HIP101: porting CUDA codes to HIP

With the upcoming LUMI supercomputer arriving this year, we start a series of training for the users. The main partition of LUMI is based on AMD GPUs, thus the users should be ready to port their applications to this new architecture. The first training will be about porting CUDA applications to HIP. It will provide an extensive dive into the Hipify tools and showcase many examples.

Moreover, we go one step further with a hands-on session. You can participate in the hands-on sessions with the applications that we will provide or you can send us your CUDA application in advance. CSC’s Puhti system with Nvidia GPUs will be used for the hands-on part. We will work on a number of selected applications in preparing for the hands-on session. The number of the evaluated applications depends on how many applications we will receive. Of course, you can work on your application without sending it to us but there can be some challenges as we will not have verified that it runs on CSC’s Puhti system.

In order to send your code, please prepare a txt file including the following:
1) How to retrieve the code and the possible input data, be sure that there is access.
2) The name of the code, the person of contact for any questions, and a list with the required dependencies.
3) How to compile and execute the code.

This training is open to the LUMI consortium. We believe that many future users of LUMI will be interested in this training. Training accounts will be provided for accessing Puhti supercomputer

The deadline for the registration is February 15th.

You can submit your code in advance, for testing on Puhti Supercomputer. In case of any problem, we will communicate this with you. Please send the txt file with all the information to georgios.markomanolis(at)

Learning Outcome

Aftet this course the participants will know how to use the AMD Hipify tools to convert CUDA codes to HIP. They will be familiar with the AMD architecture, issues that could arise from the code conversion and the differences between CUDA and HIP.


In order to take the course, the participant should be familiar with GPU concepts and CUDA. Any experience with GPU is welcome. The examples will be delivered in various programming languages among Fortran and C.

Agenda (Times are in CET)

09:00 – 10:00 Introduction to AMD architecture and HIP
10:00 – 10:15 Break
10:15 – 10:45 Deep dive to Hipify tools and examples
10:45 – 11:30 Lunch
11:30 – 16:00 Hands-on sessions