A new GPU compute-server named “spirit” was funded by MAPEX

The use of graphics processing units (GPUs) in classic simulations greatly accelerates computing times. They also form the basis for applications in the field of machine learning. 

The compute-server includes: 2 AMD EPYC 9374F central processing units (CPUs) each with 32 cores; 8 NVIDIA RTX 6000 Ada GPUs each with 48 GB GDDR6 GPU memory, 18,176 NVIDIA CUDA Cores, 568 NVIDIA Tensor Cores and 142 NVIDIA RT cores; 1,152 GB main memory.

Key benefits: The compute server allows simulations with GPU acceleration and machine learning applications to be performed on modern state-of-the-art hardware, with full access to the hardware and software.

Features: First and foremost, the compute server allows all types of simulations for any application. However, its special structure makes it particularly well suited for the use of GPUs. Accordingly, the focus is on simulations that use GPUs to accelerate calculations. The main use case, however, is machine learning methods. Here, the GPUs drastically reduce the computing times during the training of artificial neural networks and subsequent evaluation. This compute server provides local, low-threshold access to the software, enabling the installation of special software packages or similar.

Applied by: 

Prof. Dr. rer. nat. Andreas Rademacher 

(Center for Industrial Mathematics)