Conditions for Access

1. Guiding Principles

The Aphroditi access policy operates under a unified access model designed to serve the national research and innovation community. Its core guiding principles are fairness, transparency, efficiency, and compliance.

• Primary Focus on R&D&I: The facility strategically emphasizes Research, Development, and Innovation (R&D&I) to advance scientific knowledge and foster innovative entrepreneurship.

• Complementary to Larger Systems: The facility serves as a primary resource for moderate-scale projects, directing users with larger computational demands to international facilities like those provided by EuroHPC.

2. Access Modes, Policy, and Conditions

Access to the Aphroditi HPC-AI facility is provided through two distinct modes, as defined in the CyI General Framework. The mode of access is determined by the nature of the proposed work.

3. Application and Evaluation Process

In accordance with the CyI General Framework, which states that a research infrastructure may announce “periodic or open calls for applications depending on the availability of the Research Infrastructure(s)", access to Aphroditi operates under a continuous open call, with submissions accepted and evaluated on a rolling basis rather than through fixed periodic windows. This ensures rapid resource access and maximum utilisation of the infrastructure.

New users begin with a one-time account registration request through the designated HPC ticketing system, found at the following address: Submit a Request via the HPC Ticketing System

Once the account is provisioned, all project activities, such as creating a project, submitting resource allocation requests, managing team members, monitoring usage in real time and requesting renewals, are carried out by the Principal Investigator through the ColdFront self-service portal.

3.1 Application Content

To be considered for evaluation, applicants must submit a comprehensive proposal, typically 3-4 pages, providing a convincing case for the project's merit and resource needs. The proposal must include:

• Project Description: A clear description of the project's goals, its scientific excellence or potential for innovation, and its novelty and timeliness.

• Methodology: A detailed explanation of the methodology that will be used to achieve the project's objectives.

• Project Team: A brief presentation of the Principal Investigator and collaborating team members, outlining their expertise and the management structure of the project.

• Software and Technical Readiness: A statement on the availability of the necessary software. In the case of community or user-developed code, evidence must be provided that it has been tested for efficiency, scalability, and suitability for the Aphroditi architecture. For licensed software, applicants must be able to provide a valid license.

• Resource Justification: A detailed work plan that clearly justifies the number of CPU/GPU hours requested. This must include calculations based on the proposed job size and the number of job executions planned.

• Confidentiality: Applicants must identify any confidentiality issues and provide a justification for why they are necessary.

3.2 Evaluation Process and Timeline

The evaluation process operates on a continuous rolling basis and is divided into three distinct paths depending on the scale of the resources requested:

• Small-Scale Projects (Fast-Track), up to 3,000 B200 or 11,000 RTX GPU-hours over a 3-month period:

  • Process: A quick eligibility check focusing on the types of entity, activity, and completion of forms (including an AI ethics self-assessment).

  • Evaluator: The review is conducted by the Internal HPC Team.

  • Timeline: A decision is provided within 2-3 working days.

• Medium-Scale Projects (PRC Review), 3,000 to 10,000 B200 GPU-hours or 11,000 to 25,000 RTX GPU-hours over a 3-month period:

  • Process: Evaluates eligibility and technical merit, including methodology, use of resources, code details, and compliance measures.

  • Evaluator: The review is conducted by the Project Review Committee (PRC), which is composed of HPCF engineers and two PharosCY engineers.

  • Timeline: The committee evaluates proposals within 1 week.

• Large-Scale Projects, more than 10,000 B200 GPU-hours or 25,000 RTX GPU-hours over a 3-month period:

  • Process: Because requests of this scale consume a significant share of the facility's total capacity, they are evaluated on a case-by-case basis.

  • Evaluator: The review is conducted by the Project Advisory Committee (PAC), as described below.

  • Timeline: Communicated to the applicant on a case-by-case basis.

3.3 Waitlist and Scheduling Model

To maintain the continuous utilisation of the infrastructure, technically approved projects are never rejected or stalled when allocation capacity reaches 100%. Instead, they are placed on an active waitlist and granted immediate low-priority access, allowing them to utilise idle resources. They can run background jobs, such as code optimisation, compilation, and scaling tests, on nodes that are temporarily inactive between scheduled production runs. As existing production projects complete and dedicated capacity frees up, waitlisted projects are automatically upgraded to standard priority in queue order, at which point their official 3-month production run begins.

