HAICORE: computing resources for the Helmholtz AI community
Access to computing resources is key for the Helmholtz AI community to accelerate innovative AI applications.
The “Helmholtz AI computing resources” (HAICORE) provide easy and low-barrier GPU access to the entire AI community within the Helmholtz Association.
HAICORE provides two usage modalities:
1. Ad-hoc usage for an initial playing with the data and/or AI methods
2. Lightweight projects of up to 5,000 GPU hours per year
For ease of access, computing services are delivered through web portals and traditional ssh login.
How can you get access to the HAICORE resources?
For ad-hoc usage with up to 10 GPU/h per day - GPUs implemented at KIT
1. Fill in the Access Form for HAICORE System (German - English) from KIT including the following information:
- Project acronym: project or organisation
- BwIDM Account: your date of birth
2. Sign the form and send it to haicore-accountingnoSp@firstname.lastname@example.org.
After a positive formality check, an account will be generated and you will be informed that you now have access to the system. Please note that due to an overflow of work, only 5 people per week will be granted access.
In order to use the system, make sure that you have the KIT VPN turned on. You may access the system either via regular ssh connection or Jupyter Notebooks in the browser. Here are the links to the respective documentations: VPN / ForHLR II / Jupyter
Need help? Contact the HAICORE service hotline for ad-hoc usage: haicore-hotlinenoSp@email@example.com
Any work performed on the HAICORE@KIT partition should be appropriately cited:
"This work was supported by the Helmholtz Association's Initiative and Networking Fund on the HAICORE@KIT partition."
For lightweight project usage with up to 5000 GPU/h per year - GPUs implemented at FZJ
Until the HAICORE resource allocation is set up through the voucher system, compute projects must be created manually. These are the required steps:
- Create an Account at JuDoor
- Send an e-mail to s.kesselheimnoSp@firstname.lastname@example.org with the following information:
- Personal details of the Principle Investigator (PI): mail address, name and surname, title, address
- The PI’s user name in JuDoor
- Project details (project name and platform, e.g. "Helmholtz AI voucher ‘Vastly Parallel Cat and Dog Discriminator’")
- A project acronym (only [a-z] and “_”)
- A human-readable project title
- A short description of the project (<800 characters)
- Names of collaborators
- The number of GPU hours (remember, 5000 is the max!). They will be converted to MCoreHours by NMCoreH=10−6⋅(48/4)NGpuH. The limit of 5000 GpuH corresponds to a 0.06 MCoreH.
Make sure that your code makes good use of the GPUs: On Juwels Booster, one of the fastest supercomputer in Europe, your job will be assigned at least one full compute node with four GPUs. If you use only one, the usage of four will be deducted from your account.
Any work performed on the HAICORE@FZJ partition should be appropriately cited:
"This work was supported by the Helmholtz Association's Initiative and Networking Fund on the HAICORE@FZJ partition."
If a proposal exceeds 5000 GPU hours or is over one year, it will be redirected to the already well established tier 1/2 project allocation process. The Helmholtz AI consultant teams can provide support for the application process via the voucher system.
The HAICORE ramp-up policy contains all relevant information for accessing the resources during HAICORE’s ramp-up phase. If you have any further questions, we are happy to help! Send us an e-mail to contactnoSp@email@example.com.