LUMI GPU early access platform available - LUMI GPU early access platform available
LUMI GPU early access platform available
As of today, we are making the LUMI GPU early access platform (EAP) available to LUMI users. The goal of this platform is to give LUMI’s users access to AMD GPUs in order to prepare their codes and work for the upcoming LUMI-G partition.
This early access platform is a set of nodes that are available as a partition in the LUMI-C system. The EAP nodes consist of a single AMD EPYC 7662 64 core CPU, 512 GiB of memory, a single 100gbit network adapter, and 4x AMD MI100 GPUs. The main part of interest here is the MI100 GPUs as they are the predecessor to the MI250X GPUs that will be used in LUMI-G. The main point of these nodes is to give access to the AMD software environment and the option to try out codes on real AMD GPUs, and not as a place where to run full-scale simulations.
On these nodes, we are currently offering the AMD ROCm stack which includes the HIP libraries and tools for converting CUDA code to run on AMD GPUs. The ROCm stack also includes a version of the clang compiler that can offload OpenMP code to AMD GPUs. In addition to this, there is the Cray software stack that has the capability to offload OpenMP C/C++ code and OpenACC Fortran code to the AMD GPUs. The software and features available on these nodes are still evolving and we will be improving them over time.
For more in-depth documentation of how to use the EAP nodes please check LUMI Documentation for LUMI-G Early Access Platform.
Access to these nodes is available to all users that have a project on LUMI, but we will be giving priority to the pilot users of the second phase if needed. If you do not have an active project on LUMI but would like access to the EAP nodes, please contact your local resource allocator about getting a development project on LUMI. For more information check the Get Started guide. For users that are not from a LUMI consortium country, development access may be obtained from the EuroHPC JU.