3.4 The Project Advisory Committee (PAC)

The Project Advisory Committee (PAC) is the standing advisory body for the Aphroditi national HPC-AI facility. Its purpose is to oversee the fairness, transparency, and scientific quality of the facility's access policy, to advise on its operation and evolution, and to review large-scale resource requests on a case-by-case basis. By involving the PAC in the largest allocations, the facility ensures that a significant portion of national computational capacity is matched to projects of high scientific excellence and strategic national importance.

In case of rejection, applicants will receive detailed feedback in order to improve and resubmit their proposals.

4. Resource Management and Renewals

Projects operate on standard 3-month access cycles and can be renewed in 3-month increments to support ongoing research; they are not granted indefinite allocations. All extension requests must be submitted by the PI directly through the ColdFront portal, accompanied by clear scientific reasoning, a progress summary, and justification for the additional CPU/GPU hours requested. Renewal approval is strictly tied to historical usage performance: to prevent "resource hoarding", automated penalties, such as reduced allocation caps or lower queue priority, may be applied if a team has failed to meaningfully utilise its previously granted allocation. Depending on the resources requested (small/medium/large scale), the renewal request is reviewed by the corresponding body: the Internal HPC Team or the Project Review Committee. To ensure fair access, multiple recurring requests from the same applicants may also be given lower priority than those from new users.

5. Data Management and Retention Policy

• No Backup: There is NO BACKUP performed for any user directory on the system. Users are solely responsible for securing their own data.

• Data Removal: All project data must be removed from the system by the user within one month after the end of the project. One month after the agreement terminates, CyI-HPCF will remove all user accounts and any remaining files.

• Scratch Directory: Files in the scratch directory are automatically deleted 30 days after their last access time; empty directories are deleted after 30 days.

6. User and Institutional Obligations

Access to the facility is governed by the principles of two agreements: the Research Infrastructure Access Agreement (at the institutional level) and the User Access Agreement (at the individual level). The implementation of these agreements depends on the access mode.

• User Access Agreement (Individual): This agreement applies to all users (both free and paid access). It is implemented as the Acceptable Usage Policy (AUP). To be granted access, all users...

• Research Infrastructure Access Agreement (Institutional):

  • For Relevance-driven (Free) Access: This agreement is established through a process of formal communication. It is initiated when a Principal Investigator (PI) submits a proposal on behalf of their institution and is considered finalized upon the official notification of the resource award from CyI.

  • For Market-driven (Paid) Access: This agreement is formalized as a contract between CyI and the User Institution detailing the scope of work, fees, payment terms, and other legal provisions.

Key obligations covered within these agreements include:

• Final Report: The project's PI must submit a Final Report summarizing the results within two months of the allocation's completion.

• Acknowledgement: All publications resulting...

  "This work was supported by computing time awarded on the Aphroditi infrastructure of the High Performance Computing and AI Facility of The Cyprus Institute under project ID PROJECTID."

  Where technical support from staff has been received, the following additional text is recommended:

  "The assistance of [name of person/people] from the CyI HPC-AI Facility in achieving the technical requirements is gratefully acknowledged."

• Acceptable Usage: The AUP, which all individual users must accept, includes rules such as the non-sharing of personal usernames and using the system only for the approved project.

7. Intellectual Property Rights (IPR)

The allocation of Intellectual Property Rights (IPR) is governed by the principles laid out for Relevance-driven access in the CyI General Framework. As such, Results generated during the project are owned by the Party that generates them. Where Results are generated jointly and inseparably, they shall be jointly owned by the User's party (Entitled Party) and CyI.

8. Data Protection

The handling of all personal data related to the application and access process adheres strictly to the EU General Data Protection Regulation (GDPR), as detailed in the CyI General Framework. Personal information will be used only for the purposes of managing the access process and will not be disclosed without the explicit permission of those involved.