core/houг. Additionally, Qarnot provides a special offer for startups, including €6,000 in free credits and a 50% discount following credit utilization.
I also should note that the idea of writing the article on this topic was suggested me by Qarnot management - to show how their HPC solutions for FEA simulations compare to other providers in this niche.
However, I tried as much as possible to ensure an objective comparison of all leading providers in this area, including their pricing (even when it is not officially published).
Simr (UberCloud)
UberCloud was launched based on 2012's founders research of providing HPC as a Service. It was designed specifically for FEA simulations and was rebranded as Simr after getting $20M funding from BMW Group in 2024 in addition to $1.7M in earlier venture funding invested secured in 2017.
Simr defines its methodology and best practices for automating and managing simulation processes as SimOps (Simulation Operations Automation). The approach is focused on enabling engineers to run complex simulations across any public, private, or hybrid cloud environment eliminating the need for manual cluster management. Simr operates in your own cloud account (AWS, Azure, etc.) and supports a wide range of preinstalled FEA software through integration with leading commercial platforms such as Ansys, Abaqus, and Comsol.
The platform employs BYOC licensing model (Bring Your Own Cloud) - so you use your own AWS, Azure, or other cloud provider, entirely within your organization’s cloud subscription and pay your cloud provider directly for compute and storage services. Simr’s software is licensed based on a fixed, annual per-user license fee.
Simr pricing details are not publicly specified. According to the company blog, Simr annual software license cost is “just a small fraction (like 5–10%) of the annual cost of an engineer". Also, Simr’s another blog post has referenced $15,000/year as an example for license costs.
CloudHPC
CloudHPC was founded in 2018 in Italy as a cloud platform designed specifically for running FEA and CFD simulations with open source software. The platform integrates with all leading open source simulation packages, such as OpenFOAM, Code_Aster, Code_Saturne, Elmer, FreeFem, etc.
CloudHPC provides transparent pricing information for its services. Standard rates are 0.10-0.22 €/core/hour (for multicore instances) and €0.08-0.18 €/vCPU/hour (for multithread per core instances). Certain instance types are available at a 50% discount, and there are no additional charges for storage or data transfer. To facilitate platform evaluation, new users are provided with 300 free vCPU-hours for testing. CloudHPC also offers volume-based discounts, with available rates accessible via the online pricing calculator.
I would also like to highlight the content quality of the CloudHPC blog, really valuable for users seeking to maximize the capabilities of their simulation workflows, such as detailed use cases and step-by-step tutorials for the wide range of open source packages supported by the platform.
Ansys HPC Solutions - Ansys Cloud, Gateway and Access
Ansys offers a suite of high-performance computing (HPC) cloud solutions targeting primarily large enterprises with significant investments in the Ansys software ecosystem, based on Amazon Web Services (AWS) or Microsoft Azure cloud services.
Ansys Cloud was launched in 2019 as a platform enabling on-demand access to Microsoft Azure computing resources for Ansys simulation. Its pricing structure includes both hardware and software licensing costs.
Hardware costs vary based on the selected hardware configuration and geographic region - for example, for a 120-core node the cost is 7.50-9.50 AEC/node/hour depending. One AEC (Ansys Elastic Unit) is approximately equivalent to one USD in most regions.
Software licensing costs range a lot - from 3 AEC/hour for basic configurations to 105 AEC/hour for the most powerful setups.
Ansys Gateway was announced in 2022 and provides users with access to the full Ansys software suite on AWS infrastructure.
Its pricing includes $50,000 one-time fee (after 30-day trial) and $0.01/core/hour after the first year (not including costs of Ansys licensing and AWS instances).
Ansys Access is a similar system announced in 2024, offers similar functionality based on Microsoft Azure with the same pricing as for Ansys Gateway - $50,000 one-time fee (post 30-day trial) and $0.01/core/hour.
It should be noted that Ansys continues to expand its HPC offerings with more specialized solutions. Notably, shortly after the acquisition of Ansys by Synopsys in 2024, a new HPC platform for electronic component simulations based on Ansys Lumerical and Microsoft Azure was announced. This development suggests that further innovation and expansion of Ansys cloud solutions, aligned with the strategic priorities of Synopsys, can be anticipated in the future.
Managing FEA Software Licensing
HPC cloud providers mentioned here typically employ a Bring Your Own License (BYOL) model for proprietary software, requiring users to supply their own software licenses. It is important to note that additional licensing may be necessary to fully utilize multi-core capabilities for commercial packages, For example, standard Ansys licenses are limited to 4 cores - to increase this number, special licenses (Ansys HPC Pack) must be purchased (one HPC Pack for adding 8 cores, two - for 32 cores, three - for 128 cores, etc.)
Choosing the Optimal Mesh Density for FEA Simulations
Computational resources needed to simulate large-size finite element models depend significantly (usually more than linearly) on degrees of freedom (DOFs, proportional to number of nodes and elements in the mesh).
Therefore, it is reasonable to carefully choose the appropriate mesh density before initiating a series of simulations for a large-size model.
An effective approach for this is to perform a mesh convergence study. This involves running a sequence of simulations for the same scenario while incrementally adjusting the element size, typically by 10–20% between simulations - with the goal to identify the mesh density at which further refinement does not result in significant changes to key simulation results. This approach ensures that the simulation is both accurate and computationally efficient, avoiding unnecessary increases in model size and simulation time.
It sounds rather complicated, so to illustrate it more clearly, here is a picture of how I practically implemented such an approach - the same structural analysis was perfomed for the same model (of machine bed) as an Ansys APDL script with parametrized element sizing while changing mesh density between simulations (with the predominant element size changing from 20 mm to 40 mm in 5 mm steps).
Also, more sophisticated strategies can be employed to control the number of degrees of freedom (DOFs) and optimize mesh density.
Submodeling involves using a coarse global model to capture the overall behavior, then applying its results as boundary conditions to a locally refined submodel. It is particularly effective for capturing detailed responses in critical regions without incurring the computational cost of a uniformly fine mesh.
Superelements condense portions of the model into modular components defined by their interface nodes. This reduces the overall number of DOFs while preserving essential behavior. However, implementing superelements requires additional steps, such as defining its boundaries and ensuring proper connectivity with the rest of the model.
It is important to note that these advanced approaches are not universally applicable to all types of FEA simulations and typically require more complex setup, including more complex meshing and additional model partitioning.
In summary, the comparative analysis presented in this article underscores the growing adoption of HPC cloud platforms for FEA and CFD applications across a wide range of industries. The significant investments made by venture capital in these platforms in recent years further highlight the sector’s importance and future potential.
Continued innovation in cloud-based HPC solutions provides engineers with an expanding array of choices and enhanced flexibility. By carefully selecting an HPC cloud provider that aligns with their specific simulation requirements, preferred software tools, compliance standards, and budgetary constraints, organizations can achieve faster turnaround times, improved accuracy, and more sustainable engineering outcomes.
No comments:
Post a Comment