Prospective users HowTo

Prospective users HowTo

This page describes how new labs can get access to the HPC facilities and how to get funding for computing with SCITAS. It is aimed towards new Professors joining EPFL or, more generally, willing to start computing with us.

Accounts

Free accounts

Everyone can try our service out with a free account. Please note that these accounts are individual and not linked to a lab.

Paying accounts

Once you are comfortable with the provided environment and you wish to run production calculations, you can ask for a paying account (premium) for your lab/group. You will receive a monthly usage report and a quarterly invoice. When your consumption has reached cruise speed, you can commit for the usage of a certain amount of CPU time for four years. Please contact us.

CPU costs eligibility

Our costs are eligible with FNS and ERC. The current costs are:

Node type invoiced (CHF) EPFL subsidy (CHF)
HPC standard – core hour 0.0082 0.0105
HPC large memory – core hour 0.0090 0.0102
HPC accelerated – GPU hour 0.1755 0.0653

The invoiced price is that which you are asked to pay and which will be reimbursed by the FNS/ERC. The subsidy takes into account the ancillary costs of running HPC clusters and together the two give the total cost of ownership.

Costs may vary from year to year depending on market prices fluctuations of nodes costs, number of cores managed by the facility, and staff.

How to indicate costs in grant proposals

The following text may be used as is to describe the SCITAS facilities:

SCITAS

SCITAS is the High Performance Computing facility at EPFL. It counts more than 400 active users in approximately 180 labs and its user base is constantly growing. It provides the research community at EPFL with computing resources and expertise in the fields of 1) General purpose and specialized computing platforms, 2) Training on a wide range of topics – from how to use clusters to advances parallel programming techniques and 3) Specific application support – 3.1) Consulting and code architecture and best practices, 3.2) Profiling and optimization and 3.3) Parallelization.

Furthermore, storage of computational data for the duration of the project can be handled by the SCITAS platform.

The following text can be adapted and used as a resource justification.

CPU time

This project will require on average __N1__ Million core hours per year on a general purpose cluster architecture (standard HPC cores). According to the current fees (see table below), this will amount to __N2__ kCHF per year, thus __N3__ kCHF for the duration of the project. Costs are claimed via internal invoices, which do not include indirect costs elements, profit margin or mark-up.

Node type invoiced (CHF)
HPC standard – core hour  0.0082
HPC large memory – core hour  0.0090
HPC accelerated – GPU hour  0.1755

Storage costs

SCITAS provides medium term high-performance storage with backup. This storage is directly attached to the clusters. Costs of data storage are claimed via internal invoices and a the following costs apply:

Medium term storage 300 CHF / TB for 3 years (it is possible to purchase yearly storage)
Backup 500 CHF / TB for 3 years

Specific application support

Costs for specific application support for scientific software development 1) Consulting on code architecture and best practices, 2) Profiling and optimization and 3) Parallelization amount to 500 CHF / day. Costs for specific application support are claimed via internal invoices.

ERC proposals

As a general rule, costs for SCITAS services should be indicated under direct costs → “Other goods and services” → “Other” in the budget table provided in the template for part B2 of the ERC application.

FNS proposals

FNS Ambizione

You can indicate your collaboration with SCITAS under 10. Collaboration (national and international). Under 13. Requested funding → Research funds, you can add the costs for Computing time and data and application support costs are added under Additional research costs (incl. consumables)/